Capstone_Project_json.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.7.1"
    },
    "colab": {
      "name": "Capstone Project (1).json",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/hpadhuka/customizing_models_tf2/blob/master/Capstone_Project_json.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "r4d-6ciyQV8u"
      },
      "source": [
        "# Capstone Project\n",
        "## Neural translation model\n",
        "### Instructions\n",
        "\n",
        "In this notebook, you will create a neural network that translates from English to German. You will use concepts from throughout this course, including building more flexible model architectures, freezing layers, data processing pipeline and sequence modelling.\n",
        "\n",
        "This project is peer-assessed. Within this notebook you will find instructions in each section for how to complete the project. Pay close attention to the instructions as the peer review will be carried out according to a grading rubric that checks key parts of the project instructions. Feel free to add extra cells into the notebook as required.\n",
        "\n",
        "### How to submit\n",
        "\n",
        "When you have completed the Capstone project notebook, you will submit a pdf of the notebook for peer review. First ensure that the notebook has been fully executed from beginning to end, and all of the cell outputs are visible. This is important, as the grading rubric depends on the reviewer being able to view the outputs of your notebook. Save the notebook as a pdf (File -> Download as -> PDF via LaTeX). You should then submit this pdf for review.\n",
        "\n",
        "### Let's get started!\n",
        "\n",
        "We'll start by running some imports, and loading the dataset. For this project you are free to make further imports throughout the notebook as you wish. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V_epOCg-QV82"
      },
      "source": [
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "import tensorflow_hub as hub\n",
        "import unicodedata\n",
        "import re\n",
        "from sklearn.model_selection import train_test_split\n",
        "from tensorflow.keras.layers import Layer\n",
        "from tensorflow.keras.layers import Input, Dense, LSTM, Masking\n",
        "from tensorflow.keras.models import Model\n",
        "import random\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": 65,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kS5Qtu6mQV83"
      },
      "source": [
        "![Flags overview image](data/germany_uk_flags.png)\n",
        "\n",
        "For the capstone project, you will use a language dataset from http://www.manythings.org/anki/ to build a neural translation model. This dataset consists of over 200,000 pairs of sentences in English and German. In order to make the training quicker, we will restrict to our dataset to 20,000 pairs. Feel free to change this if you wish - the size of the dataset used is not part of the grading rubric.\n",
        "\n",
        "Your goal is to develop a neural translation model from English to German, making use of a pre-trained English word embedding module."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qBIoI2COQV84",
        "outputId": "5c7c6371-0d98-43a2-80f0-faa94e504c35"
      },
      "source": [
        "# Run this cell to load the dataset\n",
        "\n",
        "NUM_EXAMPLES = 20000\n",
        "from google.colab import drive\n",
        "drive.mount('/content/gdrive')\n",
        "\n",
        "data_examples = []\n",
        "with open('gdrive/My Drive/Colab Notebooks/deu.txt', 'r', encoding='utf8') as f:\n",
        "    for line in f.readlines():\n",
        "        if len(data_examples) < NUM_EXAMPLES:\n",
        "            data_examples.append(line)\n",
        "        else:\n",
        "            print(line)\n",
        "            print(line.split('\\t'))\n",
        "            break"
      ],
      "execution_count": 66,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/gdrive; to attempt to forcibly remount, call drive.mount(\"/content/gdrive\", force_remount=True).\n",
            "I've got a secret.\tIch habe ein Geheimnis.\tCC-BY 2.0 (France) Attribution: tatoeba.org #2359617 (CK) & #2824661 (raggione)\n",
            "\n",
            "[\"I've got a secret.\", 'Ich habe ein Geheimnis.', 'CC-BY 2.0 (France) Attribution: tatoeba.org #2359617 (CK) & #2824661 (raggione)\\n']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jp8gUO48QV85"
      },
      "source": [
        "# These functions preprocess English and German sentences\n",
        "\n",
        "def unicode_to_ascii(s):\n",
        "    return ''.join(c for c in unicodedata.normalize('NFD', s) if unicodedata.category(c) != 'Mn')\n",
        "\n",
        "def preprocess_sentence(sentence):\n",
        "    sentence = sentence.lower().strip()\n",
        "    sentence = re.sub(r\"ü\", 'ue', sentence)\n",
        "    sentence = re.sub(r\"ä\", 'ae', sentence)\n",
        "    sentence = re.sub(r\"ö\", 'oe', sentence)\n",
        "    sentence = re.sub(r'ß', 'ss', sentence)\n",
        "    \n",
        "    sentence = unicode_to_ascii(sentence)\n",
        "    sentence = re.sub(r\"([?.!,])\", r\" \\1 \", sentence)\n",
        "    sentence = re.sub(r\"[^a-z?.!,']+\", \" \", sentence)\n",
        "    sentence = re.sub(r'[\" \"]+', \" \", sentence)\n",
        "    \n",
        "    return sentence.strip()"
      ],
      "execution_count": 67,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Wt8hXWelQV85"
      },
      "source": [
        "#### The custom translation model\n",
        "The following is a schematic of the custom translation model architecture you will develop in this project.\n",
        "\n",
        "![Model Schematic](data/neural_translation_model.png)\n",
        "\n",
        "Key:\n",
        "![Model key](data/neural_translation_model_key.png)\n",
        "\n",
        "The custom model consists of an encoder RNN and a decoder RNN. The encoder takes words of an English sentence as input, and uses a pre-trained word embedding to embed the words into a 128-dimensional space. To indicate the end of the input sentence, a special end token (in the same 128-dimensional space) is passed in as an input. This token is a TensorFlow Variable that is learned in the training phase (unlike the pre-trained word embedding, which is frozen).\n",
        "\n",
        "The decoder RNN takes the internal state of the encoder network as its initial state. A start token is passed in as the first input, which is embedded using a learned German word embedding. The decoder RNN then makes a prediction for the next German word, which during inference is then passed in as the following input, and this process is repeated until the special `<end>` token is emitted from the decoder."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lZBs0ufOQV85"
      },
      "source": [
        "## 1. Text preprocessing\n",
        "* Create separate lists of English and German sentences, and preprocess them using the `preprocess_sentence` function provided for you above.\n",
        "* Add a special `\"<start>\"` and `\"<end>\"` token to the beginning and end of every German sentence.\n",
        "* Use the Tokenizer class from the `tf.keras.preprocessing.text` module to tokenize the German sentences, ensuring that no character filters are applied. _Hint: use the Tokenizer's \"filter\" keyword argument._\n",
        "* Print out at least 5 randomly chosen examples of (preprocessed) English and German sentence pairs. For the German sentence, print out the text (with start and end tokens) as well as the tokenized sequence.\n",
        "* Pad the end of the tokenized German sequences with zeros, and batch the complete set of sequences into one numpy array."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "kGnlgy1uQV86"
      },
      "source": [
        "english_sentences =[]\n",
        "german_sentences = []\n",
        "for sentence in data_examples:\n",
        "    eng, deu, rest = sentence.split('\\t')\n",
        "    eng = preprocess_sentence(eng)\n",
        "    deu = preprocess_sentence(deu)\n",
        "    english_sentences.append(eng)\n",
        "    german_sentences.append(\"<start> {0} <end>\".format(deu))"
      ],
      "execution_count": 68,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "dAKLHwWVQV87",
        "outputId": "49663e0e-e7b9-45ab-d15a-8ac9e2d441dd"
      },
      "source": [
        "def create_tokenizer(list_of_strings):\n",
        "    \"\"\"\n",
        "    Word tokenizer \n",
        "    \"\"\"\n",
        "    \n",
        "    tokenizer = tf.keras.preprocessing.text.Tokenizer(filters='', char_level=False, lower=False)\n",
        "    tokenizer.fit_on_texts(list_of_strings)\n",
        "    return tokenizer\n",
        "\n",
        "\n",
        "\n",
        "def strings_to_sequences(tokenizer, list_of_strings):\n",
        "    \"\"\"\n",
        "    This function takes a tokenizer object and a list of strings as its arguments.\n",
        "    It should use the tokenizer to map the text chunks to sequences of tokens and\n",
        "    then return this list of encoded sequences.\n",
        "    \"\"\"\n",
        "    \n",
        "    return tokenizer.texts_to_sequences(list_of_strings)\n",
        "\n",
        "tokenizer = create_tokenizer(german_sentences)\n",
        "deu_input_dim = len(tokenizer.word_index) + 1\n",
        "print(deu_input_dim)\n",
        "german_seq_chunks = strings_to_sequences(tokenizer, german_sentences)"
      ],
      "execution_count": 69,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "5744\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wLD4EulkQV8-",
        "outputId": "bb539676-9a38-4918-8558-e7f1a5b96a69"
      },
      "source": [
        "num_samples = 5\n",
        "inx = np.random.choice(len(english_sentences), num_samples, replace=False)\n",
        "for idx in inx:\n",
        "    print(english_sentences[idx], '\\t', german_sentences[idx], '\\t', german_seq_chunks[idx])"
      ],
      "execution_count": 70,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "i bike to work . \t <start> ich fahre mit dem rad zur arbeit . <end> \t [1, 4, 497, 55, 118, 1651, 313, 163, 3, 2]\n",
            "help me . \t <start> hilf mir . <end> \t [1, 543, 21, 3, 2]\n",
            "i've lost hope . \t <start> ich habe die hoffnung verloren . <end> \t [1, 4, 18, 26, 567, 142, 3, 2]\n",
            "step on it ! \t <start> beeil dich ! <end> \t [1, 882, 28, 9, 2]\n",
            "we're not lost . \t <start> wir sind nicht verloren . <end> \t [1, 17, 23, 12, 142, 3, 2]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ooaFRPh0QV8-"
      },
      "source": [
        "def make_padded_dataset(sequence_chunks):\n",
        "    \"\"\"\n",
        "    This function takes a list of lists of tokenized sequences, and transforms\n",
        "    them into a 2D numpy array, padding the sequences as necessary according to\n",
        "    the above specification. The function should then return the numpy array.\n",
        "    \"\"\"\n",
        "    \n",
        "    return tf.keras.preprocessing.sequence.pad_sequences(sequence_chunks, maxlen=20, padding='post', value=0.0)"
      ],
      "execution_count": 71,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zjHLBJneQV8-",
        "outputId": "36ebbaea-212e-4c33-a843-b958bbc7dbef"
      },
      "source": [
        "# Pad the token sequence chunks and get the numpy array\n",
        "\n",
        "german_padded_sequences = make_padded_dataset(german_seq_chunks)\n",
        "print(german_padded_sequences.shape)"
      ],
      "execution_count": 72,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(20000, 20)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lpO3U_m7QV8_"
      },
      "source": [
        ""
      ],
      "execution_count": 72,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8L9MH6oNQV8_"
      },
      "source": [
        "## 2. Prepare the data with tf.data.Dataset objects"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QyNY94zoQV8_"
      },
      "source": [
        "#### Load the embedding layer\n",
        "As part of the dataset preproceessing for this project, you will use a pre-trained English word embedding module from TensorFlow Hub. The URL for the module is https://tfhub.dev/google/tf2-preview/nnlm-en-dim128-with-normalization/1. This module has also been made available as a complete saved model in the folder `'./models/tf2-preview_nnlm-en-dim128_1'`. \n",
        "\n",
        "This embedding takes a batch of text tokens in a 1-D tensor of strings as input. It then embeds the separate tokens into a 128-dimensional space. \n",
        "\n",
        "The code to load and test the embedding layer is provided for you below.\n",
        "\n",
        "**NB:** this model can also be used as a sentence embedding module. The module will process each token by removing punctuation and splitting on spaces. It then averages the word embeddings over a sentence to give a single embedding vector. However, we will use it only as a word embedding module, and will pass each word in the input sentence as a separate token."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QzHahH0FQV8_"
      },
      "source": [
        "# Load embedding module from Tensorflow Hub\n",
        "\n",
        "import tensorflow_hub as hub\n",
        "\n",
        "embedding_layer = hub.KerasLayer(\"https://tfhub.dev/google/tf2-preview/nnlm-en-dim128-with-normalization/1\", output_shape=[128],\n",
        "                           input_shape=[], dtype=tf.string)"
      ],
      "execution_count": 73,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rcsb5IUKQV8_",
        "outputId": "4886f6c7-f1c6-4f33-854f-51961c4d8d06"
      },
      "source": [
        "# Test the layer\n",
        "\n",
        "embedding_layer(tf.constant([\"these\", \"aren't\", \"the\", \"droids\", \"you're\", \"looking\", \"for\"])).shape\n"
      ],
      "execution_count": 74,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "TensorShape([7, 128])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 74
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JTiy-vABQV9A"
      },
      "source": [
        "You should now prepare the training and validation Datasets.\n",
        "\n",
        "* Create a random training and validation set split of the data, reserving e.g. 20% of the data for validation (NB: each English dataset example is a single sentence string, and each German dataset example is a sequence of padded integer tokens).\n",
        "* Load the training and validation sets into a tf.data.Dataset object, passing in a tuple of English and German data for both training and validation sets.\n",
        "* Create a function to map over the datasets that splits each English sentence at spaces. Apply this function to both Dataset objects using the map method. _Hint: look at the tf.strings.split function._\n",
        "* Create a function to map over the datasets that embeds each sequence of English words using the loaded embedding layer/model. Apply this function to both Dataset objects using the map method.\n",
        "* Create a function to filter out dataset examples where the English sentence is more than 13 (embedded) tokens in length. Apply this function to both Dataset objects using the filter method.\n",
        "* Create a function to map over the datasets that pads each English sequence of embeddings with some distinct padding value before the sequence, so that each sequence is length 13. Apply this function to both Dataset objects using the map method. _Hint: look at the tf.pad function. You can extract a Tensor shape using tf.shape; you might also find the tf.math.maximum function useful._\n",
        "* Batch both training and validation Datasets with a batch size of 16.\n",
        "* Print the `element_spec` property for the training and validation Datasets. \n",
        "* Using the Dataset `.take(1)` method, print the shape of the English data example from the training Dataset.\n",
        "* Using the Dataset `.take(1)` method, print the German data example Tensor from the validation Dataset."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "KBZMIJa4QV9A",
        "outputId": "f2d895f5-7d73-4426-db95-98b8265c9228"
      },
      "source": [
        "training_data, test_data, training_label, test_label = train_test_split(english_sentences, german_padded_sequences, test_size=0.2)\n",
        "print(training_data[0])\n",
        "print(training_label[0])\n"
      ],
      "execution_count": 75,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "do you smell that ?\n",
            "[   1 2016   13   11    7    2    0    0    0    0    0    0    0    0\n",
            "    0    0    0    0    0    0]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "s1WMBYOIQV9B"
      },
      "source": [
        "def make_dataset(input_array, target_array):\n",
        "    \"\"\"\n",
        "    This function takes two 2D numpy arrays in the first two arguments, and an integer\n",
        "    batch_size in the third argument. It should create and return a Dataset object \n",
        "    using the two numpy arrays and batch size according to the above specification.\n",
        "    \"\"\"\n",
        "    \n",
        "    dataset = tf.data.Dataset.from_tensor_slices((input_array, target_array))\n",
        "    #dataset = dataset.batch(batch_size, drop_remainder=True)\n",
        "    return dataset"
      ],
      "execution_count": 76,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-czOG5aRQV9B",
        "outputId": "ecd522c0-13f3-46b5-b346-16b6b7a7045e"
      },
      "source": [
        "\n",
        "train_dataset = make_dataset(training_data, training_label)\n",
        "val_dataset = make_dataset(test_data, test_label)\n",
        "\n",
        "\n",
        "\n",
        "def split(x, y):\n",
        "    split_x = tf.strings.split(x)\n",
        "    return (split_x, y)\n",
        "\n",
        "def embed(x, y):\n",
        "    return (embedding_layer(x), y)\n",
        "\n",
        "def filter_long(x, y):\n",
        "    return len(x) < 14\n",
        "\n",
        "\n",
        "\n",
        "def pad(x, y):\n",
        "    paddings = tf.constant([[13, 0], [0, 0]])\n",
        "    new_x = tf.pad(x, paddings, \"CONSTANT\")\n",
        "    return (new_x[-13:,:], y)\n",
        "    \n",
        "split_train_dataset = train_dataset.map(split)\n",
        "split_val_dataset = val_dataset.map(split)\n",
        "\n",
        "embed_train_dataset = split_train_dataset.map(embed)\n",
        "embed_val_dataset = split_val_dataset.map(embed)\n",
        "\n",
        "filter_train_dataset = embed_train_dataset.filter(filter_long)\n",
        "filter_val_dataset = embed_val_dataset.filter(filter_long)\n",
        "\n",
        "\n",
        "pad_train_dataset = filter_train_dataset.map(pad)\n",
        "pad_val_dataset = filter_val_dataset.map(pad)\n",
        "\n",
        "batch_train = pad_train_dataset.batch(16, drop_remainder=True)\n",
        "batch_val = pad_val_dataset.batch(16, drop_remainder=True)\n",
        "\n",
        "print(batch_train.element_spec)\n",
        "print(batch_val.element_spec)\n",
        "\n",
        "for ds in batch_train.take(1):\n",
        "    print(\"English training example shape:\", ds[0].shape)\n",
        "\n",
        "for ds in batch_val.take(1):\n",
        "    print(\"German validation example shape:\", ds[1].shape)\n",
        "  "
      ],
      "execution_count": 77,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(TensorSpec(shape=(16, None, 128), dtype=tf.float32, name=None), TensorSpec(shape=(16, 20), dtype=tf.int32, name=None))\n",
            "(TensorSpec(shape=(16, None, 128), dtype=tf.float32, name=None), TensorSpec(shape=(16, 20), dtype=tf.int32, name=None))\n",
            "English training example shape: (16, 13, 128)\n",
            "German validation example shape: (16, 20)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-_lKGSIRQV9B"
      },
      "source": [
        ""
      ],
      "execution_count": 77,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0IRdRBT-QV9B"
      },
      "source": [
        ""
      ],
      "execution_count": 77,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OByUHHSiQV9C"
      },
      "source": [
        ""
      ],
      "execution_count": 77,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5f2R_AkDQV9D"
      },
      "source": [
        "## 3. Create the custom layer\n",
        "You will now create a custom layer to add the learned end token embedding to the encoder model:\n",
        "\n",
        "![Encoder schematic](data/neural_translation_model_encoder.png)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FJ8-9XefQV9D"
      },
      "source": [
        "You should now build the custom layer.\n",
        "* Using layer subclassing, create a custom layer that takes a batch of English data examples from one of the Datasets, and adds a learned embedded ‘end’ token to the end of each sequence. \n",
        "* This layer should create a TensorFlow Variable (that will be learned during training) that is 128-dimensional (the size of the embedding space). _Hint: you may find it helpful in the call method to use the tf.tile function to replicate the end token embedding across every element in the batch._\n",
        "* Using the Dataset `.take(1)` method, extract a batch of English data examples from the training Dataset and print the shape. Test the custom layer by calling the layer on the English data batch Tensor and print the resulting Tensor shape (the layer should increase the sequence length by one)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bo3R-sGYQV9H"
      },
      "source": [
        "class EndToken(Layer):\n",
        "    def __init__(self):\n",
        "        super(EndToken, self).__init__()\n",
        "        init = tf.random_normal_initializer()\n",
        "        self.end_token = tf.Variable(initial_value=init(shape=(1, 1, 128)))\n",
        "    def call(self, inputs):\n",
        "        #print(\"Inputs shape:\", inputs.shape)\n",
        "        batch_size = 16\n",
        "        if inputs.shape[0]:\n",
        "            batch_size = inputs.shape[0]\n",
        "        #print(self.end_token.shape)\n",
        "        tile_end_token = tf.tile(self.end_token, [batch_size, 1, 1])\n",
        "        concat_result = tf.concat([inputs, tile_end_token], axis=1)\n",
        "        return concat_result\n",
        "                                     "
      ],
      "execution_count": 78,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-NugqzowQV9H",
        "outputId": "b4af73a1-7f44-44ee-c5ea-1d84025c0248"
      },
      "source": [
        "et = EndToken()\n",
        "for ds_eng, ds_deu in batch_train.take(1):\n",
        "    print(\"English training example shape:\", ds_eng.shape)\n",
        "    output = et(ds_eng)\n",
        "    print(\"Shape after end token concatenation\", output.shape)\n",
        "    "
      ],
      "execution_count": 79,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "English training example shape: (16, 13, 128)\n",
            "Shape after end token concatenation (16, 14, 128)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yl9EMUqdQV9H"
      },
      "source": [
        ""
      ],
      "execution_count": 79,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hHqG6_xfQV9H"
      },
      "source": [
        "## 4. Build the encoder network\n",
        "The encoder network follows the schematic diagram above. You should now build the RNN encoder model.\n",
        "* Using the functional API, build the encoder network according to the following spec:\n",
        "    * The model will take a batch of sequences of embedded English words as input, as given by the Dataset objects.\n",
        "    * The next layer in the encoder will be the custom layer you created previously, to add a learned end token embedding to the end of the English sequence.\n",
        "    * This is followed by a Masking layer, with the `mask_value` set to the distinct padding value you used when you padded the English sequences with the Dataset preprocessing above.\n",
        "    * The final layer is an LSTM layer with 512 units, which also returns the hidden and cell states.\n",
        "    * The encoder is a multi-output model. There should be two output Tensors of this model: the hidden state and cell states of the LSTM layer. The output of the LSTM layer is unused.\n",
        "* Using the Dataset `.take(1)` method, extract a batch of English data examples from the training Dataset and test the encoder model by calling it on the English data Tensor, and print the shape of the resulting Tensor outputs.\n",
        "* Print the model summary for the encoder network."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "vtUQxx_oQV9I",
        "outputId": "b5e92520-10f3-410a-ca6f-68c8e91981bb"
      },
      "source": [
        "inputs = Input(shape=(13, 128))\n",
        "output = EndToken()(inputs)\n",
        "output = Masking()(output)\n",
        "output = LSTM(512, return_state=True)(output)\n",
        "\n",
        "encoder = Model(inputs=inputs, outputs=output, name=\"encoder\")\n",
        "encoder.summary()\n",
        "\n",
        "for ds_eng, ds_deu in batch_train.take(1):\n",
        "    final_output, final_eng_hidden_state, final_eng_cell_state = encoder(ds_eng)\n",
        "    print(f'Unused final output step shape: {final_output.shape}')\n",
        "    print(f'Hidden state: {final_eng_hidden_state.shape}')\n",
        "    print(f'Cell state: {final_eng_cell_state.shape}')\n"
      ],
      "execution_count": 80,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"encoder\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "input_3 (InputLayer)         [(None, 13, 128)]         0         \n",
            "_________________________________________________________________\n",
            "end_token_5 (EndToken)       (16, 14, 128)             128       \n",
            "_________________________________________________________________\n",
            "masking_2 (Masking)          (16, 14, 128)             0         \n",
            "_________________________________________________________________\n",
            "lstm_6 (LSTM)                [(16, 512), (16, 512), (1 1312768   \n",
            "=================================================================\n",
            "Total params: 1,312,896\n",
            "Trainable params: 1,312,896\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "Unused final output step shape: (16, 512)\n",
            "Hidden state: (16, 512)\n",
            "Cell state: (16, 512)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-M3puGTuQV9I"
      },
      "source": [
        ""
      ],
      "execution_count": 80,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SNLYo5fnQV9I"
      },
      "source": [
        ""
      ],
      "execution_count": 80,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MPbHNwsiQV9I"
      },
      "source": [
        "## 5. Build the decoder network\n",
        "The decoder network follows the schematic diagram below. \n",
        "\n",
        "![Decoder schematic](data/neural_translation_model_decoder.png)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3Dr89ymfQV9I"
      },
      "source": [
        "You should now build the RNN decoder model.\n",
        "* Using Model subclassing, build the decoder network according to the following spec:\n",
        "    * The initializer should create the following layers:\n",
        "        * An Embedding layer with vocabulary size set to the number of unique German tokens, embedding dimension 128, and set to mask zero values in the input.\n",
        "        * An LSTM layer with 512 units, that returns its hidden and cell states, and also returns sequences.\n",
        "        * A Dense layer with number of units equal to the number of unique German tokens, and no activation function.\n",
        "    * The call method should include the usual `inputs` argument, as well as the additional keyword arguments `hidden_state` and `cell_state`. The default value for these keyword arguments should be `None`.\n",
        "    * The call method should pass the inputs through the Embedding layer, and then through the LSTM layer. If the `hidden_state` and `cell_state` arguments are provided, these should be used for the initial state of the LSTM layer. _Hint: use the_ `initial_state` _keyword argument when calling the LSTM layer on its input._\n",
        "    * The call method should pass the LSTM output sequence through the Dense layer, and return the resulting Tensor, along with the hidden and cell states of the LSTM layer.\n",
        "* Using the Dataset `.take(1)` method, extract a batch of English and German data examples from the training Dataset. Test the decoder model by first calling the encoder model on the English data Tensor to get the hidden and cell states, and then call the decoder model on the German data Tensor and hidden and cell states, and print the shape of the resulting decoder Tensor outputs.\n",
        "* Print the model summary for the decoder network."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "g5MKRuV0QV9I"
      },
      "source": [
        "\n",
        "class Decoder(Model):\n",
        "    def __init__(self, **kwargs):\n",
        "        super(Decoder, self).__init__()\n",
        "        self.embedding = tf.keras.layers.Embedding(input_dim=deu_input_dim , output_dim=128, mask_zero=True)\n",
        "        self.lstm = LSTM(512, return_sequences=True, return_state=True)\n",
        "        self.dense = Dense(deu_input_dim)\n",
        "        \n",
        "    def call(self, inputs, hidden_state=None, cell_state=None):\n",
        "        embed_inputs = self.embedding(inputs)\n",
        "        lstm_output, current_hidden_state, current_cell_state = self.lstm(embed_inputs, initial_state=[hidden_state, cell_state])\n",
        "        outputs = self.dense(lstm_output)\n",
        "        return outputs, current_hidden_state, current_cell_state\n",
        "    \n",
        "    "
      ],
      "execution_count": 81,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zHIR8YuJQV9J",
        "outputId": "2ec6c14a-d39a-48b9-c905-285ea3e664c5"
      },
      "source": [
        "for ds_eng, ds_deu in batch_train.take(1):\n",
        "    output, final_eng_hidden_state, final_eng_cell_state = encoder(ds_eng)\n",
        "    d = Decoder(name=\"decoder\")\n",
        "    seq_output, hidden_state, cell_state = d(ds_deu, final_eng_hidden_state, final_eng_cell_state)\n",
        "    print(seq_output.shape)\n",
        "    d.summary()"
      ],
      "execution_count": 82,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(16, 20, 5744)\n",
            "Model: \"decoder_4\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_5 (Embedding)      multiple                  735232    \n",
            "_________________________________________________________________\n",
            "lstm_7 (LSTM)                multiple                  1312768   \n",
            "_________________________________________________________________\n",
            "dense_4 (Dense)              multiple                  2946672   \n",
            "=================================================================\n",
            "Total params: 4,994,672\n",
            "Trainable params: 4,994,672\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wdAh3u5LQV9J"
      },
      "source": [
        ""
      ],
      "execution_count": 82,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "s5_Ra7DEQV9J"
      },
      "source": [
        "## 6. Make a custom training loop\n",
        "You should now write a custom training loop to train your custom neural translation model.\n",
        "* Define a function that takes a Tensor batch of German data (as extracted from the training Dataset), and returns a tuple containing German inputs and outputs for the decoder model (refer to schematic diagram above).\n",
        "* Define a function that computes the forward and backward pass for your translation model. This function should take an English input, German input and German output as arguments, and should do the following:\n",
        "    * Pass the English input into the encoder, to get the hidden and cell states of the encoder LSTM.\n",
        "    * These hidden and cell states are then passed into the decoder, along with the German inputs, which returns a sequence of outputs (the hidden and cell state outputs of the decoder LSTM are unused in this function).\n",
        "    * The loss should then be computed between the decoder outputs and the German output function argument.\n",
        "    * The function returns the loss and gradients with respect to the encoder and decoder’s trainable variables.\n",
        "    * Decorate the function with @tf.function\n",
        "* Define and run a custom training loop for a number of epochs (for you to choose) that does the following:\n",
        "    * Iterates through the training dataset, and creates decoder inputs and outputs from the German sequences.\n",
        "    * Updates the parameters of the translation model using the gradients of the function above and an optimizer object.\n",
        "    * Every epoch, compute the validation loss on a number of batches from the validation and save the epoch training and validation losses.\n",
        "* Plot the learning curves for loss vs epoch for both training and validation sets.\n",
        "\n",
        "_Hint: This model is computationally demanding to train. The quality of the model or length of training is not a factor in the grading rubric. However, to obtain a better model we recommend using the GPU accelerator hardware on Colab._"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fJOUwVUvQV9J"
      },
      "source": [
        "def prepare_train_data(deu_batch):\n",
        "    batch_inputs = []\n",
        "    batch_outputs = []\n",
        "    for row in deu_batch:\n",
        "        y = tf.where(tf.equal(row, 2)).numpy()[0][0]\n",
        "        row1 = row[:y]\n",
        "        row2 = row[1:y+1]\n",
        "        size = 20 - tf.size(row1).numpy()\n",
        "        paddings = tf.constant([[0, size]])\n",
        "        row1 = tf.pad(row1, paddings)\n",
        "        row2 = tf.pad(row2, paddings)\n",
        "        batch_inputs.append(row1)\n",
        "        batch_outputs.append(row2)\n",
        "    \n",
        "    batch_inputs = tf.stack(batch_inputs, 0)\n",
        "    batch_outputs = tf.stack(batch_outputs, 0)\n",
        "    return(batch_inputs, batch_outputs)\n",
        "\n",
        "for eng_data, deu_data in batch_train.take(1):\n",
        "    eng_input = eng_data\n",
        "    deu_input, deu_output = prepare_train_data(deu_data)\n",
        "    #print(deu_input.shape)\n",
        "    #print(deu_input)"
      ],
      "execution_count": 83,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WPfFpfChQV9J"
      },
      "source": [
        "def loss_function(real, pred):\n",
        "  cross_entropy = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True, reduction='none')\n",
        "  loss = cross_entropy(y_true=real, y_pred=pred)\n",
        "  mask = tf.logical_not(tf.math.equal(real,0))   #output 0 for y=0 else output 1\n",
        "  mask = tf.cast(mask, dtype=loss.dtype)  \n",
        "  loss = mask * loss\n",
        "  loss = tf.reduce_mean(loss)\n",
        "  return loss"
      ],
      "execution_count": 84,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZNqQODOkQV9K"
      },
      "source": [
        "#tf.config.experimental_run_functions_eagerly(True)\n",
        "decoder = Decoder(name=\"decoder\")\n",
        "optimizer = tf.keras.optimizers.Adam(learning_rate=0.001)\n",
        "\n",
        "@tf.function\n",
        "def compute_pass(eng_input, deu_input, deu_output, training=True):    \n",
        "    with tf.GradientTape() as tape:\n",
        "        discard_output_seq, eng_final_hidden_state, eng_final_cell_state = encoder(eng_input)\n",
        "        output_seq, hidden_state, cell_state = decoder(deu_input, eng_final_hidden_state, eng_final_cell_state)\n",
        "        loss = loss_function(deu_output, output_seq)\n",
        "    variables = encoder.trainable_variables + decoder.trainable_variables\n",
        "    gradients = tape.gradient(loss, variables)\n",
        "    if training:\n",
        "        optimizer.apply_gradients(zip(gradients, variables))\n",
        "\n",
        "    return loss, gradients\n",
        "\n"
      ],
      "execution_count": 85,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "L5TPG8TjQV9K"
      },
      "source": [
        ""
      ],
      "execution_count": 85,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PA_EBSerQV9K"
      },
      "source": [
        "\n",
        "def train_translator(num_epochs, train_dataset, val_dataset):\n",
        "    \"\"\"\n",
        "    Returns training and validation losses per epoch\n",
        "    \"\"\"\n",
        "    \n",
        "    train_epoch_loss = []\n",
        "    val_epoch_loss = []\n",
        "\n",
        "    \n",
        "    for _ in range(num_epochs):\n",
        "        train_batch_loss = []\n",
        "        val_batch_loss = []\n",
        "        # Iterate over the batches of the train dataset.\n",
        "        for step, (eng_data, deu_data) in enumerate(train_dataset):\n",
        "            eng_input = eng_data\n",
        "            deu_input, deu_output = prepare_train_data(deu_data)\n",
        "            current_loss, gradients = compute_pass(eng_input, deu_input, deu_output)\n",
        "            print(\"epoch:\", _, \"step:\", step, \"train batch loss:\", current_loss)\n",
        "            train_batch_loss.append(current_loss)\n",
        "\n",
        "        \n",
        "        # Iterate over the batches of the val dataset.\n",
        "        for step, (eng_data, deu_data) in enumerate(val_dataset):\n",
        "            eng_input = eng_data\n",
        "            deu_input, deu_output = prepare_train_data(deu_data)\n",
        "            current_loss, gradients = compute_pass(eng_input, deu_input, deu_output, training=False)\n",
        "            print(\"epoch:\", _, \"step:\", step, \"val batch loss:\", current_loss)\n",
        "            val_batch_loss.append(current_loss)\n",
        "\n",
        "            \n",
        "        train_mean_loss = np.mean(train_batch_loss)\n",
        "        val_mean_loss = np.mean(val_batch_loss)\n",
        "        train_epoch_loss.append(train_mean_loss)\n",
        "        val_epoch_loss.append(val_mean_loss)\n",
        "        print('Epoch {0} training loss: {1} val loss: {2}'.format(_, train_mean_loss, val_mean_loss))    \n",
        "        \n",
        "    return train_epoch_loss, val_epoch_loss\n",
        "\n",
        "        "
      ],
      "execution_count": 86,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "V0gUU7NIQV9K",
        "outputId": "1fbf488f-a727-494e-81dc-d6625daf619a"
      },
      "source": [
        "# Train the model for 6 epochs, since anything more overfits\n",
        "\n",
        "train_loss_results, val_loss_results = train_translator(6, batch_train, batch_val)"
      ],
      "execution_count": 87,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[1;30;43mStreaming output truncated to the last 5000 lines.\u001b[0m\n",
            "epoch: 2 step: 4 train batch loss: tf.Tensor(0.74590766, shape=(), dtype=float32)\n",
            "epoch: 2 step: 5 train batch loss: tf.Tensor(0.51901805, shape=(), dtype=float32)\n",
            "epoch: 2 step: 6 train batch loss: tf.Tensor(0.771515, shape=(), dtype=float32)\n",
            "epoch: 2 step: 7 train batch loss: tf.Tensor(0.6408268, shape=(), dtype=float32)\n",
            "epoch: 2 step: 8 train batch loss: tf.Tensor(0.72491056, shape=(), dtype=float32)\n",
            "epoch: 2 step: 9 train batch loss: tf.Tensor(0.7582003, shape=(), dtype=float32)\n",
            "epoch: 2 step: 10 train batch loss: tf.Tensor(0.6683379, shape=(), dtype=float32)\n",
            "epoch: 2 step: 11 train batch loss: tf.Tensor(0.69357353, shape=(), dtype=float32)\n",
            "epoch: 2 step: 12 train batch loss: tf.Tensor(0.7009182, shape=(), dtype=float32)\n",
            "epoch: 2 step: 13 train batch loss: tf.Tensor(0.6690576, shape=(), dtype=float32)\n",
            "epoch: 2 step: 14 train batch loss: tf.Tensor(0.64655364, shape=(), dtype=float32)\n",
            "epoch: 2 step: 15 train batch loss: tf.Tensor(0.6414668, shape=(), dtype=float32)\n",
            "epoch: 2 step: 16 train batch loss: tf.Tensor(0.6668853, shape=(), dtype=float32)\n",
            "epoch: 2 step: 17 train batch loss: tf.Tensor(0.61483276, shape=(), dtype=float32)\n",
            "epoch: 2 step: 18 train batch loss: tf.Tensor(0.6084663, shape=(), dtype=float32)\n",
            "epoch: 2 step: 19 train batch loss: tf.Tensor(0.6334621, shape=(), dtype=float32)\n",
            "epoch: 2 step: 20 train batch loss: tf.Tensor(0.77729356, shape=(), dtype=float32)\n",
            "epoch: 2 step: 21 train batch loss: tf.Tensor(0.64825124, shape=(), dtype=float32)\n",
            "epoch: 2 step: 22 train batch loss: tf.Tensor(0.5985406, shape=(), dtype=float32)\n",
            "epoch: 2 step: 23 train batch loss: tf.Tensor(0.6074593, shape=(), dtype=float32)\n",
            "epoch: 2 step: 24 train batch loss: tf.Tensor(0.7061551, shape=(), dtype=float32)\n",
            "epoch: 2 step: 25 train batch loss: tf.Tensor(0.5975976, shape=(), dtype=float32)\n",
            "epoch: 2 step: 26 train batch loss: tf.Tensor(0.6343023, shape=(), dtype=float32)\n",
            "epoch: 2 step: 27 train batch loss: tf.Tensor(0.5615056, shape=(), dtype=float32)\n",
            "epoch: 2 step: 28 train batch loss: tf.Tensor(0.68682927, shape=(), dtype=float32)\n",
            "epoch: 2 step: 29 train batch loss: tf.Tensor(0.6061746, shape=(), dtype=float32)\n",
            "epoch: 2 step: 30 train batch loss: tf.Tensor(0.63510126, shape=(), dtype=float32)\n",
            "epoch: 2 step: 31 train batch loss: tf.Tensor(0.70271224, shape=(), dtype=float32)\n",
            "epoch: 2 step: 32 train batch loss: tf.Tensor(0.60828775, shape=(), dtype=float32)\n",
            "epoch: 2 step: 33 train batch loss: tf.Tensor(0.56722134, shape=(), dtype=float32)\n",
            "epoch: 2 step: 34 train batch loss: tf.Tensor(0.6207294, shape=(), dtype=float32)\n",
            "epoch: 2 step: 35 train batch loss: tf.Tensor(0.63553834, shape=(), dtype=float32)\n",
            "epoch: 2 step: 36 train batch loss: tf.Tensor(0.51111877, shape=(), dtype=float32)\n",
            "epoch: 2 step: 37 train batch loss: tf.Tensor(0.5272464, shape=(), dtype=float32)\n",
            "epoch: 2 step: 38 train batch loss: tf.Tensor(0.6249968, shape=(), dtype=float32)\n",
            "epoch: 2 step: 39 train batch loss: tf.Tensor(0.6799102, shape=(), dtype=float32)\n",
            "epoch: 2 step: 40 train batch loss: tf.Tensor(0.8987713, shape=(), dtype=float32)\n",
            "epoch: 2 step: 41 train batch loss: tf.Tensor(0.7000259, shape=(), dtype=float32)\n",
            "epoch: 2 step: 42 train batch loss: tf.Tensor(0.7014442, shape=(), dtype=float32)\n",
            "epoch: 2 step: 43 train batch loss: tf.Tensor(0.6852208, shape=(), dtype=float32)\n",
            "epoch: 2 step: 44 train batch loss: tf.Tensor(0.5847746, shape=(), dtype=float32)\n",
            "epoch: 2 step: 45 train batch loss: tf.Tensor(0.58363104, shape=(), dtype=float32)\n",
            "epoch: 2 step: 46 train batch loss: tf.Tensor(0.64590776, shape=(), dtype=float32)\n",
            "epoch: 2 step: 47 train batch loss: tf.Tensor(0.5628696, shape=(), dtype=float32)\n",
            "epoch: 2 step: 48 train batch loss: tf.Tensor(0.713586, shape=(), dtype=float32)\n",
            "epoch: 2 step: 49 train batch loss: tf.Tensor(0.7569445, shape=(), dtype=float32)\n",
            "epoch: 2 step: 50 train batch loss: tf.Tensor(0.71130717, shape=(), dtype=float32)\n",
            "epoch: 2 step: 51 train batch loss: tf.Tensor(0.55697376, shape=(), dtype=float32)\n",
            "epoch: 2 step: 52 train batch loss: tf.Tensor(0.6864471, shape=(), dtype=float32)\n",
            "epoch: 2 step: 53 train batch loss: tf.Tensor(0.5545094, shape=(), dtype=float32)\n",
            "epoch: 2 step: 54 train batch loss: tf.Tensor(0.6775796, shape=(), dtype=float32)\n",
            "epoch: 2 step: 55 train batch loss: tf.Tensor(0.6691113, shape=(), dtype=float32)\n",
            "epoch: 2 step: 56 train batch loss: tf.Tensor(0.595103, shape=(), dtype=float32)\n",
            "epoch: 2 step: 57 train batch loss: tf.Tensor(0.5337542, shape=(), dtype=float32)\n",
            "epoch: 2 step: 58 train batch loss: tf.Tensor(0.60141426, shape=(), dtype=float32)\n",
            "epoch: 2 step: 59 train batch loss: tf.Tensor(0.6908583, shape=(), dtype=float32)\n",
            "epoch: 2 step: 60 train batch loss: tf.Tensor(0.6574131, shape=(), dtype=float32)\n",
            "epoch: 2 step: 61 train batch loss: tf.Tensor(0.6006509, shape=(), dtype=float32)\n",
            "epoch: 2 step: 62 train batch loss: tf.Tensor(0.69335145, shape=(), dtype=float32)\n",
            "epoch: 2 step: 63 train batch loss: tf.Tensor(0.53884923, shape=(), dtype=float32)\n",
            "epoch: 2 step: 64 train batch loss: tf.Tensor(0.70395577, shape=(), dtype=float32)\n",
            "epoch: 2 step: 65 train batch loss: tf.Tensor(0.6106206, shape=(), dtype=float32)\n",
            "epoch: 2 step: 66 train batch loss: tf.Tensor(0.6340249, shape=(), dtype=float32)\n",
            "epoch: 2 step: 67 train batch loss: tf.Tensor(0.53797126, shape=(), dtype=float32)\n",
            "epoch: 2 step: 68 train batch loss: tf.Tensor(0.7061814, shape=(), dtype=float32)\n",
            "epoch: 2 step: 69 train batch loss: tf.Tensor(0.73867905, shape=(), dtype=float32)\n",
            "epoch: 2 step: 70 train batch loss: tf.Tensor(0.65256006, shape=(), dtype=float32)\n",
            "epoch: 2 step: 71 train batch loss: tf.Tensor(0.63611287, shape=(), dtype=float32)\n",
            "epoch: 2 step: 72 train batch loss: tf.Tensor(0.79599583, shape=(), dtype=float32)\n",
            "epoch: 2 step: 73 train batch loss: tf.Tensor(0.5006245, shape=(), dtype=float32)\n",
            "epoch: 2 step: 74 train batch loss: tf.Tensor(0.7432554, shape=(), dtype=float32)\n",
            "epoch: 2 step: 75 train batch loss: tf.Tensor(0.60220146, shape=(), dtype=float32)\n",
            "epoch: 2 step: 76 train batch loss: tf.Tensor(0.6911801, shape=(), dtype=float32)\n",
            "epoch: 2 step: 77 train batch loss: tf.Tensor(0.65458864, shape=(), dtype=float32)\n",
            "epoch: 2 step: 78 train batch loss: tf.Tensor(0.53423226, shape=(), dtype=float32)\n",
            "epoch: 2 step: 79 train batch loss: tf.Tensor(0.5917716, shape=(), dtype=float32)\n",
            "epoch: 2 step: 80 train batch loss: tf.Tensor(0.5448852, shape=(), dtype=float32)\n",
            "epoch: 2 step: 81 train batch loss: tf.Tensor(0.6895763, shape=(), dtype=float32)\n",
            "epoch: 2 step: 82 train batch loss: tf.Tensor(0.67700684, shape=(), dtype=float32)\n",
            "epoch: 2 step: 83 train batch loss: tf.Tensor(0.63208187, shape=(), dtype=float32)\n",
            "epoch: 2 step: 84 train batch loss: tf.Tensor(0.72625864, shape=(), dtype=float32)\n",
            "epoch: 2 step: 85 train batch loss: tf.Tensor(0.6622075, shape=(), dtype=float32)\n",
            "epoch: 2 step: 86 train batch loss: tf.Tensor(0.7069099, shape=(), dtype=float32)\n",
            "epoch: 2 step: 87 train batch loss: tf.Tensor(0.6288119, shape=(), dtype=float32)\n",
            "epoch: 2 step: 88 train batch loss: tf.Tensor(0.56489646, shape=(), dtype=float32)\n",
            "epoch: 2 step: 89 train batch loss: tf.Tensor(0.62903893, shape=(), dtype=float32)\n",
            "epoch: 2 step: 90 train batch loss: tf.Tensor(0.590352, shape=(), dtype=float32)\n",
            "epoch: 2 step: 91 train batch loss: tf.Tensor(0.7478464, shape=(), dtype=float32)\n",
            "epoch: 2 step: 92 train batch loss: tf.Tensor(0.8570021, shape=(), dtype=float32)\n",
            "epoch: 2 step: 93 train batch loss: tf.Tensor(0.6077279, shape=(), dtype=float32)\n",
            "epoch: 2 step: 94 train batch loss: tf.Tensor(0.628292, shape=(), dtype=float32)\n",
            "epoch: 2 step: 95 train batch loss: tf.Tensor(0.58195245, shape=(), dtype=float32)\n",
            "epoch: 2 step: 96 train batch loss: tf.Tensor(0.57158315, shape=(), dtype=float32)\n",
            "epoch: 2 step: 97 train batch loss: tf.Tensor(0.56262773, shape=(), dtype=float32)\n",
            "epoch: 2 step: 98 train batch loss: tf.Tensor(0.5085374, shape=(), dtype=float32)\n",
            "epoch: 2 step: 99 train batch loss: tf.Tensor(0.6595602, shape=(), dtype=float32)\n",
            "epoch: 2 step: 100 train batch loss: tf.Tensor(0.6227617, shape=(), dtype=float32)\n",
            "epoch: 2 step: 101 train batch loss: tf.Tensor(0.6423379, shape=(), dtype=float32)\n",
            "epoch: 2 step: 102 train batch loss: tf.Tensor(0.73145187, shape=(), dtype=float32)\n",
            "epoch: 2 step: 103 train batch loss: tf.Tensor(0.56437826, shape=(), dtype=float32)\n",
            "epoch: 2 step: 104 train batch loss: tf.Tensor(0.6814287, shape=(), dtype=float32)\n",
            "epoch: 2 step: 105 train batch loss: tf.Tensor(0.59295356, shape=(), dtype=float32)\n",
            "epoch: 2 step: 106 train batch loss: tf.Tensor(0.6321906, shape=(), dtype=float32)\n",
            "epoch: 2 step: 107 train batch loss: tf.Tensor(0.76843554, shape=(), dtype=float32)\n",
            "epoch: 2 step: 108 train batch loss: tf.Tensor(0.5692603, shape=(), dtype=float32)\n",
            "epoch: 2 step: 109 train batch loss: tf.Tensor(0.6398445, shape=(), dtype=float32)\n",
            "epoch: 2 step: 110 train batch loss: tf.Tensor(0.6478973, shape=(), dtype=float32)\n",
            "epoch: 2 step: 111 train batch loss: tf.Tensor(0.62395084, shape=(), dtype=float32)\n",
            "epoch: 2 step: 112 train batch loss: tf.Tensor(0.6696726, shape=(), dtype=float32)\n",
            "epoch: 2 step: 113 train batch loss: tf.Tensor(0.58377016, shape=(), dtype=float32)\n",
            "epoch: 2 step: 114 train batch loss: tf.Tensor(0.58708084, shape=(), dtype=float32)\n",
            "epoch: 2 step: 115 train batch loss: tf.Tensor(0.5974447, shape=(), dtype=float32)\n",
            "epoch: 2 step: 116 train batch loss: tf.Tensor(0.527469, shape=(), dtype=float32)\n",
            "epoch: 2 step: 117 train batch loss: tf.Tensor(0.5973998, shape=(), dtype=float32)\n",
            "epoch: 2 step: 118 train batch loss: tf.Tensor(0.74559015, shape=(), dtype=float32)\n",
            "epoch: 2 step: 119 train batch loss: tf.Tensor(0.58073914, shape=(), dtype=float32)\n",
            "epoch: 2 step: 120 train batch loss: tf.Tensor(0.5618846, shape=(), dtype=float32)\n",
            "epoch: 2 step: 121 train batch loss: tf.Tensor(0.62463367, shape=(), dtype=float32)\n",
            "epoch: 2 step: 122 train batch loss: tf.Tensor(0.5478628, shape=(), dtype=float32)\n",
            "epoch: 2 step: 123 train batch loss: tf.Tensor(0.6520905, shape=(), dtype=float32)\n",
            "epoch: 2 step: 124 train batch loss: tf.Tensor(0.69435006, shape=(), dtype=float32)\n",
            "epoch: 2 step: 125 train batch loss: tf.Tensor(0.62449265, shape=(), dtype=float32)\n",
            "epoch: 2 step: 126 train batch loss: tf.Tensor(0.6345606, shape=(), dtype=float32)\n",
            "epoch: 2 step: 127 train batch loss: tf.Tensor(0.86184806, shape=(), dtype=float32)\n",
            "epoch: 2 step: 128 train batch loss: tf.Tensor(0.63841695, shape=(), dtype=float32)\n",
            "epoch: 2 step: 129 train batch loss: tf.Tensor(0.5789768, shape=(), dtype=float32)\n",
            "epoch: 2 step: 130 train batch loss: tf.Tensor(0.59602255, shape=(), dtype=float32)\n",
            "epoch: 2 step: 131 train batch loss: tf.Tensor(0.64197505, shape=(), dtype=float32)\n",
            "epoch: 2 step: 132 train batch loss: tf.Tensor(0.58346736, shape=(), dtype=float32)\n",
            "epoch: 2 step: 133 train batch loss: tf.Tensor(0.513263, shape=(), dtype=float32)\n",
            "epoch: 2 step: 134 train batch loss: tf.Tensor(0.5840923, shape=(), dtype=float32)\n",
            "epoch: 2 step: 135 train batch loss: tf.Tensor(0.6679395, shape=(), dtype=float32)\n",
            "epoch: 2 step: 136 train batch loss: tf.Tensor(0.7117127, shape=(), dtype=float32)\n",
            "epoch: 2 step: 137 train batch loss: tf.Tensor(0.60160214, shape=(), dtype=float32)\n",
            "epoch: 2 step: 138 train batch loss: tf.Tensor(0.6365956, shape=(), dtype=float32)\n",
            "epoch: 2 step: 139 train batch loss: tf.Tensor(0.53350675, shape=(), dtype=float32)\n",
            "epoch: 2 step: 140 train batch loss: tf.Tensor(0.59142625, shape=(), dtype=float32)\n",
            "epoch: 2 step: 141 train batch loss: tf.Tensor(0.6363872, shape=(), dtype=float32)\n",
            "epoch: 2 step: 142 train batch loss: tf.Tensor(0.6541256, shape=(), dtype=float32)\n",
            "epoch: 2 step: 143 train batch loss: tf.Tensor(0.7734664, shape=(), dtype=float32)\n",
            "epoch: 2 step: 144 train batch loss: tf.Tensor(0.6237786, shape=(), dtype=float32)\n",
            "epoch: 2 step: 145 train batch loss: tf.Tensor(0.6593355, shape=(), dtype=float32)\n",
            "epoch: 2 step: 146 train batch loss: tf.Tensor(0.59493977, shape=(), dtype=float32)\n",
            "epoch: 2 step: 147 train batch loss: tf.Tensor(0.6098364, shape=(), dtype=float32)\n",
            "epoch: 2 step: 148 train batch loss: tf.Tensor(0.51750785, shape=(), dtype=float32)\n",
            "epoch: 2 step: 149 train batch loss: tf.Tensor(0.5977878, shape=(), dtype=float32)\n",
            "epoch: 2 step: 150 train batch loss: tf.Tensor(0.5451553, shape=(), dtype=float32)\n",
            "epoch: 2 step: 151 train batch loss: tf.Tensor(0.5678236, shape=(), dtype=float32)\n",
            "epoch: 2 step: 152 train batch loss: tf.Tensor(0.6149507, shape=(), dtype=float32)\n",
            "epoch: 2 step: 153 train batch loss: tf.Tensor(0.5266794, shape=(), dtype=float32)\n",
            "epoch: 2 step: 154 train batch loss: tf.Tensor(0.63589334, shape=(), dtype=float32)\n",
            "epoch: 2 step: 155 train batch loss: tf.Tensor(0.56803286, shape=(), dtype=float32)\n",
            "epoch: 2 step: 156 train batch loss: tf.Tensor(0.49915504, shape=(), dtype=float32)\n",
            "epoch: 2 step: 157 train batch loss: tf.Tensor(0.5845234, shape=(), dtype=float32)\n",
            "epoch: 2 step: 158 train batch loss: tf.Tensor(0.6418906, shape=(), dtype=float32)\n",
            "epoch: 2 step: 159 train batch loss: tf.Tensor(0.6380822, shape=(), dtype=float32)\n",
            "epoch: 2 step: 160 train batch loss: tf.Tensor(0.7501358, shape=(), dtype=float32)\n",
            "epoch: 2 step: 161 train batch loss: tf.Tensor(0.600507, shape=(), dtype=float32)\n",
            "epoch: 2 step: 162 train batch loss: tf.Tensor(0.6451957, shape=(), dtype=float32)\n",
            "epoch: 2 step: 163 train batch loss: tf.Tensor(0.6338408, shape=(), dtype=float32)\n",
            "epoch: 2 step: 164 train batch loss: tf.Tensor(0.61794174, shape=(), dtype=float32)\n",
            "epoch: 2 step: 165 train batch loss: tf.Tensor(0.65123665, shape=(), dtype=float32)\n",
            "epoch: 2 step: 166 train batch loss: tf.Tensor(0.59782124, shape=(), dtype=float32)\n",
            "epoch: 2 step: 167 train batch loss: tf.Tensor(0.5238674, shape=(), dtype=float32)\n",
            "epoch: 2 step: 168 train batch loss: tf.Tensor(0.5842965, shape=(), dtype=float32)\n",
            "epoch: 2 step: 169 train batch loss: tf.Tensor(0.5851474, shape=(), dtype=float32)\n",
            "epoch: 2 step: 170 train batch loss: tf.Tensor(0.65162885, shape=(), dtype=float32)\n",
            "epoch: 2 step: 171 train batch loss: tf.Tensor(0.59675324, shape=(), dtype=float32)\n",
            "epoch: 2 step: 172 train batch loss: tf.Tensor(0.57857, shape=(), dtype=float32)\n",
            "epoch: 2 step: 173 train batch loss: tf.Tensor(0.6044201, shape=(), dtype=float32)\n",
            "epoch: 2 step: 174 train batch loss: tf.Tensor(0.5001001, shape=(), dtype=float32)\n",
            "epoch: 2 step: 175 train batch loss: tf.Tensor(0.76262033, shape=(), dtype=float32)\n",
            "epoch: 2 step: 176 train batch loss: tf.Tensor(0.6333906, shape=(), dtype=float32)\n",
            "epoch: 2 step: 177 train batch loss: tf.Tensor(0.70839286, shape=(), dtype=float32)\n",
            "epoch: 2 step: 178 train batch loss: tf.Tensor(0.5726441, shape=(), dtype=float32)\n",
            "epoch: 2 step: 179 train batch loss: tf.Tensor(0.59517115, shape=(), dtype=float32)\n",
            "epoch: 2 step: 180 train batch loss: tf.Tensor(0.5918581, shape=(), dtype=float32)\n",
            "epoch: 2 step: 181 train batch loss: tf.Tensor(0.5731786, shape=(), dtype=float32)\n",
            "epoch: 2 step: 182 train batch loss: tf.Tensor(0.5128114, shape=(), dtype=float32)\n",
            "epoch: 2 step: 183 train batch loss: tf.Tensor(0.50619906, shape=(), dtype=float32)\n",
            "epoch: 2 step: 184 train batch loss: tf.Tensor(0.54662377, shape=(), dtype=float32)\n",
            "epoch: 2 step: 185 train batch loss: tf.Tensor(0.72465813, shape=(), dtype=float32)\n",
            "epoch: 2 step: 186 train batch loss: tf.Tensor(0.63437176, shape=(), dtype=float32)\n",
            "epoch: 2 step: 187 train batch loss: tf.Tensor(0.5927954, shape=(), dtype=float32)\n",
            "epoch: 2 step: 188 train batch loss: tf.Tensor(0.6127771, shape=(), dtype=float32)\n",
            "epoch: 2 step: 189 train batch loss: tf.Tensor(0.59205455, shape=(), dtype=float32)\n",
            "epoch: 2 step: 190 train batch loss: tf.Tensor(0.67704767, shape=(), dtype=float32)\n",
            "epoch: 2 step: 191 train batch loss: tf.Tensor(0.53705704, shape=(), dtype=float32)\n",
            "epoch: 2 step: 192 train batch loss: tf.Tensor(0.6085099, shape=(), dtype=float32)\n",
            "epoch: 2 step: 193 train batch loss: tf.Tensor(0.6253922, shape=(), dtype=float32)\n",
            "epoch: 2 step: 194 train batch loss: tf.Tensor(0.48796415, shape=(), dtype=float32)\n",
            "epoch: 2 step: 195 train batch loss: tf.Tensor(0.734258, shape=(), dtype=float32)\n",
            "epoch: 2 step: 196 train batch loss: tf.Tensor(0.5597696, shape=(), dtype=float32)\n",
            "epoch: 2 step: 197 train batch loss: tf.Tensor(0.61254096, shape=(), dtype=float32)\n",
            "epoch: 2 step: 198 train batch loss: tf.Tensor(0.63964427, shape=(), dtype=float32)\n",
            "epoch: 2 step: 199 train batch loss: tf.Tensor(0.512633, shape=(), dtype=float32)\n",
            "epoch: 2 step: 200 train batch loss: tf.Tensor(0.58878213, shape=(), dtype=float32)\n",
            "epoch: 2 step: 201 train batch loss: tf.Tensor(0.57502866, shape=(), dtype=float32)\n",
            "epoch: 2 step: 202 train batch loss: tf.Tensor(0.9191715, shape=(), dtype=float32)\n",
            "epoch: 2 step: 203 train batch loss: tf.Tensor(0.5449168, shape=(), dtype=float32)\n",
            "epoch: 2 step: 204 train batch loss: tf.Tensor(0.6737134, shape=(), dtype=float32)\n",
            "epoch: 2 step: 205 train batch loss: tf.Tensor(0.60405225, shape=(), dtype=float32)\n",
            "epoch: 2 step: 206 train batch loss: tf.Tensor(0.6509937, shape=(), dtype=float32)\n",
            "epoch: 2 step: 207 train batch loss: tf.Tensor(0.580491, shape=(), dtype=float32)\n",
            "epoch: 2 step: 208 train batch loss: tf.Tensor(0.6332531, shape=(), dtype=float32)\n",
            "epoch: 2 step: 209 train batch loss: tf.Tensor(0.7578624, shape=(), dtype=float32)\n",
            "epoch: 2 step: 210 train batch loss: tf.Tensor(0.6413678, shape=(), dtype=float32)\n",
            "epoch: 2 step: 211 train batch loss: tf.Tensor(0.5839123, shape=(), dtype=float32)\n",
            "epoch: 2 step: 212 train batch loss: tf.Tensor(0.5455502, shape=(), dtype=float32)\n",
            "epoch: 2 step: 213 train batch loss: tf.Tensor(0.56996834, shape=(), dtype=float32)\n",
            "epoch: 2 step: 214 train batch loss: tf.Tensor(0.6063338, shape=(), dtype=float32)\n",
            "epoch: 2 step: 215 train batch loss: tf.Tensor(0.48885292, shape=(), dtype=float32)\n",
            "epoch: 2 step: 216 train batch loss: tf.Tensor(0.87388057, shape=(), dtype=float32)\n",
            "epoch: 2 step: 217 train batch loss: tf.Tensor(0.51595074, shape=(), dtype=float32)\n",
            "epoch: 2 step: 218 train batch loss: tf.Tensor(0.58748, shape=(), dtype=float32)\n",
            "epoch: 2 step: 219 train batch loss: tf.Tensor(0.5223167, shape=(), dtype=float32)\n",
            "epoch: 2 step: 220 train batch loss: tf.Tensor(0.5610249, shape=(), dtype=float32)\n",
            "epoch: 2 step: 221 train batch loss: tf.Tensor(0.5560846, shape=(), dtype=float32)\n",
            "epoch: 2 step: 222 train batch loss: tf.Tensor(0.56009597, shape=(), dtype=float32)\n",
            "epoch: 2 step: 223 train batch loss: tf.Tensor(0.6229101, shape=(), dtype=float32)\n",
            "epoch: 2 step: 224 train batch loss: tf.Tensor(0.6893416, shape=(), dtype=float32)\n",
            "epoch: 2 step: 225 train batch loss: tf.Tensor(0.5181086, shape=(), dtype=float32)\n",
            "epoch: 2 step: 226 train batch loss: tf.Tensor(0.4817074, shape=(), dtype=float32)\n",
            "epoch: 2 step: 227 train batch loss: tf.Tensor(0.59998626, shape=(), dtype=float32)\n",
            "epoch: 2 step: 228 train batch loss: tf.Tensor(0.5827621, shape=(), dtype=float32)\n",
            "epoch: 2 step: 229 train batch loss: tf.Tensor(0.60821044, shape=(), dtype=float32)\n",
            "epoch: 2 step: 230 train batch loss: tf.Tensor(0.48208755, shape=(), dtype=float32)\n",
            "epoch: 2 step: 231 train batch loss: tf.Tensor(0.51671684, shape=(), dtype=float32)\n",
            "epoch: 2 step: 232 train batch loss: tf.Tensor(0.69338405, shape=(), dtype=float32)\n",
            "epoch: 2 step: 233 train batch loss: tf.Tensor(0.7650181, shape=(), dtype=float32)\n",
            "epoch: 2 step: 234 train batch loss: tf.Tensor(0.61635315, shape=(), dtype=float32)\n",
            "epoch: 2 step: 235 train batch loss: tf.Tensor(0.5837927, shape=(), dtype=float32)\n",
            "epoch: 2 step: 236 train batch loss: tf.Tensor(0.7224446, shape=(), dtype=float32)\n",
            "epoch: 2 step: 237 train batch loss: tf.Tensor(0.6907746, shape=(), dtype=float32)\n",
            "epoch: 2 step: 238 train batch loss: tf.Tensor(0.649576, shape=(), dtype=float32)\n",
            "epoch: 2 step: 239 train batch loss: tf.Tensor(0.6035257, shape=(), dtype=float32)\n",
            "epoch: 2 step: 240 train batch loss: tf.Tensor(0.6495986, shape=(), dtype=float32)\n",
            "epoch: 2 step: 241 train batch loss: tf.Tensor(0.74092805, shape=(), dtype=float32)\n",
            "epoch: 2 step: 242 train batch loss: tf.Tensor(0.54971635, shape=(), dtype=float32)\n",
            "epoch: 2 step: 243 train batch loss: tf.Tensor(0.49167997, shape=(), dtype=float32)\n",
            "epoch: 2 step: 244 train batch loss: tf.Tensor(0.68569005, shape=(), dtype=float32)\n",
            "epoch: 2 step: 245 train batch loss: tf.Tensor(0.59181, shape=(), dtype=float32)\n",
            "epoch: 2 step: 246 train batch loss: tf.Tensor(0.474263, shape=(), dtype=float32)\n",
            "epoch: 2 step: 247 train batch loss: tf.Tensor(0.61724913, shape=(), dtype=float32)\n",
            "epoch: 2 step: 248 train batch loss: tf.Tensor(0.6416446, shape=(), dtype=float32)\n",
            "epoch: 2 step: 249 train batch loss: tf.Tensor(0.5097802, shape=(), dtype=float32)\n",
            "epoch: 2 step: 250 train batch loss: tf.Tensor(0.6278137, shape=(), dtype=float32)\n",
            "epoch: 2 step: 251 train batch loss: tf.Tensor(0.65468633, shape=(), dtype=float32)\n",
            "epoch: 2 step: 252 train batch loss: tf.Tensor(0.77694154, shape=(), dtype=float32)\n",
            "epoch: 2 step: 253 train batch loss: tf.Tensor(0.46787015, shape=(), dtype=float32)\n",
            "epoch: 2 step: 254 train batch loss: tf.Tensor(0.58875936, shape=(), dtype=float32)\n",
            "epoch: 2 step: 255 train batch loss: tf.Tensor(0.66221964, shape=(), dtype=float32)\n",
            "epoch: 2 step: 256 train batch loss: tf.Tensor(0.53191745, shape=(), dtype=float32)\n",
            "epoch: 2 step: 257 train batch loss: tf.Tensor(0.55978787, shape=(), dtype=float32)\n",
            "epoch: 2 step: 258 train batch loss: tf.Tensor(0.4904236, shape=(), dtype=float32)\n",
            "epoch: 2 step: 259 train batch loss: tf.Tensor(0.58089745, shape=(), dtype=float32)\n",
            "epoch: 2 step: 260 train batch loss: tf.Tensor(0.62331724, shape=(), dtype=float32)\n",
            "epoch: 2 step: 261 train batch loss: tf.Tensor(0.60441613, shape=(), dtype=float32)\n",
            "epoch: 2 step: 262 train batch loss: tf.Tensor(0.7367321, shape=(), dtype=float32)\n",
            "epoch: 2 step: 263 train batch loss: tf.Tensor(0.64025176, shape=(), dtype=float32)\n",
            "epoch: 2 step: 264 train batch loss: tf.Tensor(0.6866531, shape=(), dtype=float32)\n",
            "epoch: 2 step: 265 train batch loss: tf.Tensor(0.5822623, shape=(), dtype=float32)\n",
            "epoch: 2 step: 266 train batch loss: tf.Tensor(0.6479882, shape=(), dtype=float32)\n",
            "epoch: 2 step: 267 train batch loss: tf.Tensor(0.521978, shape=(), dtype=float32)\n",
            "epoch: 2 step: 268 train batch loss: tf.Tensor(0.5011942, shape=(), dtype=float32)\n",
            "epoch: 2 step: 269 train batch loss: tf.Tensor(0.5487314, shape=(), dtype=float32)\n",
            "epoch: 2 step: 270 train batch loss: tf.Tensor(0.6100079, shape=(), dtype=float32)\n",
            "epoch: 2 step: 271 train batch loss: tf.Tensor(0.68874174, shape=(), dtype=float32)\n",
            "epoch: 2 step: 272 train batch loss: tf.Tensor(0.5499233, shape=(), dtype=float32)\n",
            "epoch: 2 step: 273 train batch loss: tf.Tensor(0.6056514, shape=(), dtype=float32)\n",
            "epoch: 2 step: 274 train batch loss: tf.Tensor(0.9067772, shape=(), dtype=float32)\n",
            "epoch: 2 step: 275 train batch loss: tf.Tensor(0.59573776, shape=(), dtype=float32)\n",
            "epoch: 2 step: 276 train batch loss: tf.Tensor(0.86281186, shape=(), dtype=float32)\n",
            "epoch: 2 step: 277 train batch loss: tf.Tensor(0.6974353, shape=(), dtype=float32)\n",
            "epoch: 2 step: 278 train batch loss: tf.Tensor(0.7247463, shape=(), dtype=float32)\n",
            "epoch: 2 step: 279 train batch loss: tf.Tensor(0.56873465, shape=(), dtype=float32)\n",
            "epoch: 2 step: 280 train batch loss: tf.Tensor(0.5548423, shape=(), dtype=float32)\n",
            "epoch: 2 step: 281 train batch loss: tf.Tensor(0.648942, shape=(), dtype=float32)\n",
            "epoch: 2 step: 282 train batch loss: tf.Tensor(0.582917, shape=(), dtype=float32)\n",
            "epoch: 2 step: 283 train batch loss: tf.Tensor(0.63125765, shape=(), dtype=float32)\n",
            "epoch: 2 step: 284 train batch loss: tf.Tensor(0.56390184, shape=(), dtype=float32)\n",
            "epoch: 2 step: 285 train batch loss: tf.Tensor(0.5456871, shape=(), dtype=float32)\n",
            "epoch: 2 step: 286 train batch loss: tf.Tensor(0.50269496, shape=(), dtype=float32)\n",
            "epoch: 2 step: 287 train batch loss: tf.Tensor(0.55998117, shape=(), dtype=float32)\n",
            "epoch: 2 step: 288 train batch loss: tf.Tensor(0.5836755, shape=(), dtype=float32)\n",
            "epoch: 2 step: 289 train batch loss: tf.Tensor(0.49013478, shape=(), dtype=float32)\n",
            "epoch: 2 step: 290 train batch loss: tf.Tensor(0.6401894, shape=(), dtype=float32)\n",
            "epoch: 2 step: 291 train batch loss: tf.Tensor(0.71732605, shape=(), dtype=float32)\n",
            "epoch: 2 step: 292 train batch loss: tf.Tensor(0.72891915, shape=(), dtype=float32)\n",
            "epoch: 2 step: 293 train batch loss: tf.Tensor(0.740291, shape=(), dtype=float32)\n",
            "epoch: 2 step: 294 train batch loss: tf.Tensor(0.60900116, shape=(), dtype=float32)\n",
            "epoch: 2 step: 295 train batch loss: tf.Tensor(0.6616687, shape=(), dtype=float32)\n",
            "epoch: 2 step: 296 train batch loss: tf.Tensor(0.5173693, shape=(), dtype=float32)\n",
            "epoch: 2 step: 297 train batch loss: tf.Tensor(0.6524669, shape=(), dtype=float32)\n",
            "epoch: 2 step: 298 train batch loss: tf.Tensor(0.6037369, shape=(), dtype=float32)\n",
            "epoch: 2 step: 299 train batch loss: tf.Tensor(0.5690112, shape=(), dtype=float32)\n",
            "epoch: 2 step: 300 train batch loss: tf.Tensor(0.6285838, shape=(), dtype=float32)\n",
            "epoch: 2 step: 301 train batch loss: tf.Tensor(0.5590255, shape=(), dtype=float32)\n",
            "epoch: 2 step: 302 train batch loss: tf.Tensor(0.48095536, shape=(), dtype=float32)\n",
            "epoch: 2 step: 303 train batch loss: tf.Tensor(0.5641804, shape=(), dtype=float32)\n",
            "epoch: 2 step: 304 train batch loss: tf.Tensor(0.61252725, shape=(), dtype=float32)\n",
            "epoch: 2 step: 305 train batch loss: tf.Tensor(0.61618817, shape=(), dtype=float32)\n",
            "epoch: 2 step: 306 train batch loss: tf.Tensor(0.5462464, shape=(), dtype=float32)\n",
            "epoch: 2 step: 307 train batch loss: tf.Tensor(0.5640742, shape=(), dtype=float32)\n",
            "epoch: 2 step: 308 train batch loss: tf.Tensor(0.58921593, shape=(), dtype=float32)\n",
            "epoch: 2 step: 309 train batch loss: tf.Tensor(0.6059842, shape=(), dtype=float32)\n",
            "epoch: 2 step: 310 train batch loss: tf.Tensor(0.6858498, shape=(), dtype=float32)\n",
            "epoch: 2 step: 311 train batch loss: tf.Tensor(0.690724, shape=(), dtype=float32)\n",
            "epoch: 2 step: 312 train batch loss: tf.Tensor(0.69331926, shape=(), dtype=float32)\n",
            "epoch: 2 step: 313 train batch loss: tf.Tensor(0.53958774, shape=(), dtype=float32)\n",
            "epoch: 2 step: 314 train batch loss: tf.Tensor(0.6189055, shape=(), dtype=float32)\n",
            "epoch: 2 step: 315 train batch loss: tf.Tensor(0.50867987, shape=(), dtype=float32)\n",
            "epoch: 2 step: 316 train batch loss: tf.Tensor(0.46375936, shape=(), dtype=float32)\n",
            "epoch: 2 step: 317 train batch loss: tf.Tensor(0.5760252, shape=(), dtype=float32)\n",
            "epoch: 2 step: 318 train batch loss: tf.Tensor(0.5945388, shape=(), dtype=float32)\n",
            "epoch: 2 step: 319 train batch loss: tf.Tensor(0.6226152, shape=(), dtype=float32)\n",
            "epoch: 2 step: 320 train batch loss: tf.Tensor(0.50833946, shape=(), dtype=float32)\n",
            "epoch: 2 step: 321 train batch loss: tf.Tensor(0.6327058, shape=(), dtype=float32)\n",
            "epoch: 2 step: 322 train batch loss: tf.Tensor(0.5404784, shape=(), dtype=float32)\n",
            "epoch: 2 step: 323 train batch loss: tf.Tensor(0.6216372, shape=(), dtype=float32)\n",
            "epoch: 2 step: 324 train batch loss: tf.Tensor(0.6792594, shape=(), dtype=float32)\n",
            "epoch: 2 step: 325 train batch loss: tf.Tensor(0.70356196, shape=(), dtype=float32)\n",
            "epoch: 2 step: 326 train batch loss: tf.Tensor(0.5502599, shape=(), dtype=float32)\n",
            "epoch: 2 step: 327 train batch loss: tf.Tensor(0.75186956, shape=(), dtype=float32)\n",
            "epoch: 2 step: 328 train batch loss: tf.Tensor(0.61963785, shape=(), dtype=float32)\n",
            "epoch: 2 step: 329 train batch loss: tf.Tensor(0.57826173, shape=(), dtype=float32)\n",
            "epoch: 2 step: 330 train batch loss: tf.Tensor(0.59116775, shape=(), dtype=float32)\n",
            "epoch: 2 step: 331 train batch loss: tf.Tensor(0.5008081, shape=(), dtype=float32)\n",
            "epoch: 2 step: 332 train batch loss: tf.Tensor(0.60921276, shape=(), dtype=float32)\n",
            "epoch: 2 step: 333 train batch loss: tf.Tensor(0.785827, shape=(), dtype=float32)\n",
            "epoch: 2 step: 334 train batch loss: tf.Tensor(0.5373906, shape=(), dtype=float32)\n",
            "epoch: 2 step: 335 train batch loss: tf.Tensor(0.5153861, shape=(), dtype=float32)\n",
            "epoch: 2 step: 336 train batch loss: tf.Tensor(0.71685326, shape=(), dtype=float32)\n",
            "epoch: 2 step: 337 train batch loss: tf.Tensor(0.51721805, shape=(), dtype=float32)\n",
            "epoch: 2 step: 338 train batch loss: tf.Tensor(0.5313177, shape=(), dtype=float32)\n",
            "epoch: 2 step: 339 train batch loss: tf.Tensor(0.5585784, shape=(), dtype=float32)\n",
            "epoch: 2 step: 340 train batch loss: tf.Tensor(0.6097011, shape=(), dtype=float32)\n",
            "epoch: 2 step: 341 train batch loss: tf.Tensor(0.5718719, shape=(), dtype=float32)\n",
            "epoch: 2 step: 342 train batch loss: tf.Tensor(0.54291236, shape=(), dtype=float32)\n",
            "epoch: 2 step: 343 train batch loss: tf.Tensor(0.47303462, shape=(), dtype=float32)\n",
            "epoch: 2 step: 344 train batch loss: tf.Tensor(0.5903381, shape=(), dtype=float32)\n",
            "epoch: 2 step: 345 train batch loss: tf.Tensor(0.64992845, shape=(), dtype=float32)\n",
            "epoch: 2 step: 346 train batch loss: tf.Tensor(0.61103785, shape=(), dtype=float32)\n",
            "epoch: 2 step: 347 train batch loss: tf.Tensor(0.7059864, shape=(), dtype=float32)\n",
            "epoch: 2 step: 348 train batch loss: tf.Tensor(0.5401639, shape=(), dtype=float32)\n",
            "epoch: 2 step: 349 train batch loss: tf.Tensor(0.47621217, shape=(), dtype=float32)\n",
            "epoch: 2 step: 350 train batch loss: tf.Tensor(0.63175356, shape=(), dtype=float32)\n",
            "epoch: 2 step: 351 train batch loss: tf.Tensor(0.49574137, shape=(), dtype=float32)\n",
            "epoch: 2 step: 352 train batch loss: tf.Tensor(0.5285044, shape=(), dtype=float32)\n",
            "epoch: 2 step: 353 train batch loss: tf.Tensor(0.68021077, shape=(), dtype=float32)\n",
            "epoch: 2 step: 354 train batch loss: tf.Tensor(0.53604126, shape=(), dtype=float32)\n",
            "epoch: 2 step: 355 train batch loss: tf.Tensor(0.4907357, shape=(), dtype=float32)\n",
            "epoch: 2 step: 356 train batch loss: tf.Tensor(0.6501881, shape=(), dtype=float32)\n",
            "epoch: 2 step: 357 train batch loss: tf.Tensor(0.6448839, shape=(), dtype=float32)\n",
            "epoch: 2 step: 358 train batch loss: tf.Tensor(0.6629195, shape=(), dtype=float32)\n",
            "epoch: 2 step: 359 train batch loss: tf.Tensor(0.5705495, shape=(), dtype=float32)\n",
            "epoch: 2 step: 360 train batch loss: tf.Tensor(0.59808576, shape=(), dtype=float32)\n",
            "epoch: 2 step: 361 train batch loss: tf.Tensor(0.59200287, shape=(), dtype=float32)\n",
            "epoch: 2 step: 362 train batch loss: tf.Tensor(0.5385429, shape=(), dtype=float32)\n",
            "epoch: 2 step: 363 train batch loss: tf.Tensor(0.47922596, shape=(), dtype=float32)\n",
            "epoch: 2 step: 364 train batch loss: tf.Tensor(0.55911833, shape=(), dtype=float32)\n",
            "epoch: 2 step: 365 train batch loss: tf.Tensor(0.5156545, shape=(), dtype=float32)\n",
            "epoch: 2 step: 366 train batch loss: tf.Tensor(0.6088703, shape=(), dtype=float32)\n",
            "epoch: 2 step: 367 train batch loss: tf.Tensor(0.5790897, shape=(), dtype=float32)\n",
            "epoch: 2 step: 368 train batch loss: tf.Tensor(0.5814408, shape=(), dtype=float32)\n",
            "epoch: 2 step: 369 train batch loss: tf.Tensor(0.57670414, shape=(), dtype=float32)\n",
            "epoch: 2 step: 370 train batch loss: tf.Tensor(0.5183656, shape=(), dtype=float32)\n",
            "epoch: 2 step: 371 train batch loss: tf.Tensor(0.5274734, shape=(), dtype=float32)\n",
            "epoch: 2 step: 372 train batch loss: tf.Tensor(0.57666755, shape=(), dtype=float32)\n",
            "epoch: 2 step: 373 train batch loss: tf.Tensor(0.6136699, shape=(), dtype=float32)\n",
            "epoch: 2 step: 374 train batch loss: tf.Tensor(0.5314817, shape=(), dtype=float32)\n",
            "epoch: 2 step: 375 train batch loss: tf.Tensor(0.527339, shape=(), dtype=float32)\n",
            "epoch: 2 step: 376 train batch loss: tf.Tensor(0.5963424, shape=(), dtype=float32)\n",
            "epoch: 2 step: 377 train batch loss: tf.Tensor(0.60616714, shape=(), dtype=float32)\n",
            "epoch: 2 step: 378 train batch loss: tf.Tensor(0.64619005, shape=(), dtype=float32)\n",
            "epoch: 2 step: 379 train batch loss: tf.Tensor(0.63395035, shape=(), dtype=float32)\n",
            "epoch: 2 step: 380 train batch loss: tf.Tensor(0.51128036, shape=(), dtype=float32)\n",
            "epoch: 2 step: 381 train batch loss: tf.Tensor(0.68485844, shape=(), dtype=float32)\n",
            "epoch: 2 step: 382 train batch loss: tf.Tensor(0.53059614, shape=(), dtype=float32)\n",
            "epoch: 2 step: 383 train batch loss: tf.Tensor(0.6841956, shape=(), dtype=float32)\n",
            "epoch: 2 step: 384 train batch loss: tf.Tensor(0.53106624, shape=(), dtype=float32)\n",
            "epoch: 2 step: 385 train batch loss: tf.Tensor(0.5476881, shape=(), dtype=float32)\n",
            "epoch: 2 step: 386 train batch loss: tf.Tensor(0.60660464, shape=(), dtype=float32)\n",
            "epoch: 2 step: 387 train batch loss: tf.Tensor(0.4736578, shape=(), dtype=float32)\n",
            "epoch: 2 step: 388 train batch loss: tf.Tensor(0.5250298, shape=(), dtype=float32)\n",
            "epoch: 2 step: 389 train batch loss: tf.Tensor(0.51607454, shape=(), dtype=float32)\n",
            "epoch: 2 step: 390 train batch loss: tf.Tensor(0.692133, shape=(), dtype=float32)\n",
            "epoch: 2 step: 391 train batch loss: tf.Tensor(0.67802095, shape=(), dtype=float32)\n",
            "epoch: 2 step: 392 train batch loss: tf.Tensor(0.6156777, shape=(), dtype=float32)\n",
            "epoch: 2 step: 393 train batch loss: tf.Tensor(0.62496394, shape=(), dtype=float32)\n",
            "epoch: 2 step: 394 train batch loss: tf.Tensor(0.6806544, shape=(), dtype=float32)\n",
            "epoch: 2 step: 395 train batch loss: tf.Tensor(0.6162085, shape=(), dtype=float32)\n",
            "epoch: 2 step: 396 train batch loss: tf.Tensor(0.5980147, shape=(), dtype=float32)\n",
            "epoch: 2 step: 397 train batch loss: tf.Tensor(0.6447294, shape=(), dtype=float32)\n",
            "epoch: 2 step: 398 train batch loss: tf.Tensor(0.5120771, shape=(), dtype=float32)\n",
            "epoch: 2 step: 399 train batch loss: tf.Tensor(0.71172094, shape=(), dtype=float32)\n",
            "epoch: 2 step: 400 train batch loss: tf.Tensor(0.56938547, shape=(), dtype=float32)\n",
            "epoch: 2 step: 401 train batch loss: tf.Tensor(0.48376364, shape=(), dtype=float32)\n",
            "epoch: 2 step: 402 train batch loss: tf.Tensor(0.69372725, shape=(), dtype=float32)\n",
            "epoch: 2 step: 403 train batch loss: tf.Tensor(0.5063213, shape=(), dtype=float32)\n",
            "epoch: 2 step: 404 train batch loss: tf.Tensor(0.57723486, shape=(), dtype=float32)\n",
            "epoch: 2 step: 405 train batch loss: tf.Tensor(0.5505687, shape=(), dtype=float32)\n",
            "epoch: 2 step: 406 train batch loss: tf.Tensor(0.49384028, shape=(), dtype=float32)\n",
            "epoch: 2 step: 407 train batch loss: tf.Tensor(0.46695948, shape=(), dtype=float32)\n",
            "epoch: 2 step: 408 train batch loss: tf.Tensor(0.5322367, shape=(), dtype=float32)\n",
            "epoch: 2 step: 409 train batch loss: tf.Tensor(0.61862415, shape=(), dtype=float32)\n",
            "epoch: 2 step: 410 train batch loss: tf.Tensor(0.59800017, shape=(), dtype=float32)\n",
            "epoch: 2 step: 411 train batch loss: tf.Tensor(0.5897845, shape=(), dtype=float32)\n",
            "epoch: 2 step: 412 train batch loss: tf.Tensor(0.58825815, shape=(), dtype=float32)\n",
            "epoch: 2 step: 413 train batch loss: tf.Tensor(0.52541745, shape=(), dtype=float32)\n",
            "epoch: 2 step: 414 train batch loss: tf.Tensor(0.79014444, shape=(), dtype=float32)\n",
            "epoch: 2 step: 415 train batch loss: tf.Tensor(0.55616325, shape=(), dtype=float32)\n",
            "epoch: 2 step: 416 train batch loss: tf.Tensor(0.5788258, shape=(), dtype=float32)\n",
            "epoch: 2 step: 417 train batch loss: tf.Tensor(0.609595, shape=(), dtype=float32)\n",
            "epoch: 2 step: 418 train batch loss: tf.Tensor(0.5613004, shape=(), dtype=float32)\n",
            "epoch: 2 step: 419 train batch loss: tf.Tensor(0.6067922, shape=(), dtype=float32)\n",
            "epoch: 2 step: 420 train batch loss: tf.Tensor(0.5650535, shape=(), dtype=float32)\n",
            "epoch: 2 step: 421 train batch loss: tf.Tensor(0.5234031, shape=(), dtype=float32)\n",
            "epoch: 2 step: 422 train batch loss: tf.Tensor(0.5263784, shape=(), dtype=float32)\n",
            "epoch: 2 step: 423 train batch loss: tf.Tensor(0.55089426, shape=(), dtype=float32)\n",
            "epoch: 2 step: 424 train batch loss: tf.Tensor(0.64297915, shape=(), dtype=float32)\n",
            "epoch: 2 step: 425 train batch loss: tf.Tensor(0.5715847, shape=(), dtype=float32)\n",
            "epoch: 2 step: 426 train batch loss: tf.Tensor(0.6232515, shape=(), dtype=float32)\n",
            "epoch: 2 step: 427 train batch loss: tf.Tensor(0.49168596, shape=(), dtype=float32)\n",
            "epoch: 2 step: 428 train batch loss: tf.Tensor(0.47278887, shape=(), dtype=float32)\n",
            "epoch: 2 step: 429 train batch loss: tf.Tensor(0.4654602, shape=(), dtype=float32)\n",
            "epoch: 2 step: 430 train batch loss: tf.Tensor(0.65280664, shape=(), dtype=float32)\n",
            "epoch: 2 step: 431 train batch loss: tf.Tensor(0.56125844, shape=(), dtype=float32)\n",
            "epoch: 2 step: 432 train batch loss: tf.Tensor(0.65810204, shape=(), dtype=float32)\n",
            "epoch: 2 step: 433 train batch loss: tf.Tensor(0.676792, shape=(), dtype=float32)\n",
            "epoch: 2 step: 434 train batch loss: tf.Tensor(0.5856098, shape=(), dtype=float32)\n",
            "epoch: 2 step: 435 train batch loss: tf.Tensor(0.6378962, shape=(), dtype=float32)\n",
            "epoch: 2 step: 436 train batch loss: tf.Tensor(0.5168315, shape=(), dtype=float32)\n",
            "epoch: 2 step: 437 train batch loss: tf.Tensor(0.7196231, shape=(), dtype=float32)\n",
            "epoch: 2 step: 438 train batch loss: tf.Tensor(0.5495473, shape=(), dtype=float32)\n",
            "epoch: 2 step: 439 train batch loss: tf.Tensor(0.6055578, shape=(), dtype=float32)\n",
            "epoch: 2 step: 440 train batch loss: tf.Tensor(0.5782052, shape=(), dtype=float32)\n",
            "epoch: 2 step: 441 train batch loss: tf.Tensor(0.7167108, shape=(), dtype=float32)\n",
            "epoch: 2 step: 442 train batch loss: tf.Tensor(0.6911695, shape=(), dtype=float32)\n",
            "epoch: 2 step: 443 train batch loss: tf.Tensor(0.45893335, shape=(), dtype=float32)\n",
            "epoch: 2 step: 444 train batch loss: tf.Tensor(0.575886, shape=(), dtype=float32)\n",
            "epoch: 2 step: 445 train batch loss: tf.Tensor(0.6513357, shape=(), dtype=float32)\n",
            "epoch: 2 step: 446 train batch loss: tf.Tensor(0.55204165, shape=(), dtype=float32)\n",
            "epoch: 2 step: 447 train batch loss: tf.Tensor(0.5948662, shape=(), dtype=float32)\n",
            "epoch: 2 step: 448 train batch loss: tf.Tensor(0.52406466, shape=(), dtype=float32)\n",
            "epoch: 2 step: 449 train batch loss: tf.Tensor(0.70901763, shape=(), dtype=float32)\n",
            "epoch: 2 step: 450 train batch loss: tf.Tensor(0.5420457, shape=(), dtype=float32)\n",
            "epoch: 2 step: 451 train batch loss: tf.Tensor(0.70543754, shape=(), dtype=float32)\n",
            "epoch: 2 step: 452 train batch loss: tf.Tensor(0.58075094, shape=(), dtype=float32)\n",
            "epoch: 2 step: 453 train batch loss: tf.Tensor(0.5357761, shape=(), dtype=float32)\n",
            "epoch: 2 step: 454 train batch loss: tf.Tensor(0.54851586, shape=(), dtype=float32)\n",
            "epoch: 2 step: 455 train batch loss: tf.Tensor(0.58217067, shape=(), dtype=float32)\n",
            "epoch: 2 step: 456 train batch loss: tf.Tensor(0.48934823, shape=(), dtype=float32)\n",
            "epoch: 2 step: 457 train batch loss: tf.Tensor(0.76549745, shape=(), dtype=float32)\n",
            "epoch: 2 step: 458 train batch loss: tf.Tensor(0.59049344, shape=(), dtype=float32)\n",
            "epoch: 2 step: 459 train batch loss: tf.Tensor(0.6274169, shape=(), dtype=float32)\n",
            "epoch: 2 step: 460 train batch loss: tf.Tensor(0.5088879, shape=(), dtype=float32)\n",
            "epoch: 2 step: 461 train batch loss: tf.Tensor(0.55060434, shape=(), dtype=float32)\n",
            "epoch: 2 step: 462 train batch loss: tf.Tensor(0.6100421, shape=(), dtype=float32)\n",
            "epoch: 2 step: 463 train batch loss: tf.Tensor(0.6253995, shape=(), dtype=float32)\n",
            "epoch: 2 step: 464 train batch loss: tf.Tensor(0.5771097, shape=(), dtype=float32)\n",
            "epoch: 2 step: 465 train batch loss: tf.Tensor(0.5233606, shape=(), dtype=float32)\n",
            "epoch: 2 step: 466 train batch loss: tf.Tensor(0.56462914, shape=(), dtype=float32)\n",
            "epoch: 2 step: 467 train batch loss: tf.Tensor(0.48473305, shape=(), dtype=float32)\n",
            "epoch: 2 step: 468 train batch loss: tf.Tensor(0.6250576, shape=(), dtype=float32)\n",
            "epoch: 2 step: 469 train batch loss: tf.Tensor(0.605672, shape=(), dtype=float32)\n",
            "epoch: 2 step: 470 train batch loss: tf.Tensor(0.58504725, shape=(), dtype=float32)\n",
            "epoch: 2 step: 471 train batch loss: tf.Tensor(0.5737105, shape=(), dtype=float32)\n",
            "epoch: 2 step: 472 train batch loss: tf.Tensor(0.60910183, shape=(), dtype=float32)\n",
            "epoch: 2 step: 473 train batch loss: tf.Tensor(0.5472273, shape=(), dtype=float32)\n",
            "epoch: 2 step: 474 train batch loss: tf.Tensor(0.5681189, shape=(), dtype=float32)\n",
            "epoch: 2 step: 475 train batch loss: tf.Tensor(0.5686927, shape=(), dtype=float32)\n",
            "epoch: 2 step: 476 train batch loss: tf.Tensor(0.54905474, shape=(), dtype=float32)\n",
            "epoch: 2 step: 477 train batch loss: tf.Tensor(0.5832789, shape=(), dtype=float32)\n",
            "epoch: 2 step: 478 train batch loss: tf.Tensor(0.57722044, shape=(), dtype=float32)\n",
            "epoch: 2 step: 479 train batch loss: tf.Tensor(0.584122, shape=(), dtype=float32)\n",
            "epoch: 2 step: 480 train batch loss: tf.Tensor(0.72343576, shape=(), dtype=float32)\n",
            "epoch: 2 step: 481 train batch loss: tf.Tensor(0.44331092, shape=(), dtype=float32)\n",
            "epoch: 2 step: 482 train batch loss: tf.Tensor(0.5830952, shape=(), dtype=float32)\n",
            "epoch: 2 step: 483 train batch loss: tf.Tensor(0.6344333, shape=(), dtype=float32)\n",
            "epoch: 2 step: 484 train batch loss: tf.Tensor(0.48157382, shape=(), dtype=float32)\n",
            "epoch: 2 step: 485 train batch loss: tf.Tensor(0.6555152, shape=(), dtype=float32)\n",
            "epoch: 2 step: 486 train batch loss: tf.Tensor(0.5415665, shape=(), dtype=float32)\n",
            "epoch: 2 step: 487 train batch loss: tf.Tensor(0.5276049, shape=(), dtype=float32)\n",
            "epoch: 2 step: 488 train batch loss: tf.Tensor(0.56652784, shape=(), dtype=float32)\n",
            "epoch: 2 step: 489 train batch loss: tf.Tensor(0.57266146, shape=(), dtype=float32)\n",
            "epoch: 2 step: 490 train batch loss: tf.Tensor(0.519412, shape=(), dtype=float32)\n",
            "epoch: 2 step: 491 train batch loss: tf.Tensor(0.5772165, shape=(), dtype=float32)\n",
            "epoch: 2 step: 492 train batch loss: tf.Tensor(0.7595317, shape=(), dtype=float32)\n",
            "epoch: 2 step: 493 train batch loss: tf.Tensor(0.5283555, shape=(), dtype=float32)\n",
            "epoch: 2 step: 494 train batch loss: tf.Tensor(0.53552943, shape=(), dtype=float32)\n",
            "epoch: 2 step: 495 train batch loss: tf.Tensor(0.42103752, shape=(), dtype=float32)\n",
            "epoch: 2 step: 496 train batch loss: tf.Tensor(0.6290377, shape=(), dtype=float32)\n",
            "epoch: 2 step: 497 train batch loss: tf.Tensor(0.60987365, shape=(), dtype=float32)\n",
            "epoch: 2 step: 498 train batch loss: tf.Tensor(0.6289882, shape=(), dtype=float32)\n",
            "epoch: 2 step: 499 train batch loss: tf.Tensor(0.53374696, shape=(), dtype=float32)\n",
            "epoch: 2 step: 500 train batch loss: tf.Tensor(0.6534988, shape=(), dtype=float32)\n",
            "epoch: 2 step: 501 train batch loss: tf.Tensor(0.49904314, shape=(), dtype=float32)\n",
            "epoch: 2 step: 502 train batch loss: tf.Tensor(0.63187397, shape=(), dtype=float32)\n",
            "epoch: 2 step: 503 train batch loss: tf.Tensor(0.5006894, shape=(), dtype=float32)\n",
            "epoch: 2 step: 504 train batch loss: tf.Tensor(0.55819154, shape=(), dtype=float32)\n",
            "epoch: 2 step: 505 train batch loss: tf.Tensor(0.54892796, shape=(), dtype=float32)\n",
            "epoch: 2 step: 506 train batch loss: tf.Tensor(0.5787121, shape=(), dtype=float32)\n",
            "epoch: 2 step: 507 train batch loss: tf.Tensor(0.41177496, shape=(), dtype=float32)\n",
            "epoch: 2 step: 508 train batch loss: tf.Tensor(0.63465255, shape=(), dtype=float32)\n",
            "epoch: 2 step: 509 train batch loss: tf.Tensor(0.606183, shape=(), dtype=float32)\n",
            "epoch: 2 step: 510 train batch loss: tf.Tensor(0.5252292, shape=(), dtype=float32)\n",
            "epoch: 2 step: 511 train batch loss: tf.Tensor(0.57814276, shape=(), dtype=float32)\n",
            "epoch: 2 step: 512 train batch loss: tf.Tensor(0.5743335, shape=(), dtype=float32)\n",
            "epoch: 2 step: 513 train batch loss: tf.Tensor(0.6727526, shape=(), dtype=float32)\n",
            "epoch: 2 step: 514 train batch loss: tf.Tensor(0.6674281, shape=(), dtype=float32)\n",
            "epoch: 2 step: 515 train batch loss: tf.Tensor(0.5873731, shape=(), dtype=float32)\n",
            "epoch: 2 step: 516 train batch loss: tf.Tensor(0.42641672, shape=(), dtype=float32)\n",
            "epoch: 2 step: 517 train batch loss: tf.Tensor(0.4982249, shape=(), dtype=float32)\n",
            "epoch: 2 step: 518 train batch loss: tf.Tensor(0.6158563, shape=(), dtype=float32)\n",
            "epoch: 2 step: 519 train batch loss: tf.Tensor(0.5020697, shape=(), dtype=float32)\n",
            "epoch: 2 step: 520 train batch loss: tf.Tensor(0.57820064, shape=(), dtype=float32)\n",
            "epoch: 2 step: 521 train batch loss: tf.Tensor(0.56299883, shape=(), dtype=float32)\n",
            "epoch: 2 step: 522 train batch loss: tf.Tensor(0.6201505, shape=(), dtype=float32)\n",
            "epoch: 2 step: 523 train batch loss: tf.Tensor(0.50887644, shape=(), dtype=float32)\n",
            "epoch: 2 step: 524 train batch loss: tf.Tensor(0.72902393, shape=(), dtype=float32)\n",
            "epoch: 2 step: 525 train batch loss: tf.Tensor(0.49739885, shape=(), dtype=float32)\n",
            "epoch: 2 step: 526 train batch loss: tf.Tensor(0.6015947, shape=(), dtype=float32)\n",
            "epoch: 2 step: 527 train batch loss: tf.Tensor(0.51148933, shape=(), dtype=float32)\n",
            "epoch: 2 step: 528 train batch loss: tf.Tensor(0.48174453, shape=(), dtype=float32)\n",
            "epoch: 2 step: 529 train batch loss: tf.Tensor(0.5525204, shape=(), dtype=float32)\n",
            "epoch: 2 step: 530 train batch loss: tf.Tensor(0.56459635, shape=(), dtype=float32)\n",
            "epoch: 2 step: 531 train batch loss: tf.Tensor(0.57641876, shape=(), dtype=float32)\n",
            "epoch: 2 step: 532 train batch loss: tf.Tensor(0.5535684, shape=(), dtype=float32)\n",
            "epoch: 2 step: 533 train batch loss: tf.Tensor(0.668179, shape=(), dtype=float32)\n",
            "epoch: 2 step: 534 train batch loss: tf.Tensor(0.5266683, shape=(), dtype=float32)\n",
            "epoch: 2 step: 535 train batch loss: tf.Tensor(0.4333845, shape=(), dtype=float32)\n",
            "epoch: 2 step: 536 train batch loss: tf.Tensor(0.49088773, shape=(), dtype=float32)\n",
            "epoch: 2 step: 537 train batch loss: tf.Tensor(0.58832496, shape=(), dtype=float32)\n",
            "epoch: 2 step: 538 train batch loss: tf.Tensor(0.4816184, shape=(), dtype=float32)\n",
            "epoch: 2 step: 539 train batch loss: tf.Tensor(0.6283678, shape=(), dtype=float32)\n",
            "epoch: 2 step: 540 train batch loss: tf.Tensor(0.6367093, shape=(), dtype=float32)\n",
            "epoch: 2 step: 541 train batch loss: tf.Tensor(0.5247806, shape=(), dtype=float32)\n",
            "epoch: 2 step: 542 train batch loss: tf.Tensor(0.6104835, shape=(), dtype=float32)\n",
            "epoch: 2 step: 543 train batch loss: tf.Tensor(0.7696917, shape=(), dtype=float32)\n",
            "epoch: 2 step: 544 train batch loss: tf.Tensor(0.60460585, shape=(), dtype=float32)\n",
            "epoch: 2 step: 545 train batch loss: tf.Tensor(0.5675453, shape=(), dtype=float32)\n",
            "epoch: 2 step: 546 train batch loss: tf.Tensor(0.4990603, shape=(), dtype=float32)\n",
            "epoch: 2 step: 547 train batch loss: tf.Tensor(0.49250287, shape=(), dtype=float32)\n",
            "epoch: 2 step: 548 train batch loss: tf.Tensor(0.6249598, shape=(), dtype=float32)\n",
            "epoch: 2 step: 549 train batch loss: tf.Tensor(0.5593436, shape=(), dtype=float32)\n",
            "epoch: 2 step: 550 train batch loss: tf.Tensor(0.52369046, shape=(), dtype=float32)\n",
            "epoch: 2 step: 551 train batch loss: tf.Tensor(0.60263693, shape=(), dtype=float32)\n",
            "epoch: 2 step: 552 train batch loss: tf.Tensor(0.55193645, shape=(), dtype=float32)\n",
            "epoch: 2 step: 553 train batch loss: tf.Tensor(0.61755157, shape=(), dtype=float32)\n",
            "epoch: 2 step: 554 train batch loss: tf.Tensor(0.71535444, shape=(), dtype=float32)\n",
            "epoch: 2 step: 555 train batch loss: tf.Tensor(0.51404047, shape=(), dtype=float32)\n",
            "epoch: 2 step: 556 train batch loss: tf.Tensor(0.6270961, shape=(), dtype=float32)\n",
            "epoch: 2 step: 557 train batch loss: tf.Tensor(0.59755355, shape=(), dtype=float32)\n",
            "epoch: 2 step: 558 train batch loss: tf.Tensor(0.5077325, shape=(), dtype=float32)\n",
            "epoch: 2 step: 559 train batch loss: tf.Tensor(0.50792927, shape=(), dtype=float32)\n",
            "epoch: 2 step: 560 train batch loss: tf.Tensor(0.46270895, shape=(), dtype=float32)\n",
            "epoch: 2 step: 561 train batch loss: tf.Tensor(0.53768355, shape=(), dtype=float32)\n",
            "epoch: 2 step: 562 train batch loss: tf.Tensor(0.56312054, shape=(), dtype=float32)\n",
            "epoch: 2 step: 563 train batch loss: tf.Tensor(0.56180316, shape=(), dtype=float32)\n",
            "epoch: 2 step: 564 train batch loss: tf.Tensor(0.542982, shape=(), dtype=float32)\n",
            "epoch: 2 step: 565 train batch loss: tf.Tensor(0.68828404, shape=(), dtype=float32)\n",
            "epoch: 2 step: 566 train batch loss: tf.Tensor(0.5328766, shape=(), dtype=float32)\n",
            "epoch: 2 step: 567 train batch loss: tf.Tensor(0.46295142, shape=(), dtype=float32)\n",
            "epoch: 2 step: 568 train batch loss: tf.Tensor(0.5307382, shape=(), dtype=float32)\n",
            "epoch: 2 step: 569 train batch loss: tf.Tensor(0.55863744, shape=(), dtype=float32)\n",
            "epoch: 2 step: 570 train batch loss: tf.Tensor(0.55364686, shape=(), dtype=float32)\n",
            "epoch: 2 step: 571 train batch loss: tf.Tensor(0.551666, shape=(), dtype=float32)\n",
            "epoch: 2 step: 572 train batch loss: tf.Tensor(0.62287885, shape=(), dtype=float32)\n",
            "epoch: 2 step: 573 train batch loss: tf.Tensor(0.5346278, shape=(), dtype=float32)\n",
            "epoch: 2 step: 574 train batch loss: tf.Tensor(0.5486148, shape=(), dtype=float32)\n",
            "epoch: 2 step: 575 train batch loss: tf.Tensor(0.51974297, shape=(), dtype=float32)\n",
            "epoch: 2 step: 576 train batch loss: tf.Tensor(0.59326017, shape=(), dtype=float32)\n",
            "epoch: 2 step: 577 train batch loss: tf.Tensor(0.53470683, shape=(), dtype=float32)\n",
            "epoch: 2 step: 578 train batch loss: tf.Tensor(0.6079193, shape=(), dtype=float32)\n",
            "epoch: 2 step: 579 train batch loss: tf.Tensor(0.74270386, shape=(), dtype=float32)\n",
            "epoch: 2 step: 580 train batch loss: tf.Tensor(0.55480534, shape=(), dtype=float32)\n",
            "epoch: 2 step: 581 train batch loss: tf.Tensor(0.53815734, shape=(), dtype=float32)\n",
            "epoch: 2 step: 582 train batch loss: tf.Tensor(0.5940788, shape=(), dtype=float32)\n",
            "epoch: 2 step: 583 train batch loss: tf.Tensor(0.5128397, shape=(), dtype=float32)\n",
            "epoch: 2 step: 584 train batch loss: tf.Tensor(0.5074445, shape=(), dtype=float32)\n",
            "epoch: 2 step: 585 train batch loss: tf.Tensor(0.5983526, shape=(), dtype=float32)\n",
            "epoch: 2 step: 586 train batch loss: tf.Tensor(0.48703808, shape=(), dtype=float32)\n",
            "epoch: 2 step: 587 train batch loss: tf.Tensor(0.52557474, shape=(), dtype=float32)\n",
            "epoch: 2 step: 588 train batch loss: tf.Tensor(0.46277422, shape=(), dtype=float32)\n",
            "epoch: 2 step: 589 train batch loss: tf.Tensor(0.54200375, shape=(), dtype=float32)\n",
            "epoch: 2 step: 590 train batch loss: tf.Tensor(0.54304194, shape=(), dtype=float32)\n",
            "epoch: 2 step: 591 train batch loss: tf.Tensor(0.53554827, shape=(), dtype=float32)\n",
            "epoch: 2 step: 592 train batch loss: tf.Tensor(0.6755027, shape=(), dtype=float32)\n",
            "epoch: 2 step: 593 train batch loss: tf.Tensor(0.50431985, shape=(), dtype=float32)\n",
            "epoch: 2 step: 594 train batch loss: tf.Tensor(0.49643818, shape=(), dtype=float32)\n",
            "epoch: 2 step: 595 train batch loss: tf.Tensor(0.5663544, shape=(), dtype=float32)\n",
            "epoch: 2 step: 596 train batch loss: tf.Tensor(0.5491944, shape=(), dtype=float32)\n",
            "epoch: 2 step: 597 train batch loss: tf.Tensor(0.48604423, shape=(), dtype=float32)\n",
            "epoch: 2 step: 598 train batch loss: tf.Tensor(0.58199066, shape=(), dtype=float32)\n",
            "epoch: 2 step: 599 train batch loss: tf.Tensor(0.44513822, shape=(), dtype=float32)\n",
            "epoch: 2 step: 600 train batch loss: tf.Tensor(0.6370051, shape=(), dtype=float32)\n",
            "epoch: 2 step: 601 train batch loss: tf.Tensor(0.61267275, shape=(), dtype=float32)\n",
            "epoch: 2 step: 602 train batch loss: tf.Tensor(0.6834773, shape=(), dtype=float32)\n",
            "epoch: 2 step: 603 train batch loss: tf.Tensor(0.6389555, shape=(), dtype=float32)\n",
            "epoch: 2 step: 604 train batch loss: tf.Tensor(0.5405228, shape=(), dtype=float32)\n",
            "epoch: 2 step: 605 train batch loss: tf.Tensor(0.6468753, shape=(), dtype=float32)\n",
            "epoch: 2 step: 606 train batch loss: tf.Tensor(0.48467216, shape=(), dtype=float32)\n",
            "epoch: 2 step: 607 train batch loss: tf.Tensor(0.65423644, shape=(), dtype=float32)\n",
            "epoch: 2 step: 608 train batch loss: tf.Tensor(0.5612256, shape=(), dtype=float32)\n",
            "epoch: 2 step: 609 train batch loss: tf.Tensor(0.5530616, shape=(), dtype=float32)\n",
            "epoch: 2 step: 610 train batch loss: tf.Tensor(0.6189423, shape=(), dtype=float32)\n",
            "epoch: 2 step: 611 train batch loss: tf.Tensor(0.5794802, shape=(), dtype=float32)\n",
            "epoch: 2 step: 612 train batch loss: tf.Tensor(0.6691253, shape=(), dtype=float32)\n",
            "epoch: 2 step: 613 train batch loss: tf.Tensor(0.50180733, shape=(), dtype=float32)\n",
            "epoch: 2 step: 614 train batch loss: tf.Tensor(0.5405807, shape=(), dtype=float32)\n",
            "epoch: 2 step: 615 train batch loss: tf.Tensor(0.5469932, shape=(), dtype=float32)\n",
            "epoch: 2 step: 616 train batch loss: tf.Tensor(0.6216031, shape=(), dtype=float32)\n",
            "epoch: 2 step: 617 train batch loss: tf.Tensor(0.48758778, shape=(), dtype=float32)\n",
            "epoch: 2 step: 618 train batch loss: tf.Tensor(0.521756, shape=(), dtype=float32)\n",
            "epoch: 2 step: 619 train batch loss: tf.Tensor(0.68646955, shape=(), dtype=float32)\n",
            "epoch: 2 step: 620 train batch loss: tf.Tensor(0.5465102, shape=(), dtype=float32)\n",
            "epoch: 2 step: 621 train batch loss: tf.Tensor(0.49869174, shape=(), dtype=float32)\n",
            "epoch: 2 step: 622 train batch loss: tf.Tensor(0.58314925, shape=(), dtype=float32)\n",
            "epoch: 2 step: 623 train batch loss: tf.Tensor(0.5521686, shape=(), dtype=float32)\n",
            "epoch: 2 step: 624 train batch loss: tf.Tensor(0.50775015, shape=(), dtype=float32)\n",
            "epoch: 2 step: 625 train batch loss: tf.Tensor(0.52529144, shape=(), dtype=float32)\n",
            "epoch: 2 step: 626 train batch loss: tf.Tensor(0.439612, shape=(), dtype=float32)\n",
            "epoch: 2 step: 627 train batch loss: tf.Tensor(0.62683976, shape=(), dtype=float32)\n",
            "epoch: 2 step: 628 train batch loss: tf.Tensor(0.5740643, shape=(), dtype=float32)\n",
            "epoch: 2 step: 629 train batch loss: tf.Tensor(0.5911065, shape=(), dtype=float32)\n",
            "epoch: 2 step: 630 train batch loss: tf.Tensor(0.50501305, shape=(), dtype=float32)\n",
            "epoch: 2 step: 631 train batch loss: tf.Tensor(0.53686064, shape=(), dtype=float32)\n",
            "epoch: 2 step: 632 train batch loss: tf.Tensor(0.4538099, shape=(), dtype=float32)\n",
            "epoch: 2 step: 633 train batch loss: tf.Tensor(0.508322, shape=(), dtype=float32)\n",
            "epoch: 2 step: 634 train batch loss: tf.Tensor(0.5123566, shape=(), dtype=float32)\n",
            "epoch: 2 step: 635 train batch loss: tf.Tensor(0.5955897, shape=(), dtype=float32)\n",
            "epoch: 2 step: 636 train batch loss: tf.Tensor(0.55607945, shape=(), dtype=float32)\n",
            "epoch: 2 step: 637 train batch loss: tf.Tensor(0.4812669, shape=(), dtype=float32)\n",
            "epoch: 2 step: 638 train batch loss: tf.Tensor(0.62528837, shape=(), dtype=float32)\n",
            "epoch: 2 step: 639 train batch loss: tf.Tensor(0.56631523, shape=(), dtype=float32)\n",
            "epoch: 2 step: 640 train batch loss: tf.Tensor(0.66109025, shape=(), dtype=float32)\n",
            "epoch: 2 step: 641 train batch loss: tf.Tensor(0.80062246, shape=(), dtype=float32)\n",
            "epoch: 2 step: 642 train batch loss: tf.Tensor(0.71817076, shape=(), dtype=float32)\n",
            "epoch: 2 step: 643 train batch loss: tf.Tensor(0.60675466, shape=(), dtype=float32)\n",
            "epoch: 2 step: 644 train batch loss: tf.Tensor(0.5720849, shape=(), dtype=float32)\n",
            "epoch: 2 step: 645 train batch loss: tf.Tensor(0.5579109, shape=(), dtype=float32)\n",
            "epoch: 2 step: 646 train batch loss: tf.Tensor(0.539791, shape=(), dtype=float32)\n",
            "epoch: 2 step: 647 train batch loss: tf.Tensor(0.558771, shape=(), dtype=float32)\n",
            "epoch: 2 step: 648 train batch loss: tf.Tensor(0.5329968, shape=(), dtype=float32)\n",
            "epoch: 2 step: 649 train batch loss: tf.Tensor(0.48806638, shape=(), dtype=float32)\n",
            "epoch: 2 step: 650 train batch loss: tf.Tensor(0.6275932, shape=(), dtype=float32)\n",
            "epoch: 2 step: 651 train batch loss: tf.Tensor(0.6639704, shape=(), dtype=float32)\n",
            "epoch: 2 step: 652 train batch loss: tf.Tensor(0.54240113, shape=(), dtype=float32)\n",
            "epoch: 2 step: 653 train batch loss: tf.Tensor(0.6390626, shape=(), dtype=float32)\n",
            "epoch: 2 step: 654 train batch loss: tf.Tensor(0.42228127, shape=(), dtype=float32)\n",
            "epoch: 2 step: 655 train batch loss: tf.Tensor(0.65739536, shape=(), dtype=float32)\n",
            "epoch: 2 step: 656 train batch loss: tf.Tensor(0.6157855, shape=(), dtype=float32)\n",
            "epoch: 2 step: 657 train batch loss: tf.Tensor(0.47401404, shape=(), dtype=float32)\n",
            "epoch: 2 step: 658 train batch loss: tf.Tensor(0.57818204, shape=(), dtype=float32)\n",
            "epoch: 2 step: 659 train batch loss: tf.Tensor(0.5544615, shape=(), dtype=float32)\n",
            "epoch: 2 step: 660 train batch loss: tf.Tensor(0.67313874, shape=(), dtype=float32)\n",
            "epoch: 2 step: 661 train batch loss: tf.Tensor(0.5259812, shape=(), dtype=float32)\n",
            "epoch: 2 step: 662 train batch loss: tf.Tensor(0.5564256, shape=(), dtype=float32)\n",
            "epoch: 2 step: 663 train batch loss: tf.Tensor(0.520563, shape=(), dtype=float32)\n",
            "epoch: 2 step: 664 train batch loss: tf.Tensor(0.546619, shape=(), dtype=float32)\n",
            "epoch: 2 step: 665 train batch loss: tf.Tensor(0.5706998, shape=(), dtype=float32)\n",
            "epoch: 2 step: 666 train batch loss: tf.Tensor(0.49009308, shape=(), dtype=float32)\n",
            "epoch: 2 step: 667 train batch loss: tf.Tensor(0.5409381, shape=(), dtype=float32)\n",
            "epoch: 2 step: 668 train batch loss: tf.Tensor(0.5901645, shape=(), dtype=float32)\n",
            "epoch: 2 step: 669 train batch loss: tf.Tensor(0.50152713, shape=(), dtype=float32)\n",
            "epoch: 2 step: 670 train batch loss: tf.Tensor(0.58855623, shape=(), dtype=float32)\n",
            "epoch: 2 step: 671 train batch loss: tf.Tensor(0.5602986, shape=(), dtype=float32)\n",
            "epoch: 2 step: 672 train batch loss: tf.Tensor(0.74389946, shape=(), dtype=float32)\n",
            "epoch: 2 step: 673 train batch loss: tf.Tensor(0.49022952, shape=(), dtype=float32)\n",
            "epoch: 2 step: 674 train batch loss: tf.Tensor(0.5320363, shape=(), dtype=float32)\n",
            "epoch: 2 step: 675 train batch loss: tf.Tensor(0.6236509, shape=(), dtype=float32)\n",
            "epoch: 2 step: 676 train batch loss: tf.Tensor(0.47148713, shape=(), dtype=float32)\n",
            "epoch: 2 step: 677 train batch loss: tf.Tensor(0.6601304, shape=(), dtype=float32)\n",
            "epoch: 2 step: 678 train batch loss: tf.Tensor(0.51611096, shape=(), dtype=float32)\n",
            "epoch: 2 step: 679 train batch loss: tf.Tensor(0.5724374, shape=(), dtype=float32)\n",
            "epoch: 2 step: 680 train batch loss: tf.Tensor(0.55206275, shape=(), dtype=float32)\n",
            "epoch: 2 step: 681 train batch loss: tf.Tensor(0.6109427, shape=(), dtype=float32)\n",
            "epoch: 2 step: 682 train batch loss: tf.Tensor(0.5394066, shape=(), dtype=float32)\n",
            "epoch: 2 step: 683 train batch loss: tf.Tensor(0.54112947, shape=(), dtype=float32)\n",
            "epoch: 2 step: 684 train batch loss: tf.Tensor(0.5669883, shape=(), dtype=float32)\n",
            "epoch: 2 step: 685 train batch loss: tf.Tensor(0.559907, shape=(), dtype=float32)\n",
            "epoch: 2 step: 686 train batch loss: tf.Tensor(0.511689, shape=(), dtype=float32)\n",
            "epoch: 2 step: 687 train batch loss: tf.Tensor(0.51130635, shape=(), dtype=float32)\n",
            "epoch: 2 step: 688 train batch loss: tf.Tensor(0.538462, shape=(), dtype=float32)\n",
            "epoch: 2 step: 689 train batch loss: tf.Tensor(0.5567835, shape=(), dtype=float32)\n",
            "epoch: 2 step: 690 train batch loss: tf.Tensor(0.563722, shape=(), dtype=float32)\n",
            "epoch: 2 step: 691 train batch loss: tf.Tensor(0.40293974, shape=(), dtype=float32)\n",
            "epoch: 2 step: 692 train batch loss: tf.Tensor(0.5252488, shape=(), dtype=float32)\n",
            "epoch: 2 step: 693 train batch loss: tf.Tensor(0.4025136, shape=(), dtype=float32)\n",
            "epoch: 2 step: 694 train batch loss: tf.Tensor(0.43332586, shape=(), dtype=float32)\n",
            "epoch: 2 step: 695 train batch loss: tf.Tensor(0.576188, shape=(), dtype=float32)\n",
            "epoch: 2 step: 696 train batch loss: tf.Tensor(0.5251077, shape=(), dtype=float32)\n",
            "epoch: 2 step: 697 train batch loss: tf.Tensor(0.5870398, shape=(), dtype=float32)\n",
            "epoch: 2 step: 698 train batch loss: tf.Tensor(0.54163617, shape=(), dtype=float32)\n",
            "epoch: 2 step: 699 train batch loss: tf.Tensor(0.58214605, shape=(), dtype=float32)\n",
            "epoch: 2 step: 700 train batch loss: tf.Tensor(0.48442286, shape=(), dtype=float32)\n",
            "epoch: 2 step: 701 train batch loss: tf.Tensor(0.54546976, shape=(), dtype=float32)\n",
            "epoch: 2 step: 702 train batch loss: tf.Tensor(0.5628854, shape=(), dtype=float32)\n",
            "epoch: 2 step: 703 train batch loss: tf.Tensor(0.47228438, shape=(), dtype=float32)\n",
            "epoch: 2 step: 704 train batch loss: tf.Tensor(0.59009796, shape=(), dtype=float32)\n",
            "epoch: 2 step: 705 train batch loss: tf.Tensor(0.55338264, shape=(), dtype=float32)\n",
            "epoch: 2 step: 706 train batch loss: tf.Tensor(0.5022732, shape=(), dtype=float32)\n",
            "epoch: 2 step: 707 train batch loss: tf.Tensor(0.46240845, shape=(), dtype=float32)\n",
            "epoch: 2 step: 708 train batch loss: tf.Tensor(0.47003216, shape=(), dtype=float32)\n",
            "epoch: 2 step: 709 train batch loss: tf.Tensor(0.47195283, shape=(), dtype=float32)\n",
            "epoch: 2 step: 710 train batch loss: tf.Tensor(0.60870624, shape=(), dtype=float32)\n",
            "epoch: 2 step: 711 train batch loss: tf.Tensor(0.4777786, shape=(), dtype=float32)\n",
            "epoch: 2 step: 712 train batch loss: tf.Tensor(0.6108173, shape=(), dtype=float32)\n",
            "epoch: 2 step: 713 train batch loss: tf.Tensor(0.6039924, shape=(), dtype=float32)\n",
            "epoch: 2 step: 714 train batch loss: tf.Tensor(0.5847824, shape=(), dtype=float32)\n",
            "epoch: 2 step: 715 train batch loss: tf.Tensor(0.5414275, shape=(), dtype=float32)\n",
            "epoch: 2 step: 716 train batch loss: tf.Tensor(0.5510077, shape=(), dtype=float32)\n",
            "epoch: 2 step: 717 train batch loss: tf.Tensor(0.5467957, shape=(), dtype=float32)\n",
            "epoch: 2 step: 718 train batch loss: tf.Tensor(0.5875879, shape=(), dtype=float32)\n",
            "epoch: 2 step: 719 train batch loss: tf.Tensor(0.65635407, shape=(), dtype=float32)\n",
            "epoch: 2 step: 720 train batch loss: tf.Tensor(0.61122644, shape=(), dtype=float32)\n",
            "epoch: 2 step: 721 train batch loss: tf.Tensor(0.4668986, shape=(), dtype=float32)\n",
            "epoch: 2 step: 722 train batch loss: tf.Tensor(0.6905173, shape=(), dtype=float32)\n",
            "epoch: 2 step: 723 train batch loss: tf.Tensor(0.5555633, shape=(), dtype=float32)\n",
            "epoch: 2 step: 724 train batch loss: tf.Tensor(0.5113901, shape=(), dtype=float32)\n",
            "epoch: 2 step: 725 train batch loss: tf.Tensor(0.51522213, shape=(), dtype=float32)\n",
            "epoch: 2 step: 726 train batch loss: tf.Tensor(0.47581768, shape=(), dtype=float32)\n",
            "epoch: 2 step: 727 train batch loss: tf.Tensor(0.5734259, shape=(), dtype=float32)\n",
            "epoch: 2 step: 728 train batch loss: tf.Tensor(0.60878193, shape=(), dtype=float32)\n",
            "epoch: 2 step: 729 train batch loss: tf.Tensor(0.5741481, shape=(), dtype=float32)\n",
            "epoch: 2 step: 730 train batch loss: tf.Tensor(0.63801825, shape=(), dtype=float32)\n",
            "epoch: 2 step: 731 train batch loss: tf.Tensor(0.46120182, shape=(), dtype=float32)\n",
            "epoch: 2 step: 732 train batch loss: tf.Tensor(0.5899576, shape=(), dtype=float32)\n",
            "epoch: 2 step: 733 train batch loss: tf.Tensor(0.7044487, shape=(), dtype=float32)\n",
            "epoch: 2 step: 734 train batch loss: tf.Tensor(0.64246523, shape=(), dtype=float32)\n",
            "epoch: 2 step: 735 train batch loss: tf.Tensor(0.55709493, shape=(), dtype=float32)\n",
            "epoch: 2 step: 736 train batch loss: tf.Tensor(0.6072259, shape=(), dtype=float32)\n",
            "epoch: 2 step: 737 train batch loss: tf.Tensor(0.5027924, shape=(), dtype=float32)\n",
            "epoch: 2 step: 738 train batch loss: tf.Tensor(0.60877514, shape=(), dtype=float32)\n",
            "epoch: 2 step: 739 train batch loss: tf.Tensor(0.42589092, shape=(), dtype=float32)\n",
            "epoch: 2 step: 740 train batch loss: tf.Tensor(0.586162, shape=(), dtype=float32)\n",
            "epoch: 2 step: 741 train batch loss: tf.Tensor(0.56627494, shape=(), dtype=float32)\n",
            "epoch: 2 step: 742 train batch loss: tf.Tensor(0.5994281, shape=(), dtype=float32)\n",
            "epoch: 2 step: 743 train batch loss: tf.Tensor(0.51751363, shape=(), dtype=float32)\n",
            "epoch: 2 step: 744 train batch loss: tf.Tensor(0.50850195, shape=(), dtype=float32)\n",
            "epoch: 2 step: 745 train batch loss: tf.Tensor(0.4500106, shape=(), dtype=float32)\n",
            "epoch: 2 step: 746 train batch loss: tf.Tensor(0.6189898, shape=(), dtype=float32)\n",
            "epoch: 2 step: 747 train batch loss: tf.Tensor(0.5998186, shape=(), dtype=float32)\n",
            "epoch: 2 step: 748 train batch loss: tf.Tensor(0.5252572, shape=(), dtype=float32)\n",
            "epoch: 2 step: 749 train batch loss: tf.Tensor(0.543187, shape=(), dtype=float32)\n",
            "epoch: 2 step: 750 train batch loss: tf.Tensor(0.42076397, shape=(), dtype=float32)\n",
            "epoch: 2 step: 751 train batch loss: tf.Tensor(0.5437356, shape=(), dtype=float32)\n",
            "epoch: 2 step: 752 train batch loss: tf.Tensor(0.60384166, shape=(), dtype=float32)\n",
            "epoch: 2 step: 753 train batch loss: tf.Tensor(0.5214151, shape=(), dtype=float32)\n",
            "epoch: 2 step: 754 train batch loss: tf.Tensor(0.56539154, shape=(), dtype=float32)\n",
            "epoch: 2 step: 755 train batch loss: tf.Tensor(0.5386537, shape=(), dtype=float32)\n",
            "epoch: 2 step: 756 train batch loss: tf.Tensor(0.47890487, shape=(), dtype=float32)\n",
            "epoch: 2 step: 757 train batch loss: tf.Tensor(0.62764317, shape=(), dtype=float32)\n",
            "epoch: 2 step: 758 train batch loss: tf.Tensor(0.6003756, shape=(), dtype=float32)\n",
            "epoch: 2 step: 759 train batch loss: tf.Tensor(0.6248578, shape=(), dtype=float32)\n",
            "epoch: 2 step: 760 train batch loss: tf.Tensor(0.48220462, shape=(), dtype=float32)\n",
            "epoch: 2 step: 761 train batch loss: tf.Tensor(0.57938874, shape=(), dtype=float32)\n",
            "epoch: 2 step: 762 train batch loss: tf.Tensor(0.63184786, shape=(), dtype=float32)\n",
            "epoch: 2 step: 763 train batch loss: tf.Tensor(0.58802265, shape=(), dtype=float32)\n",
            "epoch: 2 step: 764 train batch loss: tf.Tensor(0.62223613, shape=(), dtype=float32)\n",
            "epoch: 2 step: 765 train batch loss: tf.Tensor(0.57769513, shape=(), dtype=float32)\n",
            "epoch: 2 step: 766 train batch loss: tf.Tensor(0.5586637, shape=(), dtype=float32)\n",
            "epoch: 2 step: 767 train batch loss: tf.Tensor(0.4953391, shape=(), dtype=float32)\n",
            "epoch: 2 step: 768 train batch loss: tf.Tensor(0.5295514, shape=(), dtype=float32)\n",
            "epoch: 2 step: 769 train batch loss: tf.Tensor(0.6062926, shape=(), dtype=float32)\n",
            "epoch: 2 step: 770 train batch loss: tf.Tensor(0.4316774, shape=(), dtype=float32)\n",
            "epoch: 2 step: 771 train batch loss: tf.Tensor(0.5614561, shape=(), dtype=float32)\n",
            "epoch: 2 step: 772 train batch loss: tf.Tensor(0.5905472, shape=(), dtype=float32)\n",
            "epoch: 2 step: 773 train batch loss: tf.Tensor(0.45150536, shape=(), dtype=float32)\n",
            "epoch: 2 step: 774 train batch loss: tf.Tensor(0.47499323, shape=(), dtype=float32)\n",
            "epoch: 2 step: 775 train batch loss: tf.Tensor(0.5212367, shape=(), dtype=float32)\n",
            "epoch: 2 step: 776 train batch loss: tf.Tensor(0.57957166, shape=(), dtype=float32)\n",
            "epoch: 2 step: 777 train batch loss: tf.Tensor(0.5975961, shape=(), dtype=float32)\n",
            "epoch: 2 step: 778 train batch loss: tf.Tensor(0.65077436, shape=(), dtype=float32)\n",
            "epoch: 2 step: 779 train batch loss: tf.Tensor(0.50909585, shape=(), dtype=float32)\n",
            "epoch: 2 step: 780 train batch loss: tf.Tensor(0.6731933, shape=(), dtype=float32)\n",
            "epoch: 2 step: 781 train batch loss: tf.Tensor(0.57939756, shape=(), dtype=float32)\n",
            "epoch: 2 step: 782 train batch loss: tf.Tensor(0.63589925, shape=(), dtype=float32)\n",
            "epoch: 2 step: 783 train batch loss: tf.Tensor(0.5376521, shape=(), dtype=float32)\n",
            "epoch: 2 step: 784 train batch loss: tf.Tensor(0.5961712, shape=(), dtype=float32)\n",
            "epoch: 2 step: 785 train batch loss: tf.Tensor(0.510249, shape=(), dtype=float32)\n",
            "epoch: 2 step: 786 train batch loss: tf.Tensor(0.48686224, shape=(), dtype=float32)\n",
            "epoch: 2 step: 787 train batch loss: tf.Tensor(0.59860885, shape=(), dtype=float32)\n",
            "epoch: 2 step: 788 train batch loss: tf.Tensor(0.46518022, shape=(), dtype=float32)\n",
            "epoch: 2 step: 789 train batch loss: tf.Tensor(0.52449214, shape=(), dtype=float32)\n",
            "epoch: 2 step: 790 train batch loss: tf.Tensor(0.5216991, shape=(), dtype=float32)\n",
            "epoch: 2 step: 791 train batch loss: tf.Tensor(0.5770604, shape=(), dtype=float32)\n",
            "epoch: 2 step: 792 train batch loss: tf.Tensor(0.5319754, shape=(), dtype=float32)\n",
            "epoch: 2 step: 793 train batch loss: tf.Tensor(0.55474395, shape=(), dtype=float32)\n",
            "epoch: 2 step: 794 train batch loss: tf.Tensor(0.5433789, shape=(), dtype=float32)\n",
            "epoch: 2 step: 795 train batch loss: tf.Tensor(0.4871409, shape=(), dtype=float32)\n",
            "epoch: 2 step: 796 train batch loss: tf.Tensor(0.60916036, shape=(), dtype=float32)\n",
            "epoch: 2 step: 797 train batch loss: tf.Tensor(0.59307367, shape=(), dtype=float32)\n",
            "epoch: 2 step: 798 train batch loss: tf.Tensor(0.5628316, shape=(), dtype=float32)\n",
            "epoch: 2 step: 799 train batch loss: tf.Tensor(0.5003792, shape=(), dtype=float32)\n",
            "epoch: 2 step: 800 train batch loss: tf.Tensor(0.65597874, shape=(), dtype=float32)\n",
            "epoch: 2 step: 801 train batch loss: tf.Tensor(0.6087756, shape=(), dtype=float32)\n",
            "epoch: 2 step: 802 train batch loss: tf.Tensor(0.5705637, shape=(), dtype=float32)\n",
            "epoch: 2 step: 803 train batch loss: tf.Tensor(0.759586, shape=(), dtype=float32)\n",
            "epoch: 2 step: 804 train batch loss: tf.Tensor(0.61388016, shape=(), dtype=float32)\n",
            "epoch: 2 step: 805 train batch loss: tf.Tensor(0.51168334, shape=(), dtype=float32)\n",
            "epoch: 2 step: 806 train batch loss: tf.Tensor(0.542076, shape=(), dtype=float32)\n",
            "epoch: 2 step: 807 train batch loss: tf.Tensor(0.4387284, shape=(), dtype=float32)\n",
            "epoch: 2 step: 808 train batch loss: tf.Tensor(0.49101695, shape=(), dtype=float32)\n",
            "epoch: 2 step: 809 train batch loss: tf.Tensor(0.510117, shape=(), dtype=float32)\n",
            "epoch: 2 step: 810 train batch loss: tf.Tensor(0.5248686, shape=(), dtype=float32)\n",
            "epoch: 2 step: 811 train batch loss: tf.Tensor(0.53366375, shape=(), dtype=float32)\n",
            "epoch: 2 step: 812 train batch loss: tf.Tensor(0.5824973, shape=(), dtype=float32)\n",
            "epoch: 2 step: 813 train batch loss: tf.Tensor(0.44530255, shape=(), dtype=float32)\n",
            "epoch: 2 step: 814 train batch loss: tf.Tensor(0.45170134, shape=(), dtype=float32)\n",
            "epoch: 2 step: 815 train batch loss: tf.Tensor(0.6394031, shape=(), dtype=float32)\n",
            "epoch: 2 step: 816 train batch loss: tf.Tensor(0.4730228, shape=(), dtype=float32)\n",
            "epoch: 2 step: 817 train batch loss: tf.Tensor(0.52769554, shape=(), dtype=float32)\n",
            "epoch: 2 step: 818 train batch loss: tf.Tensor(0.5235859, shape=(), dtype=float32)\n",
            "epoch: 2 step: 819 train batch loss: tf.Tensor(0.49431705, shape=(), dtype=float32)\n",
            "epoch: 2 step: 820 train batch loss: tf.Tensor(0.54084367, shape=(), dtype=float32)\n",
            "epoch: 2 step: 821 train batch loss: tf.Tensor(0.47922412, shape=(), dtype=float32)\n",
            "epoch: 2 step: 822 train batch loss: tf.Tensor(0.54088175, shape=(), dtype=float32)\n",
            "epoch: 2 step: 823 train batch loss: tf.Tensor(0.58799046, shape=(), dtype=float32)\n",
            "epoch: 2 step: 824 train batch loss: tf.Tensor(0.52659243, shape=(), dtype=float32)\n",
            "epoch: 2 step: 825 train batch loss: tf.Tensor(0.6122065, shape=(), dtype=float32)\n",
            "epoch: 2 step: 826 train batch loss: tf.Tensor(0.6056703, shape=(), dtype=float32)\n",
            "epoch: 2 step: 827 train batch loss: tf.Tensor(0.5484971, shape=(), dtype=float32)\n",
            "epoch: 2 step: 828 train batch loss: tf.Tensor(0.46050167, shape=(), dtype=float32)\n",
            "epoch: 2 step: 829 train batch loss: tf.Tensor(0.5816781, shape=(), dtype=float32)\n",
            "epoch: 2 step: 830 train batch loss: tf.Tensor(0.51876605, shape=(), dtype=float32)\n",
            "epoch: 2 step: 831 train batch loss: tf.Tensor(0.451012, shape=(), dtype=float32)\n",
            "epoch: 2 step: 832 train batch loss: tf.Tensor(0.537166, shape=(), dtype=float32)\n",
            "epoch: 2 step: 833 train batch loss: tf.Tensor(0.5861975, shape=(), dtype=float32)\n",
            "epoch: 2 step: 834 train batch loss: tf.Tensor(0.44631004, shape=(), dtype=float32)\n",
            "epoch: 2 step: 835 train batch loss: tf.Tensor(0.50667685, shape=(), dtype=float32)\n",
            "epoch: 2 step: 836 train batch loss: tf.Tensor(0.50962013, shape=(), dtype=float32)\n",
            "epoch: 2 step: 837 train batch loss: tf.Tensor(0.6665211, shape=(), dtype=float32)\n",
            "epoch: 2 step: 838 train batch loss: tf.Tensor(0.54007095, shape=(), dtype=float32)\n",
            "epoch: 2 step: 839 train batch loss: tf.Tensor(0.4617551, shape=(), dtype=float32)\n",
            "epoch: 2 step: 840 train batch loss: tf.Tensor(0.46550885, shape=(), dtype=float32)\n",
            "epoch: 2 step: 841 train batch loss: tf.Tensor(0.5223831, shape=(), dtype=float32)\n",
            "epoch: 2 step: 842 train batch loss: tf.Tensor(0.61263, shape=(), dtype=float32)\n",
            "epoch: 2 step: 843 train batch loss: tf.Tensor(0.69530714, shape=(), dtype=float32)\n",
            "epoch: 2 step: 844 train batch loss: tf.Tensor(0.4935349, shape=(), dtype=float32)\n",
            "epoch: 2 step: 845 train batch loss: tf.Tensor(0.53018624, shape=(), dtype=float32)\n",
            "epoch: 2 step: 846 train batch loss: tf.Tensor(0.44988436, shape=(), dtype=float32)\n",
            "epoch: 2 step: 847 train batch loss: tf.Tensor(0.57168317, shape=(), dtype=float32)\n",
            "epoch: 2 step: 848 train batch loss: tf.Tensor(0.58937436, shape=(), dtype=float32)\n",
            "epoch: 2 step: 849 train batch loss: tf.Tensor(0.5243235, shape=(), dtype=float32)\n",
            "epoch: 2 step: 850 train batch loss: tf.Tensor(0.4943182, shape=(), dtype=float32)\n",
            "epoch: 2 step: 851 train batch loss: tf.Tensor(0.51481116, shape=(), dtype=float32)\n",
            "epoch: 2 step: 852 train batch loss: tf.Tensor(0.6795452, shape=(), dtype=float32)\n",
            "epoch: 2 step: 853 train batch loss: tf.Tensor(0.4750604, shape=(), dtype=float32)\n",
            "epoch: 2 step: 854 train batch loss: tf.Tensor(0.5718866, shape=(), dtype=float32)\n",
            "epoch: 2 step: 855 train batch loss: tf.Tensor(0.4906478, shape=(), dtype=float32)\n",
            "epoch: 2 step: 856 train batch loss: tf.Tensor(0.6190996, shape=(), dtype=float32)\n",
            "epoch: 2 step: 857 train batch loss: tf.Tensor(0.4448433, shape=(), dtype=float32)\n",
            "epoch: 2 step: 858 train batch loss: tf.Tensor(0.66049874, shape=(), dtype=float32)\n",
            "epoch: 2 step: 859 train batch loss: tf.Tensor(0.48171663, shape=(), dtype=float32)\n",
            "epoch: 2 step: 860 train batch loss: tf.Tensor(0.49700394, shape=(), dtype=float32)\n",
            "epoch: 2 step: 861 train batch loss: tf.Tensor(0.58454764, shape=(), dtype=float32)\n",
            "epoch: 2 step: 862 train batch loss: tf.Tensor(0.5788632, shape=(), dtype=float32)\n",
            "epoch: 2 step: 863 train batch loss: tf.Tensor(0.5527428, shape=(), dtype=float32)\n",
            "epoch: 2 step: 864 train batch loss: tf.Tensor(0.46861196, shape=(), dtype=float32)\n",
            "epoch: 2 step: 865 train batch loss: tf.Tensor(0.47057146, shape=(), dtype=float32)\n",
            "epoch: 2 step: 866 train batch loss: tf.Tensor(0.52350396, shape=(), dtype=float32)\n",
            "epoch: 2 step: 867 train batch loss: tf.Tensor(0.46853113, shape=(), dtype=float32)\n",
            "epoch: 2 step: 868 train batch loss: tf.Tensor(0.42336768, shape=(), dtype=float32)\n",
            "epoch: 2 step: 869 train batch loss: tf.Tensor(0.50857913, shape=(), dtype=float32)\n",
            "epoch: 2 step: 870 train batch loss: tf.Tensor(0.47888756, shape=(), dtype=float32)\n",
            "epoch: 2 step: 871 train batch loss: tf.Tensor(0.51677316, shape=(), dtype=float32)\n",
            "epoch: 2 step: 872 train batch loss: tf.Tensor(0.44943023, shape=(), dtype=float32)\n",
            "epoch: 2 step: 873 train batch loss: tf.Tensor(0.5791602, shape=(), dtype=float32)\n",
            "epoch: 2 step: 874 train batch loss: tf.Tensor(0.5725146, shape=(), dtype=float32)\n",
            "epoch: 2 step: 875 train batch loss: tf.Tensor(0.47712165, shape=(), dtype=float32)\n",
            "epoch: 2 step: 876 train batch loss: tf.Tensor(0.61853385, shape=(), dtype=float32)\n",
            "epoch: 2 step: 877 train batch loss: tf.Tensor(0.43720585, shape=(), dtype=float32)\n",
            "epoch: 2 step: 878 train batch loss: tf.Tensor(0.6704031, shape=(), dtype=float32)\n",
            "epoch: 2 step: 879 train batch loss: tf.Tensor(0.54014504, shape=(), dtype=float32)\n",
            "epoch: 2 step: 880 train batch loss: tf.Tensor(0.58807105, shape=(), dtype=float32)\n",
            "epoch: 2 step: 881 train batch loss: tf.Tensor(0.5091732, shape=(), dtype=float32)\n",
            "epoch: 2 step: 882 train batch loss: tf.Tensor(0.44967502, shape=(), dtype=float32)\n",
            "epoch: 2 step: 883 train batch loss: tf.Tensor(0.46980733, shape=(), dtype=float32)\n",
            "epoch: 2 step: 884 train batch loss: tf.Tensor(0.52789974, shape=(), dtype=float32)\n",
            "epoch: 2 step: 885 train batch loss: tf.Tensor(0.5766772, shape=(), dtype=float32)\n",
            "epoch: 2 step: 886 train batch loss: tf.Tensor(0.61578983, shape=(), dtype=float32)\n",
            "epoch: 2 step: 887 train batch loss: tf.Tensor(0.57623684, shape=(), dtype=float32)\n",
            "epoch: 2 step: 888 train batch loss: tf.Tensor(0.52148527, shape=(), dtype=float32)\n",
            "epoch: 2 step: 889 train batch loss: tf.Tensor(0.4147904, shape=(), dtype=float32)\n",
            "epoch: 2 step: 890 train batch loss: tf.Tensor(0.71518385, shape=(), dtype=float32)\n",
            "epoch: 2 step: 891 train batch loss: tf.Tensor(0.45820445, shape=(), dtype=float32)\n",
            "epoch: 2 step: 892 train batch loss: tf.Tensor(0.49827433, shape=(), dtype=float32)\n",
            "epoch: 2 step: 893 train batch loss: tf.Tensor(0.53851914, shape=(), dtype=float32)\n",
            "epoch: 2 step: 894 train batch loss: tf.Tensor(0.5331789, shape=(), dtype=float32)\n",
            "epoch: 2 step: 895 train batch loss: tf.Tensor(0.62400085, shape=(), dtype=float32)\n",
            "epoch: 2 step: 896 train batch loss: tf.Tensor(0.48025727, shape=(), dtype=float32)\n",
            "epoch: 2 step: 897 train batch loss: tf.Tensor(0.53013754, shape=(), dtype=float32)\n",
            "epoch: 2 step: 898 train batch loss: tf.Tensor(0.6025477, shape=(), dtype=float32)\n",
            "epoch: 2 step: 899 train batch loss: tf.Tensor(0.57578313, shape=(), dtype=float32)\n",
            "epoch: 2 step: 900 train batch loss: tf.Tensor(0.51430297, shape=(), dtype=float32)\n",
            "epoch: 2 step: 901 train batch loss: tf.Tensor(0.5869394, shape=(), dtype=float32)\n",
            "epoch: 2 step: 902 train batch loss: tf.Tensor(0.6042925, shape=(), dtype=float32)\n",
            "epoch: 2 step: 903 train batch loss: tf.Tensor(0.46258354, shape=(), dtype=float32)\n",
            "epoch: 2 step: 904 train batch loss: tf.Tensor(0.47206193, shape=(), dtype=float32)\n",
            "epoch: 2 step: 905 train batch loss: tf.Tensor(0.55402195, shape=(), dtype=float32)\n",
            "epoch: 2 step: 906 train batch loss: tf.Tensor(0.46262068, shape=(), dtype=float32)\n",
            "epoch: 2 step: 907 train batch loss: tf.Tensor(0.53310084, shape=(), dtype=float32)\n",
            "epoch: 2 step: 908 train batch loss: tf.Tensor(0.5711079, shape=(), dtype=float32)\n",
            "epoch: 2 step: 909 train batch loss: tf.Tensor(0.6317643, shape=(), dtype=float32)\n",
            "epoch: 2 step: 910 train batch loss: tf.Tensor(0.662644, shape=(), dtype=float32)\n",
            "epoch: 2 step: 911 train batch loss: tf.Tensor(0.58883905, shape=(), dtype=float32)\n",
            "epoch: 2 step: 912 train batch loss: tf.Tensor(0.5304791, shape=(), dtype=float32)\n",
            "epoch: 2 step: 913 train batch loss: tf.Tensor(0.5502631, shape=(), dtype=float32)\n",
            "epoch: 2 step: 914 train batch loss: tf.Tensor(0.612219, shape=(), dtype=float32)\n",
            "epoch: 2 step: 915 train batch loss: tf.Tensor(0.4640077, shape=(), dtype=float32)\n",
            "epoch: 2 step: 916 train batch loss: tf.Tensor(0.49099582, shape=(), dtype=float32)\n",
            "epoch: 2 step: 917 train batch loss: tf.Tensor(0.644185, shape=(), dtype=float32)\n",
            "epoch: 2 step: 918 train batch loss: tf.Tensor(0.46286005, shape=(), dtype=float32)\n",
            "epoch: 2 step: 919 train batch loss: tf.Tensor(0.41131657, shape=(), dtype=float32)\n",
            "epoch: 2 step: 920 train batch loss: tf.Tensor(0.50636613, shape=(), dtype=float32)\n",
            "epoch: 2 step: 921 train batch loss: tf.Tensor(0.4973927, shape=(), dtype=float32)\n",
            "epoch: 2 step: 922 train batch loss: tf.Tensor(0.56791705, shape=(), dtype=float32)\n",
            "epoch: 2 step: 923 train batch loss: tf.Tensor(0.56092733, shape=(), dtype=float32)\n",
            "epoch: 2 step: 924 train batch loss: tf.Tensor(0.6035429, shape=(), dtype=float32)\n",
            "epoch: 2 step: 925 train batch loss: tf.Tensor(0.48473844, shape=(), dtype=float32)\n",
            "epoch: 2 step: 926 train batch loss: tf.Tensor(0.5095227, shape=(), dtype=float32)\n",
            "epoch: 2 step: 927 train batch loss: tf.Tensor(0.5265809, shape=(), dtype=float32)\n",
            "epoch: 2 step: 928 train batch loss: tf.Tensor(0.5162693, shape=(), dtype=float32)\n",
            "epoch: 2 step: 929 train batch loss: tf.Tensor(0.577226, shape=(), dtype=float32)\n",
            "epoch: 2 step: 930 train batch loss: tf.Tensor(0.5818864, shape=(), dtype=float32)\n",
            "epoch: 2 step: 931 train batch loss: tf.Tensor(0.42757687, shape=(), dtype=float32)\n",
            "epoch: 2 step: 932 train batch loss: tf.Tensor(0.5453032, shape=(), dtype=float32)\n",
            "epoch: 2 step: 933 train batch loss: tf.Tensor(0.50587773, shape=(), dtype=float32)\n",
            "epoch: 2 step: 934 train batch loss: tf.Tensor(0.6475748, shape=(), dtype=float32)\n",
            "epoch: 2 step: 935 train batch loss: tf.Tensor(0.5254583, shape=(), dtype=float32)\n",
            "epoch: 2 step: 936 train batch loss: tf.Tensor(0.5047772, shape=(), dtype=float32)\n",
            "epoch: 2 step: 937 train batch loss: tf.Tensor(0.4558711, shape=(), dtype=float32)\n",
            "epoch: 2 step: 938 train batch loss: tf.Tensor(0.48739266, shape=(), dtype=float32)\n",
            "epoch: 2 step: 939 train batch loss: tf.Tensor(0.46066222, shape=(), dtype=float32)\n",
            "epoch: 2 step: 940 train batch loss: tf.Tensor(0.53567255, shape=(), dtype=float32)\n",
            "epoch: 2 step: 941 train batch loss: tf.Tensor(0.57059383, shape=(), dtype=float32)\n",
            "epoch: 2 step: 942 train batch loss: tf.Tensor(0.6302413, shape=(), dtype=float32)\n",
            "epoch: 2 step: 943 train batch loss: tf.Tensor(0.47903854, shape=(), dtype=float32)\n",
            "epoch: 2 step: 944 train batch loss: tf.Tensor(0.5405637, shape=(), dtype=float32)\n",
            "epoch: 2 step: 945 train batch loss: tf.Tensor(0.5837377, shape=(), dtype=float32)\n",
            "epoch: 2 step: 946 train batch loss: tf.Tensor(0.5246638, shape=(), dtype=float32)\n",
            "epoch: 2 step: 947 train batch loss: tf.Tensor(0.59481895, shape=(), dtype=float32)\n",
            "epoch: 2 step: 948 train batch loss: tf.Tensor(0.4304793, shape=(), dtype=float32)\n",
            "epoch: 2 step: 949 train batch loss: tf.Tensor(0.55184495, shape=(), dtype=float32)\n",
            "epoch: 2 step: 950 train batch loss: tf.Tensor(0.6522122, shape=(), dtype=float32)\n",
            "epoch: 2 step: 951 train batch loss: tf.Tensor(0.5572986, shape=(), dtype=float32)\n",
            "epoch: 2 step: 952 train batch loss: tf.Tensor(0.5097342, shape=(), dtype=float32)\n",
            "epoch: 2 step: 953 train batch loss: tf.Tensor(0.48474675, shape=(), dtype=float32)\n",
            "epoch: 2 step: 954 train batch loss: tf.Tensor(0.4178238, shape=(), dtype=float32)\n",
            "epoch: 2 step: 955 train batch loss: tf.Tensor(0.63547266, shape=(), dtype=float32)\n",
            "epoch: 2 step: 956 train batch loss: tf.Tensor(0.57137406, shape=(), dtype=float32)\n",
            "epoch: 2 step: 957 train batch loss: tf.Tensor(0.45502907, shape=(), dtype=float32)\n",
            "epoch: 2 step: 958 train batch loss: tf.Tensor(0.5597385, shape=(), dtype=float32)\n",
            "epoch: 2 step: 959 train batch loss: tf.Tensor(0.4499213, shape=(), dtype=float32)\n",
            "epoch: 2 step: 960 train batch loss: tf.Tensor(0.57607925, shape=(), dtype=float32)\n",
            "epoch: 2 step: 961 train batch loss: tf.Tensor(0.4321781, shape=(), dtype=float32)\n",
            "epoch: 2 step: 962 train batch loss: tf.Tensor(0.58502394, shape=(), dtype=float32)\n",
            "epoch: 2 step: 963 train batch loss: tf.Tensor(0.60890335, shape=(), dtype=float32)\n",
            "epoch: 2 step: 964 train batch loss: tf.Tensor(0.4729003, shape=(), dtype=float32)\n",
            "epoch: 2 step: 965 train batch loss: tf.Tensor(0.640212, shape=(), dtype=float32)\n",
            "epoch: 2 step: 966 train batch loss: tf.Tensor(0.54906774, shape=(), dtype=float32)\n",
            "epoch: 2 step: 967 train batch loss: tf.Tensor(0.5842046, shape=(), dtype=float32)\n",
            "epoch: 2 step: 968 train batch loss: tf.Tensor(0.5169072, shape=(), dtype=float32)\n",
            "epoch: 2 step: 969 train batch loss: tf.Tensor(0.47012335, shape=(), dtype=float32)\n",
            "epoch: 2 step: 970 train batch loss: tf.Tensor(0.5462593, shape=(), dtype=float32)\n",
            "epoch: 2 step: 971 train batch loss: tf.Tensor(0.64034766, shape=(), dtype=float32)\n",
            "epoch: 2 step: 972 train batch loss: tf.Tensor(0.4408142, shape=(), dtype=float32)\n",
            "epoch: 2 step: 973 train batch loss: tf.Tensor(0.43548623, shape=(), dtype=float32)\n",
            "epoch: 2 step: 974 train batch loss: tf.Tensor(0.5310221, shape=(), dtype=float32)\n",
            "epoch: 2 step: 975 train batch loss: tf.Tensor(0.59504646, shape=(), dtype=float32)\n",
            "epoch: 2 step: 976 train batch loss: tf.Tensor(0.53741574, shape=(), dtype=float32)\n",
            "epoch: 2 step: 977 train batch loss: tf.Tensor(0.7034879, shape=(), dtype=float32)\n",
            "epoch: 2 step: 978 train batch loss: tf.Tensor(0.57348126, shape=(), dtype=float32)\n",
            "epoch: 2 step: 979 train batch loss: tf.Tensor(0.5373009, shape=(), dtype=float32)\n",
            "epoch: 2 step: 980 train batch loss: tf.Tensor(0.51039755, shape=(), dtype=float32)\n",
            "epoch: 2 step: 981 train batch loss: tf.Tensor(0.41021934, shape=(), dtype=float32)\n",
            "epoch: 2 step: 982 train batch loss: tf.Tensor(0.5653293, shape=(), dtype=float32)\n",
            "epoch: 2 step: 983 train batch loss: tf.Tensor(0.48875308, shape=(), dtype=float32)\n",
            "epoch: 2 step: 984 train batch loss: tf.Tensor(0.5599092, shape=(), dtype=float32)\n",
            "epoch: 2 step: 985 train batch loss: tf.Tensor(0.48253745, shape=(), dtype=float32)\n",
            "epoch: 2 step: 986 train batch loss: tf.Tensor(0.55216175, shape=(), dtype=float32)\n",
            "epoch: 2 step: 987 train batch loss: tf.Tensor(0.48670077, shape=(), dtype=float32)\n",
            "epoch: 2 step: 988 train batch loss: tf.Tensor(0.48003787, shape=(), dtype=float32)\n",
            "epoch: 2 step: 989 train batch loss: tf.Tensor(0.52965444, shape=(), dtype=float32)\n",
            "epoch: 2 step: 990 train batch loss: tf.Tensor(0.47632113, shape=(), dtype=float32)\n",
            "epoch: 2 step: 991 train batch loss: tf.Tensor(0.60078466, shape=(), dtype=float32)\n",
            "epoch: 2 step: 992 train batch loss: tf.Tensor(0.57162404, shape=(), dtype=float32)\n",
            "epoch: 2 step: 993 train batch loss: tf.Tensor(0.4914454, shape=(), dtype=float32)\n",
            "epoch: 2 step: 994 train batch loss: tf.Tensor(0.521935, shape=(), dtype=float32)\n",
            "epoch: 2 step: 995 train batch loss: tf.Tensor(0.53789157, shape=(), dtype=float32)\n",
            "epoch: 2 step: 996 train batch loss: tf.Tensor(0.6249782, shape=(), dtype=float32)\n",
            "epoch: 2 step: 997 train batch loss: tf.Tensor(0.5214459, shape=(), dtype=float32)\n",
            "epoch: 2 step: 998 train batch loss: tf.Tensor(0.5232974, shape=(), dtype=float32)\n",
            "epoch: 2 step: 999 train batch loss: tf.Tensor(0.59100854, shape=(), dtype=float32)\n",
            "epoch: 2 step: 0 val batch loss: tf.Tensor(0.77185667, shape=(), dtype=float32)\n",
            "epoch: 2 step: 1 val batch loss: tf.Tensor(0.84009063, shape=(), dtype=float32)\n",
            "epoch: 2 step: 2 val batch loss: tf.Tensor(0.69545865, shape=(), dtype=float32)\n",
            "epoch: 2 step: 3 val batch loss: tf.Tensor(1.1193596, shape=(), dtype=float32)\n",
            "epoch: 2 step: 4 val batch loss: tf.Tensor(0.6050052, shape=(), dtype=float32)\n",
            "epoch: 2 step: 5 val batch loss: tf.Tensor(0.6433227, shape=(), dtype=float32)\n",
            "epoch: 2 step: 6 val batch loss: tf.Tensor(0.5465933, shape=(), dtype=float32)\n",
            "epoch: 2 step: 7 val batch loss: tf.Tensor(0.58988124, shape=(), dtype=float32)\n",
            "epoch: 2 step: 8 val batch loss: tf.Tensor(0.49606332, shape=(), dtype=float32)\n",
            "epoch: 2 step: 9 val batch loss: tf.Tensor(0.52456385, shape=(), dtype=float32)\n",
            "epoch: 2 step: 10 val batch loss: tf.Tensor(0.6856784, shape=(), dtype=float32)\n",
            "epoch: 2 step: 11 val batch loss: tf.Tensor(0.8046419, shape=(), dtype=float32)\n",
            "epoch: 2 step: 12 val batch loss: tf.Tensor(0.5873031, shape=(), dtype=float32)\n",
            "epoch: 2 step: 13 val batch loss: tf.Tensor(0.5146409, shape=(), dtype=float32)\n",
            "epoch: 2 step: 14 val batch loss: tf.Tensor(0.8712238, shape=(), dtype=float32)\n",
            "epoch: 2 step: 15 val batch loss: tf.Tensor(0.5202418, shape=(), dtype=float32)\n",
            "epoch: 2 step: 16 val batch loss: tf.Tensor(0.76454645, shape=(), dtype=float32)\n",
            "epoch: 2 step: 17 val batch loss: tf.Tensor(0.9820161, shape=(), dtype=float32)\n",
            "epoch: 2 step: 18 val batch loss: tf.Tensor(0.86119163, shape=(), dtype=float32)\n",
            "epoch: 2 step: 19 val batch loss: tf.Tensor(0.48121756, shape=(), dtype=float32)\n",
            "epoch: 2 step: 20 val batch loss: tf.Tensor(0.52544737, shape=(), dtype=float32)\n",
            "epoch: 2 step: 21 val batch loss: tf.Tensor(0.75314206, shape=(), dtype=float32)\n",
            "epoch: 2 step: 22 val batch loss: tf.Tensor(0.71559864, shape=(), dtype=float32)\n",
            "epoch: 2 step: 23 val batch loss: tf.Tensor(0.73049444, shape=(), dtype=float32)\n",
            "epoch: 2 step: 24 val batch loss: tf.Tensor(0.74042016, shape=(), dtype=float32)\n",
            "epoch: 2 step: 25 val batch loss: tf.Tensor(0.81689775, shape=(), dtype=float32)\n",
            "epoch: 2 step: 26 val batch loss: tf.Tensor(0.5757761, shape=(), dtype=float32)\n",
            "epoch: 2 step: 27 val batch loss: tf.Tensor(0.83371675, shape=(), dtype=float32)\n",
            "epoch: 2 step: 28 val batch loss: tf.Tensor(0.82857054, shape=(), dtype=float32)\n",
            "epoch: 2 step: 29 val batch loss: tf.Tensor(0.58291334, shape=(), dtype=float32)\n",
            "epoch: 2 step: 30 val batch loss: tf.Tensor(0.7842852, shape=(), dtype=float32)\n",
            "epoch: 2 step: 31 val batch loss: tf.Tensor(0.69543594, shape=(), dtype=float32)\n",
            "epoch: 2 step: 32 val batch loss: tf.Tensor(0.72729933, shape=(), dtype=float32)\n",
            "epoch: 2 step: 33 val batch loss: tf.Tensor(0.7032905, shape=(), dtype=float32)\n",
            "epoch: 2 step: 34 val batch loss: tf.Tensor(0.7690264, shape=(), dtype=float32)\n",
            "epoch: 2 step: 35 val batch loss: tf.Tensor(0.5730202, shape=(), dtype=float32)\n",
            "epoch: 2 step: 36 val batch loss: tf.Tensor(0.6148201, shape=(), dtype=float32)\n",
            "epoch: 2 step: 37 val batch loss: tf.Tensor(0.5685828, shape=(), dtype=float32)\n",
            "epoch: 2 step: 38 val batch loss: tf.Tensor(0.6993443, shape=(), dtype=float32)\n",
            "epoch: 2 step: 39 val batch loss: tf.Tensor(0.5711535, shape=(), dtype=float32)\n",
            "epoch: 2 step: 40 val batch loss: tf.Tensor(0.69101423, shape=(), dtype=float32)\n",
            "epoch: 2 step: 41 val batch loss: tf.Tensor(0.6090914, shape=(), dtype=float32)\n",
            "epoch: 2 step: 42 val batch loss: tf.Tensor(0.82173383, shape=(), dtype=float32)\n",
            "epoch: 2 step: 43 val batch loss: tf.Tensor(0.6159097, shape=(), dtype=float32)\n",
            "epoch: 2 step: 44 val batch loss: tf.Tensor(0.6560639, shape=(), dtype=float32)\n",
            "epoch: 2 step: 45 val batch loss: tf.Tensor(0.8010744, shape=(), dtype=float32)\n",
            "epoch: 2 step: 46 val batch loss: tf.Tensor(0.8684505, shape=(), dtype=float32)\n",
            "epoch: 2 step: 47 val batch loss: tf.Tensor(0.6190309, shape=(), dtype=float32)\n",
            "epoch: 2 step: 48 val batch loss: tf.Tensor(0.5906021, shape=(), dtype=float32)\n",
            "epoch: 2 step: 49 val batch loss: tf.Tensor(0.61590976, shape=(), dtype=float32)\n",
            "epoch: 2 step: 50 val batch loss: tf.Tensor(0.699101, shape=(), dtype=float32)\n",
            "epoch: 2 step: 51 val batch loss: tf.Tensor(0.65139043, shape=(), dtype=float32)\n",
            "epoch: 2 step: 52 val batch loss: tf.Tensor(0.75490296, shape=(), dtype=float32)\n",
            "epoch: 2 step: 53 val batch loss: tf.Tensor(0.70998085, shape=(), dtype=float32)\n",
            "epoch: 2 step: 54 val batch loss: tf.Tensor(0.5110354, shape=(), dtype=float32)\n",
            "epoch: 2 step: 55 val batch loss: tf.Tensor(0.7748004, shape=(), dtype=float32)\n",
            "epoch: 2 step: 56 val batch loss: tf.Tensor(0.61370754, shape=(), dtype=float32)\n",
            "epoch: 2 step: 57 val batch loss: tf.Tensor(0.6823634, shape=(), dtype=float32)\n",
            "epoch: 2 step: 58 val batch loss: tf.Tensor(0.77712023, shape=(), dtype=float32)\n",
            "epoch: 2 step: 59 val batch loss: tf.Tensor(0.70937234, shape=(), dtype=float32)\n",
            "epoch: 2 step: 60 val batch loss: tf.Tensor(0.62372506, shape=(), dtype=float32)\n",
            "epoch: 2 step: 61 val batch loss: tf.Tensor(0.64129794, shape=(), dtype=float32)\n",
            "epoch: 2 step: 62 val batch loss: tf.Tensor(0.7670596, shape=(), dtype=float32)\n",
            "epoch: 2 step: 63 val batch loss: tf.Tensor(0.57809705, shape=(), dtype=float32)\n",
            "epoch: 2 step: 64 val batch loss: tf.Tensor(0.64656055, shape=(), dtype=float32)\n",
            "epoch: 2 step: 65 val batch loss: tf.Tensor(0.6234711, shape=(), dtype=float32)\n",
            "epoch: 2 step: 66 val batch loss: tf.Tensor(0.7140752, shape=(), dtype=float32)\n",
            "epoch: 2 step: 67 val batch loss: tf.Tensor(0.80447996, shape=(), dtype=float32)\n",
            "epoch: 2 step: 68 val batch loss: tf.Tensor(0.67034304, shape=(), dtype=float32)\n",
            "epoch: 2 step: 69 val batch loss: tf.Tensor(0.5926706, shape=(), dtype=float32)\n",
            "epoch: 2 step: 70 val batch loss: tf.Tensor(0.8103241, shape=(), dtype=float32)\n",
            "epoch: 2 step: 71 val batch loss: tf.Tensor(0.6950645, shape=(), dtype=float32)\n",
            "epoch: 2 step: 72 val batch loss: tf.Tensor(0.7044744, shape=(), dtype=float32)\n",
            "epoch: 2 step: 73 val batch loss: tf.Tensor(0.7832746, shape=(), dtype=float32)\n",
            "epoch: 2 step: 74 val batch loss: tf.Tensor(0.7787563, shape=(), dtype=float32)\n",
            "epoch: 2 step: 75 val batch loss: tf.Tensor(0.60907257, shape=(), dtype=float32)\n",
            "epoch: 2 step: 76 val batch loss: tf.Tensor(0.77776676, shape=(), dtype=float32)\n",
            "epoch: 2 step: 77 val batch loss: tf.Tensor(0.7505722, shape=(), dtype=float32)\n",
            "epoch: 2 step: 78 val batch loss: tf.Tensor(0.6332997, shape=(), dtype=float32)\n",
            "epoch: 2 step: 79 val batch loss: tf.Tensor(0.73461926, shape=(), dtype=float32)\n",
            "epoch: 2 step: 80 val batch loss: tf.Tensor(0.60258865, shape=(), dtype=float32)\n",
            "epoch: 2 step: 81 val batch loss: tf.Tensor(0.7577384, shape=(), dtype=float32)\n",
            "epoch: 2 step: 82 val batch loss: tf.Tensor(0.6879877, shape=(), dtype=float32)\n",
            "epoch: 2 step: 83 val batch loss: tf.Tensor(0.64378554, shape=(), dtype=float32)\n",
            "epoch: 2 step: 84 val batch loss: tf.Tensor(0.5843431, shape=(), dtype=float32)\n",
            "epoch: 2 step: 85 val batch loss: tf.Tensor(0.6672335, shape=(), dtype=float32)\n",
            "epoch: 2 step: 86 val batch loss: tf.Tensor(0.6964216, shape=(), dtype=float32)\n",
            "epoch: 2 step: 87 val batch loss: tf.Tensor(0.7010955, shape=(), dtype=float32)\n",
            "epoch: 2 step: 88 val batch loss: tf.Tensor(0.706374, shape=(), dtype=float32)\n",
            "epoch: 2 step: 89 val batch loss: tf.Tensor(0.5619149, shape=(), dtype=float32)\n",
            "epoch: 2 step: 90 val batch loss: tf.Tensor(0.86979055, shape=(), dtype=float32)\n",
            "epoch: 2 step: 91 val batch loss: tf.Tensor(0.72138053, shape=(), dtype=float32)\n",
            "epoch: 2 step: 92 val batch loss: tf.Tensor(0.7267098, shape=(), dtype=float32)\n",
            "epoch: 2 step: 93 val batch loss: tf.Tensor(0.6633295, shape=(), dtype=float32)\n",
            "epoch: 2 step: 94 val batch loss: tf.Tensor(0.52984655, shape=(), dtype=float32)\n",
            "epoch: 2 step: 95 val batch loss: tf.Tensor(0.5443605, shape=(), dtype=float32)\n",
            "epoch: 2 step: 96 val batch loss: tf.Tensor(0.51189643, shape=(), dtype=float32)\n",
            "epoch: 2 step: 97 val batch loss: tf.Tensor(0.7899, shape=(), dtype=float32)\n",
            "epoch: 2 step: 98 val batch loss: tf.Tensor(0.84981984, shape=(), dtype=float32)\n",
            "epoch: 2 step: 99 val batch loss: tf.Tensor(0.7062156, shape=(), dtype=float32)\n",
            "epoch: 2 step: 100 val batch loss: tf.Tensor(0.87823313, shape=(), dtype=float32)\n",
            "epoch: 2 step: 101 val batch loss: tf.Tensor(0.65875727, shape=(), dtype=float32)\n",
            "epoch: 2 step: 102 val batch loss: tf.Tensor(0.76398593, shape=(), dtype=float32)\n",
            "epoch: 2 step: 103 val batch loss: tf.Tensor(0.64876753, shape=(), dtype=float32)\n",
            "epoch: 2 step: 104 val batch loss: tf.Tensor(0.74089235, shape=(), dtype=float32)\n",
            "epoch: 2 step: 105 val batch loss: tf.Tensor(0.51046056, shape=(), dtype=float32)\n",
            "epoch: 2 step: 106 val batch loss: tf.Tensor(0.6761503, shape=(), dtype=float32)\n",
            "epoch: 2 step: 107 val batch loss: tf.Tensor(0.61296844, shape=(), dtype=float32)\n",
            "epoch: 2 step: 108 val batch loss: tf.Tensor(0.6781379, shape=(), dtype=float32)\n",
            "epoch: 2 step: 109 val batch loss: tf.Tensor(0.70578796, shape=(), dtype=float32)\n",
            "epoch: 2 step: 110 val batch loss: tf.Tensor(0.5953102, shape=(), dtype=float32)\n",
            "epoch: 2 step: 111 val batch loss: tf.Tensor(0.63148963, shape=(), dtype=float32)\n",
            "epoch: 2 step: 112 val batch loss: tf.Tensor(0.7864677, shape=(), dtype=float32)\n",
            "epoch: 2 step: 113 val batch loss: tf.Tensor(0.7093466, shape=(), dtype=float32)\n",
            "epoch: 2 step: 114 val batch loss: tf.Tensor(0.73020667, shape=(), dtype=float32)\n",
            "epoch: 2 step: 115 val batch loss: tf.Tensor(0.59042037, shape=(), dtype=float32)\n",
            "epoch: 2 step: 116 val batch loss: tf.Tensor(0.74632925, shape=(), dtype=float32)\n",
            "epoch: 2 step: 117 val batch loss: tf.Tensor(0.5502705, shape=(), dtype=float32)\n",
            "epoch: 2 step: 118 val batch loss: tf.Tensor(0.5970816, shape=(), dtype=float32)\n",
            "epoch: 2 step: 119 val batch loss: tf.Tensor(0.75131273, shape=(), dtype=float32)\n",
            "epoch: 2 step: 120 val batch loss: tf.Tensor(0.59351194, shape=(), dtype=float32)\n",
            "epoch: 2 step: 121 val batch loss: tf.Tensor(0.94173324, shape=(), dtype=float32)\n",
            "epoch: 2 step: 122 val batch loss: tf.Tensor(0.61854994, shape=(), dtype=float32)\n",
            "epoch: 2 step: 123 val batch loss: tf.Tensor(0.70992935, shape=(), dtype=float32)\n",
            "epoch: 2 step: 124 val batch loss: tf.Tensor(0.7214096, shape=(), dtype=float32)\n",
            "epoch: 2 step: 125 val batch loss: tf.Tensor(0.7118123, shape=(), dtype=float32)\n",
            "epoch: 2 step: 126 val batch loss: tf.Tensor(0.80202353, shape=(), dtype=float32)\n",
            "epoch: 2 step: 127 val batch loss: tf.Tensor(0.70702994, shape=(), dtype=float32)\n",
            "epoch: 2 step: 128 val batch loss: tf.Tensor(0.6948372, shape=(), dtype=float32)\n",
            "epoch: 2 step: 129 val batch loss: tf.Tensor(0.6615217, shape=(), dtype=float32)\n",
            "epoch: 2 step: 130 val batch loss: tf.Tensor(0.6639286, shape=(), dtype=float32)\n",
            "epoch: 2 step: 131 val batch loss: tf.Tensor(0.7460212, shape=(), dtype=float32)\n",
            "epoch: 2 step: 132 val batch loss: tf.Tensor(0.5896117, shape=(), dtype=float32)\n",
            "epoch: 2 step: 133 val batch loss: tf.Tensor(0.62089026, shape=(), dtype=float32)\n",
            "epoch: 2 step: 134 val batch loss: tf.Tensor(0.73336715, shape=(), dtype=float32)\n",
            "epoch: 2 step: 135 val batch loss: tf.Tensor(0.55773884, shape=(), dtype=float32)\n",
            "epoch: 2 step: 136 val batch loss: tf.Tensor(0.60223, shape=(), dtype=float32)\n",
            "epoch: 2 step: 137 val batch loss: tf.Tensor(0.5730175, shape=(), dtype=float32)\n",
            "epoch: 2 step: 138 val batch loss: tf.Tensor(0.60712016, shape=(), dtype=float32)\n",
            "epoch: 2 step: 139 val batch loss: tf.Tensor(0.76058763, shape=(), dtype=float32)\n",
            "epoch: 2 step: 140 val batch loss: tf.Tensor(0.5469717, shape=(), dtype=float32)\n",
            "epoch: 2 step: 141 val batch loss: tf.Tensor(0.60323787, shape=(), dtype=float32)\n",
            "epoch: 2 step: 142 val batch loss: tf.Tensor(0.7630881, shape=(), dtype=float32)\n",
            "epoch: 2 step: 143 val batch loss: tf.Tensor(0.72245705, shape=(), dtype=float32)\n",
            "epoch: 2 step: 144 val batch loss: tf.Tensor(0.7015963, shape=(), dtype=float32)\n",
            "epoch: 2 step: 145 val batch loss: tf.Tensor(0.69809663, shape=(), dtype=float32)\n",
            "epoch: 2 step: 146 val batch loss: tf.Tensor(0.64713925, shape=(), dtype=float32)\n",
            "epoch: 2 step: 147 val batch loss: tf.Tensor(0.71636736, shape=(), dtype=float32)\n",
            "epoch: 2 step: 148 val batch loss: tf.Tensor(0.60557526, shape=(), dtype=float32)\n",
            "epoch: 2 step: 149 val batch loss: tf.Tensor(0.6986552, shape=(), dtype=float32)\n",
            "epoch: 2 step: 150 val batch loss: tf.Tensor(0.6023706, shape=(), dtype=float32)\n",
            "epoch: 2 step: 151 val batch loss: tf.Tensor(0.71405137, shape=(), dtype=float32)\n",
            "epoch: 2 step: 152 val batch loss: tf.Tensor(0.6203111, shape=(), dtype=float32)\n",
            "epoch: 2 step: 153 val batch loss: tf.Tensor(0.58845675, shape=(), dtype=float32)\n",
            "epoch: 2 step: 154 val batch loss: tf.Tensor(0.6273932, shape=(), dtype=float32)\n",
            "epoch: 2 step: 155 val batch loss: tf.Tensor(0.7561129, shape=(), dtype=float32)\n",
            "epoch: 2 step: 156 val batch loss: tf.Tensor(0.575337, shape=(), dtype=float32)\n",
            "epoch: 2 step: 157 val batch loss: tf.Tensor(0.6119437, shape=(), dtype=float32)\n",
            "epoch: 2 step: 158 val batch loss: tf.Tensor(0.63407815, shape=(), dtype=float32)\n",
            "epoch: 2 step: 159 val batch loss: tf.Tensor(0.5320412, shape=(), dtype=float32)\n",
            "epoch: 2 step: 160 val batch loss: tf.Tensor(0.6963458, shape=(), dtype=float32)\n",
            "epoch: 2 step: 161 val batch loss: tf.Tensor(0.5107604, shape=(), dtype=float32)\n",
            "epoch: 2 step: 162 val batch loss: tf.Tensor(0.7340275, shape=(), dtype=float32)\n",
            "epoch: 2 step: 163 val batch loss: tf.Tensor(0.7020133, shape=(), dtype=float32)\n",
            "epoch: 2 step: 164 val batch loss: tf.Tensor(0.81670046, shape=(), dtype=float32)\n",
            "epoch: 2 step: 165 val batch loss: tf.Tensor(0.76401603, shape=(), dtype=float32)\n",
            "epoch: 2 step: 166 val batch loss: tf.Tensor(0.67472994, shape=(), dtype=float32)\n",
            "epoch: 2 step: 167 val batch loss: tf.Tensor(0.6716889, shape=(), dtype=float32)\n",
            "epoch: 2 step: 168 val batch loss: tf.Tensor(0.54170674, shape=(), dtype=float32)\n",
            "epoch: 2 step: 169 val batch loss: tf.Tensor(0.57574785, shape=(), dtype=float32)\n",
            "epoch: 2 step: 170 val batch loss: tf.Tensor(0.6744722, shape=(), dtype=float32)\n",
            "epoch: 2 step: 171 val batch loss: tf.Tensor(0.6987575, shape=(), dtype=float32)\n",
            "epoch: 2 step: 172 val batch loss: tf.Tensor(0.53925323, shape=(), dtype=float32)\n",
            "epoch: 2 step: 173 val batch loss: tf.Tensor(0.66625035, shape=(), dtype=float32)\n",
            "epoch: 2 step: 174 val batch loss: tf.Tensor(0.82213926, shape=(), dtype=float32)\n",
            "epoch: 2 step: 175 val batch loss: tf.Tensor(0.76849836, shape=(), dtype=float32)\n",
            "epoch: 2 step: 176 val batch loss: tf.Tensor(0.6375407, shape=(), dtype=float32)\n",
            "epoch: 2 step: 177 val batch loss: tf.Tensor(0.64811754, shape=(), dtype=float32)\n",
            "epoch: 2 step: 178 val batch loss: tf.Tensor(0.52574605, shape=(), dtype=float32)\n",
            "epoch: 2 step: 179 val batch loss: tf.Tensor(0.62618905, shape=(), dtype=float32)\n",
            "epoch: 2 step: 180 val batch loss: tf.Tensor(0.7220542, shape=(), dtype=float32)\n",
            "epoch: 2 step: 181 val batch loss: tf.Tensor(0.5716969, shape=(), dtype=float32)\n",
            "epoch: 2 step: 182 val batch loss: tf.Tensor(0.66275686, shape=(), dtype=float32)\n",
            "epoch: 2 step: 183 val batch loss: tf.Tensor(0.6706114, shape=(), dtype=float32)\n",
            "epoch: 2 step: 184 val batch loss: tf.Tensor(0.55298597, shape=(), dtype=float32)\n",
            "epoch: 2 step: 185 val batch loss: tf.Tensor(0.72312796, shape=(), dtype=float32)\n",
            "epoch: 2 step: 186 val batch loss: tf.Tensor(0.75661457, shape=(), dtype=float32)\n",
            "epoch: 2 step: 187 val batch loss: tf.Tensor(0.7567645, shape=(), dtype=float32)\n",
            "epoch: 2 step: 188 val batch loss: tf.Tensor(0.6367178, shape=(), dtype=float32)\n",
            "epoch: 2 step: 189 val batch loss: tf.Tensor(0.58404285, shape=(), dtype=float32)\n",
            "epoch: 2 step: 190 val batch loss: tf.Tensor(0.6189705, shape=(), dtype=float32)\n",
            "epoch: 2 step: 191 val batch loss: tf.Tensor(0.6430596, shape=(), dtype=float32)\n",
            "epoch: 2 step: 192 val batch loss: tf.Tensor(0.76097643, shape=(), dtype=float32)\n",
            "epoch: 2 step: 193 val batch loss: tf.Tensor(0.5695765, shape=(), dtype=float32)\n",
            "epoch: 2 step: 194 val batch loss: tf.Tensor(0.59433967, shape=(), dtype=float32)\n",
            "epoch: 2 step: 195 val batch loss: tf.Tensor(0.6340977, shape=(), dtype=float32)\n",
            "epoch: 2 step: 196 val batch loss: tf.Tensor(0.7030457, shape=(), dtype=float32)\n",
            "epoch: 2 step: 197 val batch loss: tf.Tensor(0.6505309, shape=(), dtype=float32)\n",
            "epoch: 2 step: 198 val batch loss: tf.Tensor(0.6658497, shape=(), dtype=float32)\n",
            "epoch: 2 step: 199 val batch loss: tf.Tensor(0.5698132, shape=(), dtype=float32)\n",
            "epoch: 2 step: 200 val batch loss: tf.Tensor(0.60214716, shape=(), dtype=float32)\n",
            "epoch: 2 step: 201 val batch loss: tf.Tensor(0.7115077, shape=(), dtype=float32)\n",
            "epoch: 2 step: 202 val batch loss: tf.Tensor(0.6430516, shape=(), dtype=float32)\n",
            "epoch: 2 step: 203 val batch loss: tf.Tensor(0.89458656, shape=(), dtype=float32)\n",
            "epoch: 2 step: 204 val batch loss: tf.Tensor(0.5829216, shape=(), dtype=float32)\n",
            "epoch: 2 step: 205 val batch loss: tf.Tensor(0.64238906, shape=(), dtype=float32)\n",
            "epoch: 2 step: 206 val batch loss: tf.Tensor(0.5301746, shape=(), dtype=float32)\n",
            "epoch: 2 step: 207 val batch loss: tf.Tensor(0.5955053, shape=(), dtype=float32)\n",
            "epoch: 2 step: 208 val batch loss: tf.Tensor(0.53714883, shape=(), dtype=float32)\n",
            "epoch: 2 step: 209 val batch loss: tf.Tensor(0.51537275, shape=(), dtype=float32)\n",
            "epoch: 2 step: 210 val batch loss: tf.Tensor(0.61656356, shape=(), dtype=float32)\n",
            "epoch: 2 step: 211 val batch loss: tf.Tensor(0.59584, shape=(), dtype=float32)\n",
            "epoch: 2 step: 212 val batch loss: tf.Tensor(0.6797339, shape=(), dtype=float32)\n",
            "epoch: 2 step: 213 val batch loss: tf.Tensor(0.798148, shape=(), dtype=float32)\n",
            "epoch: 2 step: 214 val batch loss: tf.Tensor(0.6599916, shape=(), dtype=float32)\n",
            "epoch: 2 step: 215 val batch loss: tf.Tensor(0.59882194, shape=(), dtype=float32)\n",
            "epoch: 2 step: 216 val batch loss: tf.Tensor(0.56735456, shape=(), dtype=float32)\n",
            "epoch: 2 step: 217 val batch loss: tf.Tensor(0.630861, shape=(), dtype=float32)\n",
            "epoch: 2 step: 218 val batch loss: tf.Tensor(0.66201323, shape=(), dtype=float32)\n",
            "epoch: 2 step: 219 val batch loss: tf.Tensor(0.49726978, shape=(), dtype=float32)\n",
            "epoch: 2 step: 220 val batch loss: tf.Tensor(0.5372703, shape=(), dtype=float32)\n",
            "epoch: 2 step: 221 val batch loss: tf.Tensor(0.7513734, shape=(), dtype=float32)\n",
            "epoch: 2 step: 222 val batch loss: tf.Tensor(0.7705304, shape=(), dtype=float32)\n",
            "epoch: 2 step: 223 val batch loss: tf.Tensor(0.8559791, shape=(), dtype=float32)\n",
            "epoch: 2 step: 224 val batch loss: tf.Tensor(0.7760791, shape=(), dtype=float32)\n",
            "epoch: 2 step: 225 val batch loss: tf.Tensor(0.6771109, shape=(), dtype=float32)\n",
            "epoch: 2 step: 226 val batch loss: tf.Tensor(0.7382515, shape=(), dtype=float32)\n",
            "epoch: 2 step: 227 val batch loss: tf.Tensor(0.653164, shape=(), dtype=float32)\n",
            "epoch: 2 step: 228 val batch loss: tf.Tensor(0.5271411, shape=(), dtype=float32)\n",
            "epoch: 2 step: 229 val batch loss: tf.Tensor(1.0097188, shape=(), dtype=float32)\n",
            "epoch: 2 step: 230 val batch loss: tf.Tensor(0.7626753, shape=(), dtype=float32)\n",
            "epoch: 2 step: 231 val batch loss: tf.Tensor(0.51975286, shape=(), dtype=float32)\n",
            "epoch: 2 step: 232 val batch loss: tf.Tensor(0.77646494, shape=(), dtype=float32)\n",
            "epoch: 2 step: 233 val batch loss: tf.Tensor(0.6323601, shape=(), dtype=float32)\n",
            "epoch: 2 step: 234 val batch loss: tf.Tensor(0.69939184, shape=(), dtype=float32)\n",
            "epoch: 2 step: 235 val batch loss: tf.Tensor(0.6196942, shape=(), dtype=float32)\n",
            "epoch: 2 step: 236 val batch loss: tf.Tensor(0.7400107, shape=(), dtype=float32)\n",
            "epoch: 2 step: 237 val batch loss: tf.Tensor(0.8101842, shape=(), dtype=float32)\n",
            "epoch: 2 step: 238 val batch loss: tf.Tensor(0.72346216, shape=(), dtype=float32)\n",
            "epoch: 2 step: 239 val batch loss: tf.Tensor(0.6983984, shape=(), dtype=float32)\n",
            "epoch: 2 step: 240 val batch loss: tf.Tensor(0.66477, shape=(), dtype=float32)\n",
            "epoch: 2 step: 241 val batch loss: tf.Tensor(0.6792201, shape=(), dtype=float32)\n",
            "epoch: 2 step: 242 val batch loss: tf.Tensor(0.6360415, shape=(), dtype=float32)\n",
            "epoch: 2 step: 243 val batch loss: tf.Tensor(0.78549856, shape=(), dtype=float32)\n",
            "epoch: 2 step: 244 val batch loss: tf.Tensor(0.6131303, shape=(), dtype=float32)\n",
            "epoch: 2 step: 245 val batch loss: tf.Tensor(0.5295189, shape=(), dtype=float32)\n",
            "epoch: 2 step: 246 val batch loss: tf.Tensor(0.55350494, shape=(), dtype=float32)\n",
            "epoch: 2 step: 247 val batch loss: tf.Tensor(0.78044766, shape=(), dtype=float32)\n",
            "epoch: 2 step: 248 val batch loss: tf.Tensor(0.8103334, shape=(), dtype=float32)\n",
            "epoch: 2 step: 249 val batch loss: tf.Tensor(0.70508426, shape=(), dtype=float32)\n",
            "Epoch 2 training loss: 0.579232931137085 val loss: 0.676133394241333\n",
            "epoch: 3 step: 0 train batch loss: tf.Tensor(0.47318068, shape=(), dtype=float32)\n",
            "epoch: 3 step: 1 train batch loss: tf.Tensor(0.47294313, shape=(), dtype=float32)\n",
            "epoch: 3 step: 2 train batch loss: tf.Tensor(0.5983586, shape=(), dtype=float32)\n",
            "epoch: 3 step: 3 train batch loss: tf.Tensor(0.46103278, shape=(), dtype=float32)\n",
            "epoch: 3 step: 4 train batch loss: tf.Tensor(0.6230073, shape=(), dtype=float32)\n",
            "epoch: 3 step: 5 train batch loss: tf.Tensor(0.41835198, shape=(), dtype=float32)\n",
            "epoch: 3 step: 6 train batch loss: tf.Tensor(0.5997542, shape=(), dtype=float32)\n",
            "epoch: 3 step: 7 train batch loss: tf.Tensor(0.517215, shape=(), dtype=float32)\n",
            "epoch: 3 step: 8 train batch loss: tf.Tensor(0.5557904, shape=(), dtype=float32)\n",
            "epoch: 3 step: 9 train batch loss: tf.Tensor(0.6061314, shape=(), dtype=float32)\n",
            "epoch: 3 step: 10 train batch loss: tf.Tensor(0.5365628, shape=(), dtype=float32)\n",
            "epoch: 3 step: 11 train batch loss: tf.Tensor(0.555225, shape=(), dtype=float32)\n",
            "epoch: 3 step: 12 train batch loss: tf.Tensor(0.5502924, shape=(), dtype=float32)\n",
            "epoch: 3 step: 13 train batch loss: tf.Tensor(0.52981216, shape=(), dtype=float32)\n",
            "epoch: 3 step: 14 train batch loss: tf.Tensor(0.5049406, shape=(), dtype=float32)\n",
            "epoch: 3 step: 15 train batch loss: tf.Tensor(0.48453385, shape=(), dtype=float32)\n",
            "epoch: 3 step: 16 train batch loss: tf.Tensor(0.54964685, shape=(), dtype=float32)\n",
            "epoch: 3 step: 17 train batch loss: tf.Tensor(0.5173588, shape=(), dtype=float32)\n",
            "epoch: 3 step: 18 train batch loss: tf.Tensor(0.47163382, shape=(), dtype=float32)\n",
            "epoch: 3 step: 19 train batch loss: tf.Tensor(0.51395833, shape=(), dtype=float32)\n",
            "epoch: 3 step: 20 train batch loss: tf.Tensor(0.60481656, shape=(), dtype=float32)\n",
            "epoch: 3 step: 21 train batch loss: tf.Tensor(0.5261572, shape=(), dtype=float32)\n",
            "epoch: 3 step: 22 train batch loss: tf.Tensor(0.4773628, shape=(), dtype=float32)\n",
            "epoch: 3 step: 23 train batch loss: tf.Tensor(0.47909155, shape=(), dtype=float32)\n",
            "epoch: 3 step: 24 train batch loss: tf.Tensor(0.5812529, shape=(), dtype=float32)\n",
            "epoch: 3 step: 25 train batch loss: tf.Tensor(0.46821672, shape=(), dtype=float32)\n",
            "epoch: 3 step: 26 train batch loss: tf.Tensor(0.52335155, shape=(), dtype=float32)\n",
            "epoch: 3 step: 27 train batch loss: tf.Tensor(0.4294601, shape=(), dtype=float32)\n",
            "epoch: 3 step: 28 train batch loss: tf.Tensor(0.5576492, shape=(), dtype=float32)\n",
            "epoch: 3 step: 29 train batch loss: tf.Tensor(0.4907822, shape=(), dtype=float32)\n",
            "epoch: 3 step: 30 train batch loss: tf.Tensor(0.5075923, shape=(), dtype=float32)\n",
            "epoch: 3 step: 31 train batch loss: tf.Tensor(0.56042826, shape=(), dtype=float32)\n",
            "epoch: 3 step: 32 train batch loss: tf.Tensor(0.47166672, shape=(), dtype=float32)\n",
            "epoch: 3 step: 33 train batch loss: tf.Tensor(0.43525046, shape=(), dtype=float32)\n",
            "epoch: 3 step: 34 train batch loss: tf.Tensor(0.48939642, shape=(), dtype=float32)\n",
            "epoch: 3 step: 35 train batch loss: tf.Tensor(0.49110818, shape=(), dtype=float32)\n",
            "epoch: 3 step: 36 train batch loss: tf.Tensor(0.40313235, shape=(), dtype=float32)\n",
            "epoch: 3 step: 37 train batch loss: tf.Tensor(0.44035754, shape=(), dtype=float32)\n",
            "epoch: 3 step: 38 train batch loss: tf.Tensor(0.50479627, shape=(), dtype=float32)\n",
            "epoch: 3 step: 39 train batch loss: tf.Tensor(0.5735789, shape=(), dtype=float32)\n",
            "epoch: 3 step: 40 train batch loss: tf.Tensor(0.75019187, shape=(), dtype=float32)\n",
            "epoch: 3 step: 41 train batch loss: tf.Tensor(0.54512507, shape=(), dtype=float32)\n",
            "epoch: 3 step: 42 train batch loss: tf.Tensor(0.56433266, shape=(), dtype=float32)\n",
            "epoch: 3 step: 43 train batch loss: tf.Tensor(0.5887839, shape=(), dtype=float32)\n",
            "epoch: 3 step: 44 train batch loss: tf.Tensor(0.45734105, shape=(), dtype=float32)\n",
            "epoch: 3 step: 45 train batch loss: tf.Tensor(0.4549087, shape=(), dtype=float32)\n",
            "epoch: 3 step: 46 train batch loss: tf.Tensor(0.4879412, shape=(), dtype=float32)\n",
            "epoch: 3 step: 47 train batch loss: tf.Tensor(0.4417366, shape=(), dtype=float32)\n",
            "epoch: 3 step: 48 train batch loss: tf.Tensor(0.58506817, shape=(), dtype=float32)\n",
            "epoch: 3 step: 49 train batch loss: tf.Tensor(0.60672, shape=(), dtype=float32)\n",
            "epoch: 3 step: 50 train batch loss: tf.Tensor(0.5653869, shape=(), dtype=float32)\n",
            "epoch: 3 step: 51 train batch loss: tf.Tensor(0.44470614, shape=(), dtype=float32)\n",
            "epoch: 3 step: 52 train batch loss: tf.Tensor(0.5087193, shape=(), dtype=float32)\n",
            "epoch: 3 step: 53 train batch loss: tf.Tensor(0.44789037, shape=(), dtype=float32)\n",
            "epoch: 3 step: 54 train batch loss: tf.Tensor(0.54983824, shape=(), dtype=float32)\n",
            "epoch: 3 step: 55 train batch loss: tf.Tensor(0.51865864, shape=(), dtype=float32)\n",
            "epoch: 3 step: 56 train batch loss: tf.Tensor(0.470287, shape=(), dtype=float32)\n",
            "epoch: 3 step: 57 train batch loss: tf.Tensor(0.41846234, shape=(), dtype=float32)\n",
            "epoch: 3 step: 58 train batch loss: tf.Tensor(0.5269921, shape=(), dtype=float32)\n",
            "epoch: 3 step: 59 train batch loss: tf.Tensor(0.573125, shape=(), dtype=float32)\n",
            "epoch: 3 step: 60 train batch loss: tf.Tensor(0.5189194, shape=(), dtype=float32)\n",
            "epoch: 3 step: 61 train batch loss: tf.Tensor(0.5168378, shape=(), dtype=float32)\n",
            "epoch: 3 step: 62 train batch loss: tf.Tensor(0.5569058, shape=(), dtype=float32)\n",
            "epoch: 3 step: 63 train batch loss: tf.Tensor(0.4048764, shape=(), dtype=float32)\n",
            "epoch: 3 step: 64 train batch loss: tf.Tensor(0.56978524, shape=(), dtype=float32)\n",
            "epoch: 3 step: 65 train batch loss: tf.Tensor(0.5078466, shape=(), dtype=float32)\n",
            "epoch: 3 step: 66 train batch loss: tf.Tensor(0.49697772, shape=(), dtype=float32)\n",
            "epoch: 3 step: 67 train batch loss: tf.Tensor(0.41816187, shape=(), dtype=float32)\n",
            "epoch: 3 step: 68 train batch loss: tf.Tensor(0.6000987, shape=(), dtype=float32)\n",
            "epoch: 3 step: 69 train batch loss: tf.Tensor(0.58326036, shape=(), dtype=float32)\n",
            "epoch: 3 step: 70 train batch loss: tf.Tensor(0.4904171, shape=(), dtype=float32)\n",
            "epoch: 3 step: 71 train batch loss: tf.Tensor(0.5043248, shape=(), dtype=float32)\n",
            "epoch: 3 step: 72 train batch loss: tf.Tensor(0.6524386, shape=(), dtype=float32)\n",
            "epoch: 3 step: 73 train batch loss: tf.Tensor(0.3737734, shape=(), dtype=float32)\n",
            "epoch: 3 step: 74 train batch loss: tf.Tensor(0.60428035, shape=(), dtype=float32)\n",
            "epoch: 3 step: 75 train batch loss: tf.Tensor(0.46062407, shape=(), dtype=float32)\n",
            "epoch: 3 step: 76 train batch loss: tf.Tensor(0.5457726, shape=(), dtype=float32)\n",
            "epoch: 3 step: 77 train batch loss: tf.Tensor(0.5422203, shape=(), dtype=float32)\n",
            "epoch: 3 step: 78 train batch loss: tf.Tensor(0.4053441, shape=(), dtype=float32)\n",
            "epoch: 3 step: 79 train batch loss: tf.Tensor(0.44942522, shape=(), dtype=float32)\n",
            "epoch: 3 step: 80 train batch loss: tf.Tensor(0.44778055, shape=(), dtype=float32)\n",
            "epoch: 3 step: 81 train batch loss: tf.Tensor(0.5209343, shape=(), dtype=float32)\n",
            "epoch: 3 step: 82 train batch loss: tf.Tensor(0.536457, shape=(), dtype=float32)\n",
            "epoch: 3 step: 83 train batch loss: tf.Tensor(0.49614888, shape=(), dtype=float32)\n",
            "epoch: 3 step: 84 train batch loss: tf.Tensor(0.5854096, shape=(), dtype=float32)\n",
            "epoch: 3 step: 85 train batch loss: tf.Tensor(0.5431962, shape=(), dtype=float32)\n",
            "epoch: 3 step: 86 train batch loss: tf.Tensor(0.5629846, shape=(), dtype=float32)\n",
            "epoch: 3 step: 87 train batch loss: tf.Tensor(0.5032173, shape=(), dtype=float32)\n",
            "epoch: 3 step: 88 train batch loss: tf.Tensor(0.4522509, shape=(), dtype=float32)\n",
            "epoch: 3 step: 89 train batch loss: tf.Tensor(0.50505227, shape=(), dtype=float32)\n",
            "epoch: 3 step: 90 train batch loss: tf.Tensor(0.45751256, shape=(), dtype=float32)\n",
            "epoch: 3 step: 91 train batch loss: tf.Tensor(0.59859353, shape=(), dtype=float32)\n",
            "epoch: 3 step: 92 train batch loss: tf.Tensor(0.6926008, shape=(), dtype=float32)\n",
            "epoch: 3 step: 93 train batch loss: tf.Tensor(0.44706148, shape=(), dtype=float32)\n",
            "epoch: 3 step: 94 train batch loss: tf.Tensor(0.49395186, shape=(), dtype=float32)\n",
            "epoch: 3 step: 95 train batch loss: tf.Tensor(0.45474023, shape=(), dtype=float32)\n",
            "epoch: 3 step: 96 train batch loss: tf.Tensor(0.47409835, shape=(), dtype=float32)\n",
            "epoch: 3 step: 97 train batch loss: tf.Tensor(0.4465476, shape=(), dtype=float32)\n",
            "epoch: 3 step: 98 train batch loss: tf.Tensor(0.40973884, shape=(), dtype=float32)\n",
            "epoch: 3 step: 99 train batch loss: tf.Tensor(0.53702563, shape=(), dtype=float32)\n",
            "epoch: 3 step: 100 train batch loss: tf.Tensor(0.5119935, shape=(), dtype=float32)\n",
            "epoch: 3 step: 101 train batch loss: tf.Tensor(0.49799198, shape=(), dtype=float32)\n",
            "epoch: 3 step: 102 train batch loss: tf.Tensor(0.58408356, shape=(), dtype=float32)\n",
            "epoch: 3 step: 103 train batch loss: tf.Tensor(0.47554198, shape=(), dtype=float32)\n",
            "epoch: 3 step: 104 train batch loss: tf.Tensor(0.5435089, shape=(), dtype=float32)\n",
            "epoch: 3 step: 105 train batch loss: tf.Tensor(0.47108585, shape=(), dtype=float32)\n",
            "epoch: 3 step: 106 train batch loss: tf.Tensor(0.5033586, shape=(), dtype=float32)\n",
            "epoch: 3 step: 107 train batch loss: tf.Tensor(0.582403, shape=(), dtype=float32)\n",
            "epoch: 3 step: 108 train batch loss: tf.Tensor(0.4401346, shape=(), dtype=float32)\n",
            "epoch: 3 step: 109 train batch loss: tf.Tensor(0.5216061, shape=(), dtype=float32)\n",
            "epoch: 3 step: 110 train batch loss: tf.Tensor(0.51850873, shape=(), dtype=float32)\n",
            "epoch: 3 step: 111 train batch loss: tf.Tensor(0.5001962, shape=(), dtype=float32)\n",
            "epoch: 3 step: 112 train batch loss: tf.Tensor(0.55629456, shape=(), dtype=float32)\n",
            "epoch: 3 step: 113 train batch loss: tf.Tensor(0.45098191, shape=(), dtype=float32)\n",
            "epoch: 3 step: 114 train batch loss: tf.Tensor(0.44219095, shape=(), dtype=float32)\n",
            "epoch: 3 step: 115 train batch loss: tf.Tensor(0.47922224, shape=(), dtype=float32)\n",
            "epoch: 3 step: 116 train batch loss: tf.Tensor(0.44374293, shape=(), dtype=float32)\n",
            "epoch: 3 step: 117 train batch loss: tf.Tensor(0.47250852, shape=(), dtype=float32)\n",
            "epoch: 3 step: 118 train batch loss: tf.Tensor(0.60483015, shape=(), dtype=float32)\n",
            "epoch: 3 step: 119 train batch loss: tf.Tensor(0.4642393, shape=(), dtype=float32)\n",
            "epoch: 3 step: 120 train batch loss: tf.Tensor(0.43239766, shape=(), dtype=float32)\n",
            "epoch: 3 step: 121 train batch loss: tf.Tensor(0.47575182, shape=(), dtype=float32)\n",
            "epoch: 3 step: 122 train batch loss: tf.Tensor(0.42527217, shape=(), dtype=float32)\n",
            "epoch: 3 step: 123 train batch loss: tf.Tensor(0.52552485, shape=(), dtype=float32)\n",
            "epoch: 3 step: 124 train batch loss: tf.Tensor(0.55690944, shape=(), dtype=float32)\n",
            "epoch: 3 step: 125 train batch loss: tf.Tensor(0.4688178, shape=(), dtype=float32)\n",
            "epoch: 3 step: 126 train batch loss: tf.Tensor(0.48929492, shape=(), dtype=float32)\n",
            "epoch: 3 step: 127 train batch loss: tf.Tensor(0.71085626, shape=(), dtype=float32)\n",
            "epoch: 3 step: 128 train batch loss: tf.Tensor(0.5163568, shape=(), dtype=float32)\n",
            "epoch: 3 step: 129 train batch loss: tf.Tensor(0.4588608, shape=(), dtype=float32)\n",
            "epoch: 3 step: 130 train batch loss: tf.Tensor(0.5030814, shape=(), dtype=float32)\n",
            "epoch: 3 step: 131 train batch loss: tf.Tensor(0.51796067, shape=(), dtype=float32)\n",
            "epoch: 3 step: 132 train batch loss: tf.Tensor(0.4593048, shape=(), dtype=float32)\n",
            "epoch: 3 step: 133 train batch loss: tf.Tensor(0.39795753, shape=(), dtype=float32)\n",
            "epoch: 3 step: 134 train batch loss: tf.Tensor(0.44288832, shape=(), dtype=float32)\n",
            "epoch: 3 step: 135 train batch loss: tf.Tensor(0.52214783, shape=(), dtype=float32)\n",
            "epoch: 3 step: 136 train batch loss: tf.Tensor(0.598621, shape=(), dtype=float32)\n",
            "epoch: 3 step: 137 train batch loss: tf.Tensor(0.49293002, shape=(), dtype=float32)\n",
            "epoch: 3 step: 138 train batch loss: tf.Tensor(0.500685, shape=(), dtype=float32)\n",
            "epoch: 3 step: 139 train batch loss: tf.Tensor(0.3874698, shape=(), dtype=float32)\n",
            "epoch: 3 step: 140 train batch loss: tf.Tensor(0.46144786, shape=(), dtype=float32)\n",
            "epoch: 3 step: 141 train batch loss: tf.Tensor(0.5202449, shape=(), dtype=float32)\n",
            "epoch: 3 step: 142 train batch loss: tf.Tensor(0.51311487, shape=(), dtype=float32)\n",
            "epoch: 3 step: 143 train batch loss: tf.Tensor(0.6098594, shape=(), dtype=float32)\n",
            "epoch: 3 step: 144 train batch loss: tf.Tensor(0.4846801, shape=(), dtype=float32)\n",
            "epoch: 3 step: 145 train batch loss: tf.Tensor(0.53334814, shape=(), dtype=float32)\n",
            "epoch: 3 step: 146 train batch loss: tf.Tensor(0.45283636, shape=(), dtype=float32)\n",
            "epoch: 3 step: 147 train batch loss: tf.Tensor(0.484554, shape=(), dtype=float32)\n",
            "epoch: 3 step: 148 train batch loss: tf.Tensor(0.4248581, shape=(), dtype=float32)\n",
            "epoch: 3 step: 149 train batch loss: tf.Tensor(0.49560246, shape=(), dtype=float32)\n",
            "epoch: 3 step: 150 train batch loss: tf.Tensor(0.42571107, shape=(), dtype=float32)\n",
            "epoch: 3 step: 151 train batch loss: tf.Tensor(0.45305306, shape=(), dtype=float32)\n",
            "epoch: 3 step: 152 train batch loss: tf.Tensor(0.474029, shape=(), dtype=float32)\n",
            "epoch: 3 step: 153 train batch loss: tf.Tensor(0.41019553, shape=(), dtype=float32)\n",
            "epoch: 3 step: 154 train batch loss: tf.Tensor(0.510149, shape=(), dtype=float32)\n",
            "epoch: 3 step: 155 train batch loss: tf.Tensor(0.45602036, shape=(), dtype=float32)\n",
            "epoch: 3 step: 156 train batch loss: tf.Tensor(0.39777684, shape=(), dtype=float32)\n",
            "epoch: 3 step: 157 train batch loss: tf.Tensor(0.44119558, shape=(), dtype=float32)\n",
            "epoch: 3 step: 158 train batch loss: tf.Tensor(0.53451604, shape=(), dtype=float32)\n",
            "epoch: 3 step: 159 train batch loss: tf.Tensor(0.5251339, shape=(), dtype=float32)\n",
            "epoch: 3 step: 160 train batch loss: tf.Tensor(0.5880864, shape=(), dtype=float32)\n",
            "epoch: 3 step: 161 train batch loss: tf.Tensor(0.4747587, shape=(), dtype=float32)\n",
            "epoch: 3 step: 162 train batch loss: tf.Tensor(0.50841415, shape=(), dtype=float32)\n",
            "epoch: 3 step: 163 train batch loss: tf.Tensor(0.50258994, shape=(), dtype=float32)\n",
            "epoch: 3 step: 164 train batch loss: tf.Tensor(0.49997887, shape=(), dtype=float32)\n",
            "epoch: 3 step: 165 train batch loss: tf.Tensor(0.530257, shape=(), dtype=float32)\n",
            "epoch: 3 step: 166 train batch loss: tf.Tensor(0.47073784, shape=(), dtype=float32)\n",
            "epoch: 3 step: 167 train batch loss: tf.Tensor(0.4230463, shape=(), dtype=float32)\n",
            "epoch: 3 step: 168 train batch loss: tf.Tensor(0.4752131, shape=(), dtype=float32)\n",
            "epoch: 3 step: 169 train batch loss: tf.Tensor(0.45423737, shape=(), dtype=float32)\n",
            "epoch: 3 step: 170 train batch loss: tf.Tensor(0.52570885, shape=(), dtype=float32)\n",
            "epoch: 3 step: 171 train batch loss: tf.Tensor(0.4697873, shape=(), dtype=float32)\n",
            "epoch: 3 step: 172 train batch loss: tf.Tensor(0.45203218, shape=(), dtype=float32)\n",
            "epoch: 3 step: 173 train batch loss: tf.Tensor(0.46166164, shape=(), dtype=float32)\n",
            "epoch: 3 step: 174 train batch loss: tf.Tensor(0.39518523, shape=(), dtype=float32)\n",
            "epoch: 3 step: 175 train batch loss: tf.Tensor(0.59434795, shape=(), dtype=float32)\n",
            "epoch: 3 step: 176 train batch loss: tf.Tensor(0.5017802, shape=(), dtype=float32)\n",
            "epoch: 3 step: 177 train batch loss: tf.Tensor(0.5559452, shape=(), dtype=float32)\n",
            "epoch: 3 step: 178 train batch loss: tf.Tensor(0.45019078, shape=(), dtype=float32)\n",
            "epoch: 3 step: 179 train batch loss: tf.Tensor(0.4601562, shape=(), dtype=float32)\n",
            "epoch: 3 step: 180 train batch loss: tf.Tensor(0.46041068, shape=(), dtype=float32)\n",
            "epoch: 3 step: 181 train batch loss: tf.Tensor(0.45987254, shape=(), dtype=float32)\n",
            "epoch: 3 step: 182 train batch loss: tf.Tensor(0.4047285, shape=(), dtype=float32)\n",
            "epoch: 3 step: 183 train batch loss: tf.Tensor(0.3879098, shape=(), dtype=float32)\n",
            "epoch: 3 step: 184 train batch loss: tf.Tensor(0.4225142, shape=(), dtype=float32)\n",
            "epoch: 3 step: 185 train batch loss: tf.Tensor(0.54479706, shape=(), dtype=float32)\n",
            "epoch: 3 step: 186 train batch loss: tf.Tensor(0.4689662, shape=(), dtype=float32)\n",
            "epoch: 3 step: 187 train batch loss: tf.Tensor(0.47380525, shape=(), dtype=float32)\n",
            "epoch: 3 step: 188 train batch loss: tf.Tensor(0.48407245, shape=(), dtype=float32)\n",
            "epoch: 3 step: 189 train batch loss: tf.Tensor(0.48061028, shape=(), dtype=float32)\n",
            "epoch: 3 step: 190 train batch loss: tf.Tensor(0.5389923, shape=(), dtype=float32)\n",
            "epoch: 3 step: 191 train batch loss: tf.Tensor(0.40508085, shape=(), dtype=float32)\n",
            "epoch: 3 step: 192 train batch loss: tf.Tensor(0.50387776, shape=(), dtype=float32)\n",
            "epoch: 3 step: 193 train batch loss: tf.Tensor(0.49735317, shape=(), dtype=float32)\n",
            "epoch: 3 step: 194 train batch loss: tf.Tensor(0.3827663, shape=(), dtype=float32)\n",
            "epoch: 3 step: 195 train batch loss: tf.Tensor(0.5756385, shape=(), dtype=float32)\n",
            "epoch: 3 step: 196 train batch loss: tf.Tensor(0.4570704, shape=(), dtype=float32)\n",
            "epoch: 3 step: 197 train batch loss: tf.Tensor(0.5081892, shape=(), dtype=float32)\n",
            "epoch: 3 step: 198 train batch loss: tf.Tensor(0.50672996, shape=(), dtype=float32)\n",
            "epoch: 3 step: 199 train batch loss: tf.Tensor(0.4023448, shape=(), dtype=float32)\n",
            "epoch: 3 step: 200 train batch loss: tf.Tensor(0.46214595, shape=(), dtype=float32)\n",
            "epoch: 3 step: 201 train batch loss: tf.Tensor(0.45624894, shape=(), dtype=float32)\n",
            "epoch: 3 step: 202 train batch loss: tf.Tensor(0.7947491, shape=(), dtype=float32)\n",
            "epoch: 3 step: 203 train batch loss: tf.Tensor(0.43578595, shape=(), dtype=float32)\n",
            "epoch: 3 step: 204 train batch loss: tf.Tensor(0.52094114, shape=(), dtype=float32)\n",
            "epoch: 3 step: 205 train batch loss: tf.Tensor(0.46977192, shape=(), dtype=float32)\n",
            "epoch: 3 step: 206 train batch loss: tf.Tensor(0.54024035, shape=(), dtype=float32)\n",
            "epoch: 3 step: 207 train batch loss: tf.Tensor(0.4768576, shape=(), dtype=float32)\n",
            "epoch: 3 step: 208 train batch loss: tf.Tensor(0.48421612, shape=(), dtype=float32)\n",
            "epoch: 3 step: 209 train batch loss: tf.Tensor(0.5811458, shape=(), dtype=float32)\n",
            "epoch: 3 step: 210 train batch loss: tf.Tensor(0.5157964, shape=(), dtype=float32)\n",
            "epoch: 3 step: 211 train batch loss: tf.Tensor(0.45339674, shape=(), dtype=float32)\n",
            "epoch: 3 step: 212 train batch loss: tf.Tensor(0.39978975, shape=(), dtype=float32)\n",
            "epoch: 3 step: 213 train batch loss: tf.Tensor(0.45051512, shape=(), dtype=float32)\n",
            "epoch: 3 step: 214 train batch loss: tf.Tensor(0.49250323, shape=(), dtype=float32)\n",
            "epoch: 3 step: 215 train batch loss: tf.Tensor(0.3782321, shape=(), dtype=float32)\n",
            "epoch: 3 step: 216 train batch loss: tf.Tensor(0.71506864, shape=(), dtype=float32)\n",
            "epoch: 3 step: 217 train batch loss: tf.Tensor(0.41314068, shape=(), dtype=float32)\n",
            "epoch: 3 step: 218 train batch loss: tf.Tensor(0.44354373, shape=(), dtype=float32)\n",
            "epoch: 3 step: 219 train batch loss: tf.Tensor(0.42582417, shape=(), dtype=float32)\n",
            "epoch: 3 step: 220 train batch loss: tf.Tensor(0.43736005, shape=(), dtype=float32)\n",
            "epoch: 3 step: 221 train batch loss: tf.Tensor(0.42914933, shape=(), dtype=float32)\n",
            "epoch: 3 step: 222 train batch loss: tf.Tensor(0.44307113, shape=(), dtype=float32)\n",
            "epoch: 3 step: 223 train batch loss: tf.Tensor(0.48201093, shape=(), dtype=float32)\n",
            "epoch: 3 step: 224 train batch loss: tf.Tensor(0.5531759, shape=(), dtype=float32)\n",
            "epoch: 3 step: 225 train batch loss: tf.Tensor(0.4256969, shape=(), dtype=float32)\n",
            "epoch: 3 step: 226 train batch loss: tf.Tensor(0.37219495, shape=(), dtype=float32)\n",
            "epoch: 3 step: 227 train batch loss: tf.Tensor(0.45516118, shape=(), dtype=float32)\n",
            "epoch: 3 step: 228 train batch loss: tf.Tensor(0.47670907, shape=(), dtype=float32)\n",
            "epoch: 3 step: 229 train batch loss: tf.Tensor(0.48594752, shape=(), dtype=float32)\n",
            "epoch: 3 step: 230 train batch loss: tf.Tensor(0.37698287, shape=(), dtype=float32)\n",
            "epoch: 3 step: 231 train batch loss: tf.Tensor(0.39984196, shape=(), dtype=float32)\n",
            "epoch: 3 step: 232 train batch loss: tf.Tensor(0.54830587, shape=(), dtype=float32)\n",
            "epoch: 3 step: 233 train batch loss: tf.Tensor(0.62646, shape=(), dtype=float32)\n",
            "epoch: 3 step: 234 train batch loss: tf.Tensor(0.48699585, shape=(), dtype=float32)\n",
            "epoch: 3 step: 235 train batch loss: tf.Tensor(0.4245428, shape=(), dtype=float32)\n",
            "epoch: 3 step: 236 train batch loss: tf.Tensor(0.5496909, shape=(), dtype=float32)\n",
            "epoch: 3 step: 237 train batch loss: tf.Tensor(0.54852766, shape=(), dtype=float32)\n",
            "epoch: 3 step: 238 train batch loss: tf.Tensor(0.53196657, shape=(), dtype=float32)\n",
            "epoch: 3 step: 239 train batch loss: tf.Tensor(0.48028564, shape=(), dtype=float32)\n",
            "epoch: 3 step: 240 train batch loss: tf.Tensor(0.529629, shape=(), dtype=float32)\n",
            "epoch: 3 step: 241 train batch loss: tf.Tensor(0.5897455, shape=(), dtype=float32)\n",
            "epoch: 3 step: 242 train batch loss: tf.Tensor(0.4360568, shape=(), dtype=float32)\n",
            "epoch: 3 step: 243 train batch loss: tf.Tensor(0.39201605, shape=(), dtype=float32)\n",
            "epoch: 3 step: 244 train batch loss: tf.Tensor(0.5444006, shape=(), dtype=float32)\n",
            "epoch: 3 step: 245 train batch loss: tf.Tensor(0.4546002, shape=(), dtype=float32)\n",
            "epoch: 3 step: 246 train batch loss: tf.Tensor(0.3751694, shape=(), dtype=float32)\n",
            "epoch: 3 step: 247 train batch loss: tf.Tensor(0.4982012, shape=(), dtype=float32)\n",
            "epoch: 3 step: 248 train batch loss: tf.Tensor(0.52797663, shape=(), dtype=float32)\n",
            "epoch: 3 step: 249 train batch loss: tf.Tensor(0.4134244, shape=(), dtype=float32)\n",
            "epoch: 3 step: 250 train batch loss: tf.Tensor(0.49381837, shape=(), dtype=float32)\n",
            "epoch: 3 step: 251 train batch loss: tf.Tensor(0.5024992, shape=(), dtype=float32)\n",
            "epoch: 3 step: 252 train batch loss: tf.Tensor(0.6346526, shape=(), dtype=float32)\n",
            "epoch: 3 step: 253 train batch loss: tf.Tensor(0.38617116, shape=(), dtype=float32)\n",
            "epoch: 3 step: 254 train batch loss: tf.Tensor(0.4694026, shape=(), dtype=float32)\n",
            "epoch: 3 step: 255 train batch loss: tf.Tensor(0.54230624, shape=(), dtype=float32)\n",
            "epoch: 3 step: 256 train batch loss: tf.Tensor(0.423518, shape=(), dtype=float32)\n",
            "epoch: 3 step: 257 train batch loss: tf.Tensor(0.4445396, shape=(), dtype=float32)\n",
            "epoch: 3 step: 258 train batch loss: tf.Tensor(0.39295626, shape=(), dtype=float32)\n",
            "epoch: 3 step: 259 train batch loss: tf.Tensor(0.434272, shape=(), dtype=float32)\n",
            "epoch: 3 step: 260 train batch loss: tf.Tensor(0.51226676, shape=(), dtype=float32)\n",
            "epoch: 3 step: 261 train batch loss: tf.Tensor(0.46925527, shape=(), dtype=float32)\n",
            "epoch: 3 step: 262 train batch loss: tf.Tensor(0.5855699, shape=(), dtype=float32)\n",
            "epoch: 3 step: 263 train batch loss: tf.Tensor(0.5082864, shape=(), dtype=float32)\n",
            "epoch: 3 step: 264 train batch loss: tf.Tensor(0.52863395, shape=(), dtype=float32)\n",
            "epoch: 3 step: 265 train batch loss: tf.Tensor(0.4738554, shape=(), dtype=float32)\n",
            "epoch: 3 step: 266 train batch loss: tf.Tensor(0.51320064, shape=(), dtype=float32)\n",
            "epoch: 3 step: 267 train batch loss: tf.Tensor(0.40737525, shape=(), dtype=float32)\n",
            "epoch: 3 step: 268 train batch loss: tf.Tensor(0.40288442, shape=(), dtype=float32)\n",
            "epoch: 3 step: 269 train batch loss: tf.Tensor(0.4376115, shape=(), dtype=float32)\n",
            "epoch: 3 step: 270 train batch loss: tf.Tensor(0.46658644, shape=(), dtype=float32)\n",
            "epoch: 3 step: 271 train batch loss: tf.Tensor(0.54456913, shape=(), dtype=float32)\n",
            "epoch: 3 step: 272 train batch loss: tf.Tensor(0.42409903, shape=(), dtype=float32)\n",
            "epoch: 3 step: 273 train batch loss: tf.Tensor(0.45863065, shape=(), dtype=float32)\n",
            "epoch: 3 step: 274 train batch loss: tf.Tensor(0.73605305, shape=(), dtype=float32)\n",
            "epoch: 3 step: 275 train batch loss: tf.Tensor(0.45040798, shape=(), dtype=float32)\n",
            "epoch: 3 step: 276 train batch loss: tf.Tensor(0.65885645, shape=(), dtype=float32)\n",
            "epoch: 3 step: 277 train batch loss: tf.Tensor(0.548777, shape=(), dtype=float32)\n",
            "epoch: 3 step: 278 train batch loss: tf.Tensor(0.59321934, shape=(), dtype=float32)\n",
            "epoch: 3 step: 279 train batch loss: tf.Tensor(0.44986838, shape=(), dtype=float32)\n",
            "epoch: 3 step: 280 train batch loss: tf.Tensor(0.441011, shape=(), dtype=float32)\n",
            "epoch: 3 step: 281 train batch loss: tf.Tensor(0.5371507, shape=(), dtype=float32)\n",
            "epoch: 3 step: 282 train batch loss: tf.Tensor(0.4395178, shape=(), dtype=float32)\n",
            "epoch: 3 step: 283 train batch loss: tf.Tensor(0.4907751, shape=(), dtype=float32)\n",
            "epoch: 3 step: 284 train batch loss: tf.Tensor(0.4312355, shape=(), dtype=float32)\n",
            "epoch: 3 step: 285 train batch loss: tf.Tensor(0.4475049, shape=(), dtype=float32)\n",
            "epoch: 3 step: 286 train batch loss: tf.Tensor(0.40310436, shape=(), dtype=float32)\n",
            "epoch: 3 step: 287 train batch loss: tf.Tensor(0.45516434, shape=(), dtype=float32)\n",
            "epoch: 3 step: 288 train batch loss: tf.Tensor(0.45380554, shape=(), dtype=float32)\n",
            "epoch: 3 step: 289 train batch loss: tf.Tensor(0.38463074, shape=(), dtype=float32)\n",
            "epoch: 3 step: 290 train batch loss: tf.Tensor(0.5342712, shape=(), dtype=float32)\n",
            "epoch: 3 step: 291 train batch loss: tf.Tensor(0.5942522, shape=(), dtype=float32)\n",
            "epoch: 3 step: 292 train batch loss: tf.Tensor(0.5987354, shape=(), dtype=float32)\n",
            "epoch: 3 step: 293 train batch loss: tf.Tensor(0.59002894, shape=(), dtype=float32)\n",
            "epoch: 3 step: 294 train batch loss: tf.Tensor(0.45862642, shape=(), dtype=float32)\n",
            "epoch: 3 step: 295 train batch loss: tf.Tensor(0.5200349, shape=(), dtype=float32)\n",
            "epoch: 3 step: 296 train batch loss: tf.Tensor(0.4115165, shape=(), dtype=float32)\n",
            "epoch: 3 step: 297 train batch loss: tf.Tensor(0.5034921, shape=(), dtype=float32)\n",
            "epoch: 3 step: 298 train batch loss: tf.Tensor(0.45001826, shape=(), dtype=float32)\n",
            "epoch: 3 step: 299 train batch loss: tf.Tensor(0.45364076, shape=(), dtype=float32)\n",
            "epoch: 3 step: 300 train batch loss: tf.Tensor(0.5186218, shape=(), dtype=float32)\n",
            "epoch: 3 step: 301 train batch loss: tf.Tensor(0.4309701, shape=(), dtype=float32)\n",
            "epoch: 3 step: 302 train batch loss: tf.Tensor(0.3752218, shape=(), dtype=float32)\n",
            "epoch: 3 step: 303 train batch loss: tf.Tensor(0.43663755, shape=(), dtype=float32)\n",
            "epoch: 3 step: 304 train batch loss: tf.Tensor(0.5033562, shape=(), dtype=float32)\n",
            "epoch: 3 step: 305 train batch loss: tf.Tensor(0.49261218, shape=(), dtype=float32)\n",
            "epoch: 3 step: 306 train batch loss: tf.Tensor(0.42424384, shape=(), dtype=float32)\n",
            "epoch: 3 step: 307 train batch loss: tf.Tensor(0.43976235, shape=(), dtype=float32)\n",
            "epoch: 3 step: 308 train batch loss: tf.Tensor(0.45136318, shape=(), dtype=float32)\n",
            "epoch: 3 step: 309 train batch loss: tf.Tensor(0.46873188, shape=(), dtype=float32)\n",
            "epoch: 3 step: 310 train batch loss: tf.Tensor(0.54566836, shape=(), dtype=float32)\n",
            "epoch: 3 step: 311 train batch loss: tf.Tensor(0.53544456, shape=(), dtype=float32)\n",
            "epoch: 3 step: 312 train batch loss: tf.Tensor(0.53897643, shape=(), dtype=float32)\n",
            "epoch: 3 step: 313 train batch loss: tf.Tensor(0.43666118, shape=(), dtype=float32)\n",
            "epoch: 3 step: 314 train batch loss: tf.Tensor(0.49410945, shape=(), dtype=float32)\n",
            "epoch: 3 step: 315 train batch loss: tf.Tensor(0.40733084, shape=(), dtype=float32)\n",
            "epoch: 3 step: 316 train batch loss: tf.Tensor(0.38090724, shape=(), dtype=float32)\n",
            "epoch: 3 step: 317 train batch loss: tf.Tensor(0.46483952, shape=(), dtype=float32)\n",
            "epoch: 3 step: 318 train batch loss: tf.Tensor(0.44596085, shape=(), dtype=float32)\n",
            "epoch: 3 step: 319 train batch loss: tf.Tensor(0.4838141, shape=(), dtype=float32)\n",
            "epoch: 3 step: 320 train batch loss: tf.Tensor(0.42279476, shape=(), dtype=float32)\n",
            "epoch: 3 step: 321 train batch loss: tf.Tensor(0.4705742, shape=(), dtype=float32)\n",
            "epoch: 3 step: 322 train batch loss: tf.Tensor(0.43378744, shape=(), dtype=float32)\n",
            "epoch: 3 step: 323 train batch loss: tf.Tensor(0.48953542, shape=(), dtype=float32)\n",
            "epoch: 3 step: 324 train batch loss: tf.Tensor(0.5441792, shape=(), dtype=float32)\n",
            "epoch: 3 step: 325 train batch loss: tf.Tensor(0.5739053, shape=(), dtype=float32)\n",
            "epoch: 3 step: 326 train batch loss: tf.Tensor(0.4536311, shape=(), dtype=float32)\n",
            "epoch: 3 step: 327 train batch loss: tf.Tensor(0.60144824, shape=(), dtype=float32)\n",
            "epoch: 3 step: 328 train batch loss: tf.Tensor(0.44973797, shape=(), dtype=float32)\n",
            "epoch: 3 step: 329 train batch loss: tf.Tensor(0.46420008, shape=(), dtype=float32)\n",
            "epoch: 3 step: 330 train batch loss: tf.Tensor(0.4718964, shape=(), dtype=float32)\n",
            "epoch: 3 step: 331 train batch loss: tf.Tensor(0.40073738, shape=(), dtype=float32)\n",
            "epoch: 3 step: 332 train batch loss: tf.Tensor(0.48800763, shape=(), dtype=float32)\n",
            "epoch: 3 step: 333 train batch loss: tf.Tensor(0.61346465, shape=(), dtype=float32)\n",
            "epoch: 3 step: 334 train batch loss: tf.Tensor(0.43983108, shape=(), dtype=float32)\n",
            "epoch: 3 step: 335 train batch loss: tf.Tensor(0.41084918, shape=(), dtype=float32)\n",
            "epoch: 3 step: 336 train batch loss: tf.Tensor(0.6109446, shape=(), dtype=float32)\n",
            "epoch: 3 step: 337 train batch loss: tf.Tensor(0.39493766, shape=(), dtype=float32)\n",
            "epoch: 3 step: 338 train batch loss: tf.Tensor(0.420407, shape=(), dtype=float32)\n",
            "epoch: 3 step: 339 train batch loss: tf.Tensor(0.45491195, shape=(), dtype=float32)\n",
            "epoch: 3 step: 340 train batch loss: tf.Tensor(0.47221446, shape=(), dtype=float32)\n",
            "epoch: 3 step: 341 train batch loss: tf.Tensor(0.42523965, shape=(), dtype=float32)\n",
            "epoch: 3 step: 342 train batch loss: tf.Tensor(0.41227525, shape=(), dtype=float32)\n",
            "epoch: 3 step: 343 train batch loss: tf.Tensor(0.36872458, shape=(), dtype=float32)\n",
            "epoch: 3 step: 344 train batch loss: tf.Tensor(0.4565309, shape=(), dtype=float32)\n",
            "epoch: 3 step: 345 train batch loss: tf.Tensor(0.5272321, shape=(), dtype=float32)\n",
            "epoch: 3 step: 346 train batch loss: tf.Tensor(0.4663844, shape=(), dtype=float32)\n",
            "epoch: 3 step: 347 train batch loss: tf.Tensor(0.52035403, shape=(), dtype=float32)\n",
            "epoch: 3 step: 348 train batch loss: tf.Tensor(0.424158, shape=(), dtype=float32)\n",
            "epoch: 3 step: 349 train batch loss: tf.Tensor(0.3865381, shape=(), dtype=float32)\n",
            "epoch: 3 step: 350 train batch loss: tf.Tensor(0.4921414, shape=(), dtype=float32)\n",
            "epoch: 3 step: 351 train batch loss: tf.Tensor(0.3761047, shape=(), dtype=float32)\n",
            "epoch: 3 step: 352 train batch loss: tf.Tensor(0.43156314, shape=(), dtype=float32)\n",
            "epoch: 3 step: 353 train batch loss: tf.Tensor(0.53971636, shape=(), dtype=float32)\n",
            "epoch: 3 step: 354 train batch loss: tf.Tensor(0.406596, shape=(), dtype=float32)\n",
            "epoch: 3 step: 355 train batch loss: tf.Tensor(0.39526594, shape=(), dtype=float32)\n",
            "epoch: 3 step: 356 train batch loss: tf.Tensor(0.49109894, shape=(), dtype=float32)\n",
            "epoch: 3 step: 357 train batch loss: tf.Tensor(0.49647784, shape=(), dtype=float32)\n",
            "epoch: 3 step: 358 train batch loss: tf.Tensor(0.523226, shape=(), dtype=float32)\n",
            "epoch: 3 step: 359 train batch loss: tf.Tensor(0.44813752, shape=(), dtype=float32)\n",
            "epoch: 3 step: 360 train batch loss: tf.Tensor(0.50438976, shape=(), dtype=float32)\n",
            "epoch: 3 step: 361 train batch loss: tf.Tensor(0.47344804, shape=(), dtype=float32)\n",
            "epoch: 3 step: 362 train batch loss: tf.Tensor(0.432549, shape=(), dtype=float32)\n",
            "epoch: 3 step: 363 train batch loss: tf.Tensor(0.37837586, shape=(), dtype=float32)\n",
            "epoch: 3 step: 364 train batch loss: tf.Tensor(0.43038496, shape=(), dtype=float32)\n",
            "epoch: 3 step: 365 train batch loss: tf.Tensor(0.39679474, shape=(), dtype=float32)\n",
            "epoch: 3 step: 366 train batch loss: tf.Tensor(0.49565086, shape=(), dtype=float32)\n",
            "epoch: 3 step: 367 train batch loss: tf.Tensor(0.43785158, shape=(), dtype=float32)\n",
            "epoch: 3 step: 368 train batch loss: tf.Tensor(0.45711794, shape=(), dtype=float32)\n",
            "epoch: 3 step: 369 train batch loss: tf.Tensor(0.46807456, shape=(), dtype=float32)\n",
            "epoch: 3 step: 370 train batch loss: tf.Tensor(0.4164098, shape=(), dtype=float32)\n",
            "epoch: 3 step: 371 train batch loss: tf.Tensor(0.4026584, shape=(), dtype=float32)\n",
            "epoch: 3 step: 372 train batch loss: tf.Tensor(0.46939975, shape=(), dtype=float32)\n",
            "epoch: 3 step: 373 train batch loss: tf.Tensor(0.4714303, shape=(), dtype=float32)\n",
            "epoch: 3 step: 374 train batch loss: tf.Tensor(0.41967803, shape=(), dtype=float32)\n",
            "epoch: 3 step: 375 train batch loss: tf.Tensor(0.42562586, shape=(), dtype=float32)\n",
            "epoch: 3 step: 376 train batch loss: tf.Tensor(0.457726, shape=(), dtype=float32)\n",
            "epoch: 3 step: 377 train batch loss: tf.Tensor(0.4935701, shape=(), dtype=float32)\n",
            "epoch: 3 step: 378 train batch loss: tf.Tensor(0.48035377, shape=(), dtype=float32)\n",
            "epoch: 3 step: 379 train batch loss: tf.Tensor(0.50663555, shape=(), dtype=float32)\n",
            "epoch: 3 step: 380 train batch loss: tf.Tensor(0.41208237, shape=(), dtype=float32)\n",
            "epoch: 3 step: 381 train batch loss: tf.Tensor(0.5535999, shape=(), dtype=float32)\n",
            "epoch: 3 step: 382 train batch loss: tf.Tensor(0.42681026, shape=(), dtype=float32)\n",
            "epoch: 3 step: 383 train batch loss: tf.Tensor(0.5632262, shape=(), dtype=float32)\n",
            "epoch: 3 step: 384 train batch loss: tf.Tensor(0.41990823, shape=(), dtype=float32)\n",
            "epoch: 3 step: 385 train batch loss: tf.Tensor(0.43546215, shape=(), dtype=float32)\n",
            "epoch: 3 step: 386 train batch loss: tf.Tensor(0.490582, shape=(), dtype=float32)\n",
            "epoch: 3 step: 387 train batch loss: tf.Tensor(0.36529648, shape=(), dtype=float32)\n",
            "epoch: 3 step: 388 train batch loss: tf.Tensor(0.43310446, shape=(), dtype=float32)\n",
            "epoch: 3 step: 389 train batch loss: tf.Tensor(0.4321216, shape=(), dtype=float32)\n",
            "epoch: 3 step: 390 train batch loss: tf.Tensor(0.56573504, shape=(), dtype=float32)\n",
            "epoch: 3 step: 391 train batch loss: tf.Tensor(0.5078993, shape=(), dtype=float32)\n",
            "epoch: 3 step: 392 train batch loss: tf.Tensor(0.5139541, shape=(), dtype=float32)\n",
            "epoch: 3 step: 393 train batch loss: tf.Tensor(0.47604832, shape=(), dtype=float32)\n",
            "epoch: 3 step: 394 train batch loss: tf.Tensor(0.53129464, shape=(), dtype=float32)\n",
            "epoch: 3 step: 395 train batch loss: tf.Tensor(0.5242224, shape=(), dtype=float32)\n",
            "epoch: 3 step: 396 train batch loss: tf.Tensor(0.4775981, shape=(), dtype=float32)\n",
            "epoch: 3 step: 397 train batch loss: tf.Tensor(0.49975243, shape=(), dtype=float32)\n",
            "epoch: 3 step: 398 train batch loss: tf.Tensor(0.39538497, shape=(), dtype=float32)\n",
            "epoch: 3 step: 399 train batch loss: tf.Tensor(0.56385016, shape=(), dtype=float32)\n",
            "epoch: 3 step: 400 train batch loss: tf.Tensor(0.4549976, shape=(), dtype=float32)\n",
            "epoch: 3 step: 401 train batch loss: tf.Tensor(0.40557298, shape=(), dtype=float32)\n",
            "epoch: 3 step: 402 train batch loss: tf.Tensor(0.5555887, shape=(), dtype=float32)\n",
            "epoch: 3 step: 403 train batch loss: tf.Tensor(0.41238317, shape=(), dtype=float32)\n",
            "epoch: 3 step: 404 train batch loss: tf.Tensor(0.44947925, shape=(), dtype=float32)\n",
            "epoch: 3 step: 405 train batch loss: tf.Tensor(0.41546687, shape=(), dtype=float32)\n",
            "epoch: 3 step: 406 train batch loss: tf.Tensor(0.36985683, shape=(), dtype=float32)\n",
            "epoch: 3 step: 407 train batch loss: tf.Tensor(0.3570977, shape=(), dtype=float32)\n",
            "epoch: 3 step: 408 train batch loss: tf.Tensor(0.42282638, shape=(), dtype=float32)\n",
            "epoch: 3 step: 409 train batch loss: tf.Tensor(0.49736685, shape=(), dtype=float32)\n",
            "epoch: 3 step: 410 train batch loss: tf.Tensor(0.4602539, shape=(), dtype=float32)\n",
            "epoch: 3 step: 411 train batch loss: tf.Tensor(0.4865368, shape=(), dtype=float32)\n",
            "epoch: 3 step: 412 train batch loss: tf.Tensor(0.46284848, shape=(), dtype=float32)\n",
            "epoch: 3 step: 413 train batch loss: tf.Tensor(0.41137105, shape=(), dtype=float32)\n",
            "epoch: 3 step: 414 train batch loss: tf.Tensor(0.6086765, shape=(), dtype=float32)\n",
            "epoch: 3 step: 415 train batch loss: tf.Tensor(0.44968557, shape=(), dtype=float32)\n",
            "epoch: 3 step: 416 train batch loss: tf.Tensor(0.448541, shape=(), dtype=float32)\n",
            "epoch: 3 step: 417 train batch loss: tf.Tensor(0.47460365, shape=(), dtype=float32)\n",
            "epoch: 3 step: 418 train batch loss: tf.Tensor(0.4283362, shape=(), dtype=float32)\n",
            "epoch: 3 step: 419 train batch loss: tf.Tensor(0.4524073, shape=(), dtype=float32)\n",
            "epoch: 3 step: 420 train batch loss: tf.Tensor(0.45608586, shape=(), dtype=float32)\n",
            "epoch: 3 step: 421 train batch loss: tf.Tensor(0.42342573, shape=(), dtype=float32)\n",
            "epoch: 3 step: 422 train batch loss: tf.Tensor(0.40529054, shape=(), dtype=float32)\n",
            "epoch: 3 step: 423 train batch loss: tf.Tensor(0.41599813, shape=(), dtype=float32)\n",
            "epoch: 3 step: 424 train batch loss: tf.Tensor(0.50903016, shape=(), dtype=float32)\n",
            "epoch: 3 step: 425 train batch loss: tf.Tensor(0.44162124, shape=(), dtype=float32)\n",
            "epoch: 3 step: 426 train batch loss: tf.Tensor(0.5072802, shape=(), dtype=float32)\n",
            "epoch: 3 step: 427 train batch loss: tf.Tensor(0.38967997, shape=(), dtype=float32)\n",
            "epoch: 3 step: 428 train batch loss: tf.Tensor(0.37459904, shape=(), dtype=float32)\n",
            "epoch: 3 step: 429 train batch loss: tf.Tensor(0.3757447, shape=(), dtype=float32)\n",
            "epoch: 3 step: 430 train batch loss: tf.Tensor(0.49432078, shape=(), dtype=float32)\n",
            "epoch: 3 step: 431 train batch loss: tf.Tensor(0.46494412, shape=(), dtype=float32)\n",
            "epoch: 3 step: 432 train batch loss: tf.Tensor(0.50320774, shape=(), dtype=float32)\n",
            "epoch: 3 step: 433 train batch loss: tf.Tensor(0.52713, shape=(), dtype=float32)\n",
            "epoch: 3 step: 434 train batch loss: tf.Tensor(0.4637223, shape=(), dtype=float32)\n",
            "epoch: 3 step: 435 train batch loss: tf.Tensor(0.5373963, shape=(), dtype=float32)\n",
            "epoch: 3 step: 436 train batch loss: tf.Tensor(0.41829053, shape=(), dtype=float32)\n",
            "epoch: 3 step: 437 train batch loss: tf.Tensor(0.62086666, shape=(), dtype=float32)\n",
            "epoch: 3 step: 438 train batch loss: tf.Tensor(0.43251458, shape=(), dtype=float32)\n",
            "epoch: 3 step: 439 train batch loss: tf.Tensor(0.49240762, shape=(), dtype=float32)\n",
            "epoch: 3 step: 440 train batch loss: tf.Tensor(0.44709778, shape=(), dtype=float32)\n",
            "epoch: 3 step: 441 train batch loss: tf.Tensor(0.6010072, shape=(), dtype=float32)\n",
            "epoch: 3 step: 442 train batch loss: tf.Tensor(0.5621809, shape=(), dtype=float32)\n",
            "epoch: 3 step: 443 train batch loss: tf.Tensor(0.37527832, shape=(), dtype=float32)\n",
            "epoch: 3 step: 444 train batch loss: tf.Tensor(0.45803174, shape=(), dtype=float32)\n",
            "epoch: 3 step: 445 train batch loss: tf.Tensor(0.5452983, shape=(), dtype=float32)\n",
            "epoch: 3 step: 446 train batch loss: tf.Tensor(0.43976927, shape=(), dtype=float32)\n",
            "epoch: 3 step: 447 train batch loss: tf.Tensor(0.4898161, shape=(), dtype=float32)\n",
            "epoch: 3 step: 448 train batch loss: tf.Tensor(0.39968953, shape=(), dtype=float32)\n",
            "epoch: 3 step: 449 train batch loss: tf.Tensor(0.551712, shape=(), dtype=float32)\n",
            "epoch: 3 step: 450 train batch loss: tf.Tensor(0.42758712, shape=(), dtype=float32)\n",
            "epoch: 3 step: 451 train batch loss: tf.Tensor(0.5485245, shape=(), dtype=float32)\n",
            "epoch: 3 step: 452 train batch loss: tf.Tensor(0.46040988, shape=(), dtype=float32)\n",
            "epoch: 3 step: 453 train batch loss: tf.Tensor(0.43543768, shape=(), dtype=float32)\n",
            "epoch: 3 step: 454 train batch loss: tf.Tensor(0.44567674, shape=(), dtype=float32)\n",
            "epoch: 3 step: 455 train batch loss: tf.Tensor(0.48295194, shape=(), dtype=float32)\n",
            "epoch: 3 step: 456 train batch loss: tf.Tensor(0.3922614, shape=(), dtype=float32)\n",
            "epoch: 3 step: 457 train batch loss: tf.Tensor(0.5972504, shape=(), dtype=float32)\n",
            "epoch: 3 step: 458 train batch loss: tf.Tensor(0.4890039, shape=(), dtype=float32)\n",
            "epoch: 3 step: 459 train batch loss: tf.Tensor(0.4888622, shape=(), dtype=float32)\n",
            "epoch: 3 step: 460 train batch loss: tf.Tensor(0.40292263, shape=(), dtype=float32)\n",
            "epoch: 3 step: 461 train batch loss: tf.Tensor(0.43244624, shape=(), dtype=float32)\n",
            "epoch: 3 step: 462 train batch loss: tf.Tensor(0.4813799, shape=(), dtype=float32)\n",
            "epoch: 3 step: 463 train batch loss: tf.Tensor(0.48983806, shape=(), dtype=float32)\n",
            "epoch: 3 step: 464 train batch loss: tf.Tensor(0.45926708, shape=(), dtype=float32)\n",
            "epoch: 3 step: 465 train batch loss: tf.Tensor(0.43055636, shape=(), dtype=float32)\n",
            "epoch: 3 step: 466 train batch loss: tf.Tensor(0.454264, shape=(), dtype=float32)\n",
            "epoch: 3 step: 467 train batch loss: tf.Tensor(0.38881433, shape=(), dtype=float32)\n",
            "epoch: 3 step: 468 train batch loss: tf.Tensor(0.5123634, shape=(), dtype=float32)\n",
            "epoch: 3 step: 469 train batch loss: tf.Tensor(0.5007552, shape=(), dtype=float32)\n",
            "epoch: 3 step: 470 train batch loss: tf.Tensor(0.47950754, shape=(), dtype=float32)\n",
            "epoch: 3 step: 471 train batch loss: tf.Tensor(0.46640572, shape=(), dtype=float32)\n",
            "epoch: 3 step: 472 train batch loss: tf.Tensor(0.48877478, shape=(), dtype=float32)\n",
            "epoch: 3 step: 473 train batch loss: tf.Tensor(0.438837, shape=(), dtype=float32)\n",
            "epoch: 3 step: 474 train batch loss: tf.Tensor(0.47470784, shape=(), dtype=float32)\n",
            "epoch: 3 step: 475 train batch loss: tf.Tensor(0.44754595, shape=(), dtype=float32)\n",
            "epoch: 3 step: 476 train batch loss: tf.Tensor(0.44151616, shape=(), dtype=float32)\n",
            "epoch: 3 step: 477 train batch loss: tf.Tensor(0.43893456, shape=(), dtype=float32)\n",
            "epoch: 3 step: 478 train batch loss: tf.Tensor(0.45805937, shape=(), dtype=float32)\n",
            "epoch: 3 step: 479 train batch loss: tf.Tensor(0.474229, shape=(), dtype=float32)\n",
            "epoch: 3 step: 480 train batch loss: tf.Tensor(0.58488214, shape=(), dtype=float32)\n",
            "epoch: 3 step: 481 train batch loss: tf.Tensor(0.35069242, shape=(), dtype=float32)\n",
            "epoch: 3 step: 482 train batch loss: tf.Tensor(0.45804533, shape=(), dtype=float32)\n",
            "epoch: 3 step: 483 train batch loss: tf.Tensor(0.5006994, shape=(), dtype=float32)\n",
            "epoch: 3 step: 484 train batch loss: tf.Tensor(0.38402778, shape=(), dtype=float32)\n",
            "epoch: 3 step: 485 train batch loss: tf.Tensor(0.5261429, shape=(), dtype=float32)\n",
            "epoch: 3 step: 486 train batch loss: tf.Tensor(0.4155349, shape=(), dtype=float32)\n",
            "epoch: 3 step: 487 train batch loss: tf.Tensor(0.40378046, shape=(), dtype=float32)\n",
            "epoch: 3 step: 488 train batch loss: tf.Tensor(0.46930653, shape=(), dtype=float32)\n",
            "epoch: 3 step: 489 train batch loss: tf.Tensor(0.43304962, shape=(), dtype=float32)\n",
            "epoch: 3 step: 490 train batch loss: tf.Tensor(0.41521508, shape=(), dtype=float32)\n",
            "epoch: 3 step: 491 train batch loss: tf.Tensor(0.44305134, shape=(), dtype=float32)\n",
            "epoch: 3 step: 492 train batch loss: tf.Tensor(0.59335774, shape=(), dtype=float32)\n",
            "epoch: 3 step: 493 train batch loss: tf.Tensor(0.4208594, shape=(), dtype=float32)\n",
            "epoch: 3 step: 494 train batch loss: tf.Tensor(0.42027774, shape=(), dtype=float32)\n",
            "epoch: 3 step: 495 train batch loss: tf.Tensor(0.33852357, shape=(), dtype=float32)\n",
            "epoch: 3 step: 496 train batch loss: tf.Tensor(0.51353204, shape=(), dtype=float32)\n",
            "epoch: 3 step: 497 train batch loss: tf.Tensor(0.5062922, shape=(), dtype=float32)\n",
            "epoch: 3 step: 498 train batch loss: tf.Tensor(0.46502858, shape=(), dtype=float32)\n",
            "epoch: 3 step: 499 train batch loss: tf.Tensor(0.43878135, shape=(), dtype=float32)\n",
            "epoch: 3 step: 500 train batch loss: tf.Tensor(0.53329057, shape=(), dtype=float32)\n",
            "epoch: 3 step: 501 train batch loss: tf.Tensor(0.38924685, shape=(), dtype=float32)\n",
            "epoch: 3 step: 502 train batch loss: tf.Tensor(0.49920917, shape=(), dtype=float32)\n",
            "epoch: 3 step: 503 train batch loss: tf.Tensor(0.40682268, shape=(), dtype=float32)\n",
            "epoch: 3 step: 504 train batch loss: tf.Tensor(0.41224355, shape=(), dtype=float32)\n",
            "epoch: 3 step: 505 train batch loss: tf.Tensor(0.42267513, shape=(), dtype=float32)\n",
            "epoch: 3 step: 506 train batch loss: tf.Tensor(0.4539371, shape=(), dtype=float32)\n",
            "epoch: 3 step: 507 train batch loss: tf.Tensor(0.32915413, shape=(), dtype=float32)\n",
            "epoch: 3 step: 508 train batch loss: tf.Tensor(0.49879703, shape=(), dtype=float32)\n",
            "epoch: 3 step: 509 train batch loss: tf.Tensor(0.47322226, shape=(), dtype=float32)\n",
            "epoch: 3 step: 510 train batch loss: tf.Tensor(0.43589002, shape=(), dtype=float32)\n",
            "epoch: 3 step: 511 train batch loss: tf.Tensor(0.4841543, shape=(), dtype=float32)\n",
            "epoch: 3 step: 512 train batch loss: tf.Tensor(0.46023178, shape=(), dtype=float32)\n",
            "epoch: 3 step: 513 train batch loss: tf.Tensor(0.5368289, shape=(), dtype=float32)\n",
            "epoch: 3 step: 514 train batch loss: tf.Tensor(0.50922954, shape=(), dtype=float32)\n",
            "epoch: 3 step: 515 train batch loss: tf.Tensor(0.44410467, shape=(), dtype=float32)\n",
            "epoch: 3 step: 516 train batch loss: tf.Tensor(0.33893588, shape=(), dtype=float32)\n",
            "epoch: 3 step: 517 train batch loss: tf.Tensor(0.35508892, shape=(), dtype=float32)\n",
            "epoch: 3 step: 518 train batch loss: tf.Tensor(0.5021258, shape=(), dtype=float32)\n",
            "epoch: 3 step: 519 train batch loss: tf.Tensor(0.41284308, shape=(), dtype=float32)\n",
            "epoch: 3 step: 520 train batch loss: tf.Tensor(0.4698299, shape=(), dtype=float32)\n",
            "epoch: 3 step: 521 train batch loss: tf.Tensor(0.43868795, shape=(), dtype=float32)\n",
            "epoch: 3 step: 522 train batch loss: tf.Tensor(0.50526893, shape=(), dtype=float32)\n",
            "epoch: 3 step: 523 train batch loss: tf.Tensor(0.40903664, shape=(), dtype=float32)\n",
            "epoch: 3 step: 524 train batch loss: tf.Tensor(0.5614254, shape=(), dtype=float32)\n",
            "epoch: 3 step: 525 train batch loss: tf.Tensor(0.4103529, shape=(), dtype=float32)\n",
            "epoch: 3 step: 526 train batch loss: tf.Tensor(0.4804706, shape=(), dtype=float32)\n",
            "epoch: 3 step: 527 train batch loss: tf.Tensor(0.37710547, shape=(), dtype=float32)\n",
            "epoch: 3 step: 528 train batch loss: tf.Tensor(0.37529674, shape=(), dtype=float32)\n",
            "epoch: 3 step: 529 train batch loss: tf.Tensor(0.43285698, shape=(), dtype=float32)\n",
            "epoch: 3 step: 530 train batch loss: tf.Tensor(0.44868922, shape=(), dtype=float32)\n",
            "epoch: 3 step: 531 train batch loss: tf.Tensor(0.49604782, shape=(), dtype=float32)\n",
            "epoch: 3 step: 532 train batch loss: tf.Tensor(0.4179462, shape=(), dtype=float32)\n",
            "epoch: 3 step: 533 train batch loss: tf.Tensor(0.52622944, shape=(), dtype=float32)\n",
            "epoch: 3 step: 534 train batch loss: tf.Tensor(0.41010898, shape=(), dtype=float32)\n",
            "epoch: 3 step: 535 train batch loss: tf.Tensor(0.3331044, shape=(), dtype=float32)\n",
            "epoch: 3 step: 536 train batch loss: tf.Tensor(0.38488308, shape=(), dtype=float32)\n",
            "epoch: 3 step: 537 train batch loss: tf.Tensor(0.46183246, shape=(), dtype=float32)\n",
            "epoch: 3 step: 538 train batch loss: tf.Tensor(0.38163495, shape=(), dtype=float32)\n",
            "epoch: 3 step: 539 train batch loss: tf.Tensor(0.49729267, shape=(), dtype=float32)\n",
            "epoch: 3 step: 540 train batch loss: tf.Tensor(0.4894247, shape=(), dtype=float32)\n",
            "epoch: 3 step: 541 train batch loss: tf.Tensor(0.39843026, shape=(), dtype=float32)\n",
            "epoch: 3 step: 542 train batch loss: tf.Tensor(0.481779, shape=(), dtype=float32)\n",
            "epoch: 3 step: 543 train batch loss: tf.Tensor(0.60457593, shape=(), dtype=float32)\n",
            "epoch: 3 step: 544 train batch loss: tf.Tensor(0.4761599, shape=(), dtype=float32)\n",
            "epoch: 3 step: 545 train batch loss: tf.Tensor(0.44889894, shape=(), dtype=float32)\n",
            "epoch: 3 step: 546 train batch loss: tf.Tensor(0.37312153, shape=(), dtype=float32)\n",
            "epoch: 3 step: 547 train batch loss: tf.Tensor(0.36902094, shape=(), dtype=float32)\n",
            "epoch: 3 step: 548 train batch loss: tf.Tensor(0.47937328, shape=(), dtype=float32)\n",
            "epoch: 3 step: 549 train batch loss: tf.Tensor(0.41905785, shape=(), dtype=float32)\n",
            "epoch: 3 step: 550 train batch loss: tf.Tensor(0.41389155, shape=(), dtype=float32)\n",
            "epoch: 3 step: 551 train batch loss: tf.Tensor(0.45309958, shape=(), dtype=float32)\n",
            "epoch: 3 step: 552 train batch loss: tf.Tensor(0.42864427, shape=(), dtype=float32)\n",
            "epoch: 3 step: 553 train batch loss: tf.Tensor(0.51474637, shape=(), dtype=float32)\n",
            "epoch: 3 step: 554 train batch loss: tf.Tensor(0.586482, shape=(), dtype=float32)\n",
            "epoch: 3 step: 555 train batch loss: tf.Tensor(0.42030078, shape=(), dtype=float32)\n",
            "epoch: 3 step: 556 train batch loss: tf.Tensor(0.49920225, shape=(), dtype=float32)\n",
            "epoch: 3 step: 557 train batch loss: tf.Tensor(0.43866697, shape=(), dtype=float32)\n",
            "epoch: 3 step: 558 train batch loss: tf.Tensor(0.3945981, shape=(), dtype=float32)\n",
            "epoch: 3 step: 559 train batch loss: tf.Tensor(0.4209051, shape=(), dtype=float32)\n",
            "epoch: 3 step: 560 train batch loss: tf.Tensor(0.3657062, shape=(), dtype=float32)\n",
            "epoch: 3 step: 561 train batch loss: tf.Tensor(0.42398182, shape=(), dtype=float32)\n",
            "epoch: 3 step: 562 train batch loss: tf.Tensor(0.44741434, shape=(), dtype=float32)\n",
            "epoch: 3 step: 563 train batch loss: tf.Tensor(0.43053192, shape=(), dtype=float32)\n",
            "epoch: 3 step: 564 train batch loss: tf.Tensor(0.44271684, shape=(), dtype=float32)\n",
            "epoch: 3 step: 565 train batch loss: tf.Tensor(0.55163056, shape=(), dtype=float32)\n",
            "epoch: 3 step: 566 train batch loss: tf.Tensor(0.43601283, shape=(), dtype=float32)\n",
            "epoch: 3 step: 567 train batch loss: tf.Tensor(0.37025672, shape=(), dtype=float32)\n",
            "epoch: 3 step: 568 train batch loss: tf.Tensor(0.42817506, shape=(), dtype=float32)\n",
            "epoch: 3 step: 569 train batch loss: tf.Tensor(0.45203343, shape=(), dtype=float32)\n",
            "epoch: 3 step: 570 train batch loss: tf.Tensor(0.44564137, shape=(), dtype=float32)\n",
            "epoch: 3 step: 571 train batch loss: tf.Tensor(0.45115107, shape=(), dtype=float32)\n",
            "epoch: 3 step: 572 train batch loss: tf.Tensor(0.49517375, shape=(), dtype=float32)\n",
            "epoch: 3 step: 573 train batch loss: tf.Tensor(0.40969762, shape=(), dtype=float32)\n",
            "epoch: 3 step: 574 train batch loss: tf.Tensor(0.4284749, shape=(), dtype=float32)\n",
            "epoch: 3 step: 575 train batch loss: tf.Tensor(0.41260576, shape=(), dtype=float32)\n",
            "epoch: 3 step: 576 train batch loss: tf.Tensor(0.4753748, shape=(), dtype=float32)\n",
            "epoch: 3 step: 577 train batch loss: tf.Tensor(0.42016625, shape=(), dtype=float32)\n",
            "epoch: 3 step: 578 train batch loss: tf.Tensor(0.48511785, shape=(), dtype=float32)\n",
            "epoch: 3 step: 579 train batch loss: tf.Tensor(0.59221, shape=(), dtype=float32)\n",
            "epoch: 3 step: 580 train batch loss: tf.Tensor(0.43592033, shape=(), dtype=float32)\n",
            "epoch: 3 step: 581 train batch loss: tf.Tensor(0.43442875, shape=(), dtype=float32)\n",
            "epoch: 3 step: 582 train batch loss: tf.Tensor(0.45862022, shape=(), dtype=float32)\n",
            "epoch: 3 step: 583 train batch loss: tf.Tensor(0.40197045, shape=(), dtype=float32)\n",
            "epoch: 3 step: 584 train batch loss: tf.Tensor(0.43334508, shape=(), dtype=float32)\n",
            "epoch: 3 step: 585 train batch loss: tf.Tensor(0.4792146, shape=(), dtype=float32)\n",
            "epoch: 3 step: 586 train batch loss: tf.Tensor(0.39233756, shape=(), dtype=float32)\n",
            "epoch: 3 step: 587 train batch loss: tf.Tensor(0.40512267, shape=(), dtype=float32)\n",
            "epoch: 3 step: 588 train batch loss: tf.Tensor(0.36987, shape=(), dtype=float32)\n",
            "epoch: 3 step: 589 train batch loss: tf.Tensor(0.4167405, shape=(), dtype=float32)\n",
            "epoch: 3 step: 590 train batch loss: tf.Tensor(0.42117992, shape=(), dtype=float32)\n",
            "epoch: 3 step: 591 train batch loss: tf.Tensor(0.43858367, shape=(), dtype=float32)\n",
            "epoch: 3 step: 592 train batch loss: tf.Tensor(0.57531583, shape=(), dtype=float32)\n",
            "epoch: 3 step: 593 train batch loss: tf.Tensor(0.4197529, shape=(), dtype=float32)\n",
            "epoch: 3 step: 594 train batch loss: tf.Tensor(0.39084446, shape=(), dtype=float32)\n",
            "epoch: 3 step: 595 train batch loss: tf.Tensor(0.4487156, shape=(), dtype=float32)\n",
            "epoch: 3 step: 596 train batch loss: tf.Tensor(0.43389624, shape=(), dtype=float32)\n",
            "epoch: 3 step: 597 train batch loss: tf.Tensor(0.3916638, shape=(), dtype=float32)\n",
            "epoch: 3 step: 598 train batch loss: tf.Tensor(0.50698024, shape=(), dtype=float32)\n",
            "epoch: 3 step: 599 train batch loss: tf.Tensor(0.3365737, shape=(), dtype=float32)\n",
            "epoch: 3 step: 600 train batch loss: tf.Tensor(0.5099291, shape=(), dtype=float32)\n",
            "epoch: 3 step: 601 train batch loss: tf.Tensor(0.48528558, shape=(), dtype=float32)\n",
            "epoch: 3 step: 602 train batch loss: tf.Tensor(0.54238063, shape=(), dtype=float32)\n",
            "epoch: 3 step: 603 train batch loss: tf.Tensor(0.50339234, shape=(), dtype=float32)\n",
            "epoch: 3 step: 604 train batch loss: tf.Tensor(0.44195667, shape=(), dtype=float32)\n",
            "epoch: 3 step: 605 train batch loss: tf.Tensor(0.49175972, shape=(), dtype=float32)\n",
            "epoch: 3 step: 606 train batch loss: tf.Tensor(0.38604861, shape=(), dtype=float32)\n",
            "epoch: 3 step: 607 train batch loss: tf.Tensor(0.53529507, shape=(), dtype=float32)\n",
            "epoch: 3 step: 608 train batch loss: tf.Tensor(0.4501546, shape=(), dtype=float32)\n",
            "epoch: 3 step: 609 train batch loss: tf.Tensor(0.451272, shape=(), dtype=float32)\n",
            "epoch: 3 step: 610 train batch loss: tf.Tensor(0.4908051, shape=(), dtype=float32)\n",
            "epoch: 3 step: 611 train batch loss: tf.Tensor(0.4521331, shape=(), dtype=float32)\n",
            "epoch: 3 step: 612 train batch loss: tf.Tensor(0.52154577, shape=(), dtype=float32)\n",
            "epoch: 3 step: 613 train batch loss: tf.Tensor(0.38285345, shape=(), dtype=float32)\n",
            "epoch: 3 step: 614 train batch loss: tf.Tensor(0.42308626, shape=(), dtype=float32)\n",
            "epoch: 3 step: 615 train batch loss: tf.Tensor(0.4573722, shape=(), dtype=float32)\n",
            "epoch: 3 step: 616 train batch loss: tf.Tensor(0.48757887, shape=(), dtype=float32)\n",
            "epoch: 3 step: 617 train batch loss: tf.Tensor(0.37225652, shape=(), dtype=float32)\n",
            "epoch: 3 step: 618 train batch loss: tf.Tensor(0.3849844, shape=(), dtype=float32)\n",
            "epoch: 3 step: 619 train batch loss: tf.Tensor(0.538251, shape=(), dtype=float32)\n",
            "epoch: 3 step: 620 train batch loss: tf.Tensor(0.42735475, shape=(), dtype=float32)\n",
            "epoch: 3 step: 621 train batch loss: tf.Tensor(0.38171107, shape=(), dtype=float32)\n",
            "epoch: 3 step: 622 train batch loss: tf.Tensor(0.4639511, shape=(), dtype=float32)\n",
            "epoch: 3 step: 623 train batch loss: tf.Tensor(0.44590768, shape=(), dtype=float32)\n",
            "epoch: 3 step: 624 train batch loss: tf.Tensor(0.425076, shape=(), dtype=float32)\n",
            "epoch: 3 step: 625 train batch loss: tf.Tensor(0.3860383, shape=(), dtype=float32)\n",
            "epoch: 3 step: 626 train batch loss: tf.Tensor(0.35013255, shape=(), dtype=float32)\n",
            "epoch: 3 step: 627 train batch loss: tf.Tensor(0.5041354, shape=(), dtype=float32)\n",
            "epoch: 3 step: 628 train batch loss: tf.Tensor(0.46215853, shape=(), dtype=float32)\n",
            "epoch: 3 step: 629 train batch loss: tf.Tensor(0.46251908, shape=(), dtype=float32)\n",
            "epoch: 3 step: 630 train batch loss: tf.Tensor(0.41244268, shape=(), dtype=float32)\n",
            "epoch: 3 step: 631 train batch loss: tf.Tensor(0.45520502, shape=(), dtype=float32)\n",
            "epoch: 3 step: 632 train batch loss: tf.Tensor(0.3592718, shape=(), dtype=float32)\n",
            "epoch: 3 step: 633 train batch loss: tf.Tensor(0.38992903, shape=(), dtype=float32)\n",
            "epoch: 3 step: 634 train batch loss: tf.Tensor(0.41990298, shape=(), dtype=float32)\n",
            "epoch: 3 step: 635 train batch loss: tf.Tensor(0.4967106, shape=(), dtype=float32)\n",
            "epoch: 3 step: 636 train batch loss: tf.Tensor(0.4640863, shape=(), dtype=float32)\n",
            "epoch: 3 step: 637 train batch loss: tf.Tensor(0.365489, shape=(), dtype=float32)\n",
            "epoch: 3 step: 638 train batch loss: tf.Tensor(0.50399566, shape=(), dtype=float32)\n",
            "epoch: 3 step: 639 train batch loss: tf.Tensor(0.43791366, shape=(), dtype=float32)\n",
            "epoch: 3 step: 640 train batch loss: tf.Tensor(0.5074154, shape=(), dtype=float32)\n",
            "epoch: 3 step: 641 train batch loss: tf.Tensor(0.6409747, shape=(), dtype=float32)\n",
            "epoch: 3 step: 642 train batch loss: tf.Tensor(0.553622, shape=(), dtype=float32)\n",
            "epoch: 3 step: 643 train batch loss: tf.Tensor(0.4834706, shape=(), dtype=float32)\n",
            "epoch: 3 step: 644 train batch loss: tf.Tensor(0.4457702, shape=(), dtype=float32)\n",
            "epoch: 3 step: 645 train batch loss: tf.Tensor(0.44952807, shape=(), dtype=float32)\n",
            "epoch: 3 step: 646 train batch loss: tf.Tensor(0.41420108, shape=(), dtype=float32)\n",
            "epoch: 3 step: 647 train batch loss: tf.Tensor(0.40855747, shape=(), dtype=float32)\n",
            "epoch: 3 step: 648 train batch loss: tf.Tensor(0.41016135, shape=(), dtype=float32)\n",
            "epoch: 3 step: 649 train batch loss: tf.Tensor(0.38680595, shape=(), dtype=float32)\n",
            "epoch: 3 step: 650 train batch loss: tf.Tensor(0.49777395, shape=(), dtype=float32)\n",
            "epoch: 3 step: 651 train batch loss: tf.Tensor(0.5586607, shape=(), dtype=float32)\n",
            "epoch: 3 step: 652 train batch loss: tf.Tensor(0.4545389, shape=(), dtype=float32)\n",
            "epoch: 3 step: 653 train batch loss: tf.Tensor(0.48646206, shape=(), dtype=float32)\n",
            "epoch: 3 step: 654 train batch loss: tf.Tensor(0.32079583, shape=(), dtype=float32)\n",
            "epoch: 3 step: 655 train batch loss: tf.Tensor(0.52743423, shape=(), dtype=float32)\n",
            "epoch: 3 step: 656 train batch loss: tf.Tensor(0.49209395, shape=(), dtype=float32)\n",
            "epoch: 3 step: 657 train batch loss: tf.Tensor(0.36968908, shape=(), dtype=float32)\n",
            "epoch: 3 step: 658 train batch loss: tf.Tensor(0.43550986, shape=(), dtype=float32)\n",
            "epoch: 3 step: 659 train batch loss: tf.Tensor(0.44896966, shape=(), dtype=float32)\n",
            "epoch: 3 step: 660 train batch loss: tf.Tensor(0.5364121, shape=(), dtype=float32)\n",
            "epoch: 3 step: 661 train batch loss: tf.Tensor(0.41303796, shape=(), dtype=float32)\n",
            "epoch: 3 step: 662 train batch loss: tf.Tensor(0.43966836, shape=(), dtype=float32)\n",
            "epoch: 3 step: 663 train batch loss: tf.Tensor(0.42093578, shape=(), dtype=float32)\n",
            "epoch: 3 step: 664 train batch loss: tf.Tensor(0.4277553, shape=(), dtype=float32)\n",
            "epoch: 3 step: 665 train batch loss: tf.Tensor(0.45630646, shape=(), dtype=float32)\n",
            "epoch: 3 step: 666 train batch loss: tf.Tensor(0.3966362, shape=(), dtype=float32)\n",
            "epoch: 3 step: 667 train batch loss: tf.Tensor(0.44242224, shape=(), dtype=float32)\n",
            "epoch: 3 step: 668 train batch loss: tf.Tensor(0.45854798, shape=(), dtype=float32)\n",
            "epoch: 3 step: 669 train batch loss: tf.Tensor(0.38451183, shape=(), dtype=float32)\n",
            "epoch: 3 step: 670 train batch loss: tf.Tensor(0.46649784, shape=(), dtype=float32)\n",
            "epoch: 3 step: 671 train batch loss: tf.Tensor(0.42669687, shape=(), dtype=float32)\n",
            "epoch: 3 step: 672 train batch loss: tf.Tensor(0.59306705, shape=(), dtype=float32)\n",
            "epoch: 3 step: 673 train batch loss: tf.Tensor(0.3654744, shape=(), dtype=float32)\n",
            "epoch: 3 step: 674 train batch loss: tf.Tensor(0.4291977, shape=(), dtype=float32)\n",
            "epoch: 3 step: 675 train batch loss: tf.Tensor(0.5043239, shape=(), dtype=float32)\n",
            "epoch: 3 step: 676 train batch loss: tf.Tensor(0.34777516, shape=(), dtype=float32)\n",
            "epoch: 3 step: 677 train batch loss: tf.Tensor(0.55699754, shape=(), dtype=float32)\n",
            "epoch: 3 step: 678 train batch loss: tf.Tensor(0.41834792, shape=(), dtype=float32)\n",
            "epoch: 3 step: 679 train batch loss: tf.Tensor(0.43771896, shape=(), dtype=float32)\n",
            "epoch: 3 step: 680 train batch loss: tf.Tensor(0.44177732, shape=(), dtype=float32)\n",
            "epoch: 3 step: 681 train batch loss: tf.Tensor(0.5022076, shape=(), dtype=float32)\n",
            "epoch: 3 step: 682 train batch loss: tf.Tensor(0.4291547, shape=(), dtype=float32)\n",
            "epoch: 3 step: 683 train batch loss: tf.Tensor(0.43158317, shape=(), dtype=float32)\n",
            "epoch: 3 step: 684 train batch loss: tf.Tensor(0.45613655, shape=(), dtype=float32)\n",
            "epoch: 3 step: 685 train batch loss: tf.Tensor(0.44884458, shape=(), dtype=float32)\n",
            "epoch: 3 step: 686 train batch loss: tf.Tensor(0.40644446, shape=(), dtype=float32)\n",
            "epoch: 3 step: 687 train batch loss: tf.Tensor(0.41114983, shape=(), dtype=float32)\n",
            "epoch: 3 step: 688 train batch loss: tf.Tensor(0.41510314, shape=(), dtype=float32)\n",
            "epoch: 3 step: 689 train batch loss: tf.Tensor(0.43955296, shape=(), dtype=float32)\n",
            "epoch: 3 step: 690 train batch loss: tf.Tensor(0.4303519, shape=(), dtype=float32)\n",
            "epoch: 3 step: 691 train batch loss: tf.Tensor(0.31703788, shape=(), dtype=float32)\n",
            "epoch: 3 step: 692 train batch loss: tf.Tensor(0.4161738, shape=(), dtype=float32)\n",
            "epoch: 3 step: 693 train batch loss: tf.Tensor(0.3077672, shape=(), dtype=float32)\n",
            "epoch: 3 step: 694 train batch loss: tf.Tensor(0.33729452, shape=(), dtype=float32)\n",
            "epoch: 3 step: 695 train batch loss: tf.Tensor(0.43850094, shape=(), dtype=float32)\n",
            "epoch: 3 step: 696 train batch loss: tf.Tensor(0.42327458, shape=(), dtype=float32)\n",
            "epoch: 3 step: 697 train batch loss: tf.Tensor(0.4932283, shape=(), dtype=float32)\n",
            "epoch: 3 step: 698 train batch loss: tf.Tensor(0.42100826, shape=(), dtype=float32)\n",
            "epoch: 3 step: 699 train batch loss: tf.Tensor(0.4721387, shape=(), dtype=float32)\n",
            "epoch: 3 step: 700 train batch loss: tf.Tensor(0.39233935, shape=(), dtype=float32)\n",
            "epoch: 3 step: 701 train batch loss: tf.Tensor(0.44073242, shape=(), dtype=float32)\n",
            "epoch: 3 step: 702 train batch loss: tf.Tensor(0.45575076, shape=(), dtype=float32)\n",
            "epoch: 3 step: 703 train batch loss: tf.Tensor(0.38096407, shape=(), dtype=float32)\n",
            "epoch: 3 step: 704 train batch loss: tf.Tensor(0.49362117, shape=(), dtype=float32)\n",
            "epoch: 3 step: 705 train batch loss: tf.Tensor(0.44690543, shape=(), dtype=float32)\n",
            "epoch: 3 step: 706 train batch loss: tf.Tensor(0.40647572, shape=(), dtype=float32)\n",
            "epoch: 3 step: 707 train batch loss: tf.Tensor(0.35352632, shape=(), dtype=float32)\n",
            "epoch: 3 step: 708 train batch loss: tf.Tensor(0.3543801, shape=(), dtype=float32)\n",
            "epoch: 3 step: 709 train batch loss: tf.Tensor(0.39275604, shape=(), dtype=float32)\n",
            "epoch: 3 step: 710 train batch loss: tf.Tensor(0.49070817, shape=(), dtype=float32)\n",
            "epoch: 3 step: 711 train batch loss: tf.Tensor(0.39231303, shape=(), dtype=float32)\n",
            "epoch: 3 step: 712 train batch loss: tf.Tensor(0.46118125, shape=(), dtype=float32)\n",
            "epoch: 3 step: 713 train batch loss: tf.Tensor(0.48676553, shape=(), dtype=float32)\n",
            "epoch: 3 step: 714 train batch loss: tf.Tensor(0.47727767, shape=(), dtype=float32)\n",
            "epoch: 3 step: 715 train batch loss: tf.Tensor(0.4335319, shape=(), dtype=float32)\n",
            "epoch: 3 step: 716 train batch loss: tf.Tensor(0.42466062, shape=(), dtype=float32)\n",
            "epoch: 3 step: 717 train batch loss: tf.Tensor(0.4478507, shape=(), dtype=float32)\n",
            "epoch: 3 step: 718 train batch loss: tf.Tensor(0.4823329, shape=(), dtype=float32)\n",
            "epoch: 3 step: 719 train batch loss: tf.Tensor(0.5506812, shape=(), dtype=float32)\n",
            "epoch: 3 step: 720 train batch loss: tf.Tensor(0.45764613, shape=(), dtype=float32)\n",
            "epoch: 3 step: 721 train batch loss: tf.Tensor(0.37299272, shape=(), dtype=float32)\n",
            "epoch: 3 step: 722 train batch loss: tf.Tensor(0.5221974, shape=(), dtype=float32)\n",
            "epoch: 3 step: 723 train batch loss: tf.Tensor(0.4333766, shape=(), dtype=float32)\n",
            "epoch: 3 step: 724 train batch loss: tf.Tensor(0.3924063, shape=(), dtype=float32)\n",
            "epoch: 3 step: 725 train batch loss: tf.Tensor(0.38368273, shape=(), dtype=float32)\n",
            "epoch: 3 step: 726 train batch loss: tf.Tensor(0.36688477, shape=(), dtype=float32)\n",
            "epoch: 3 step: 727 train batch loss: tf.Tensor(0.47224706, shape=(), dtype=float32)\n",
            "epoch: 3 step: 728 train batch loss: tf.Tensor(0.481638, shape=(), dtype=float32)\n",
            "epoch: 3 step: 729 train batch loss: tf.Tensor(0.44410688, shape=(), dtype=float32)\n",
            "epoch: 3 step: 730 train batch loss: tf.Tensor(0.4852758, shape=(), dtype=float32)\n",
            "epoch: 3 step: 731 train batch loss: tf.Tensor(0.37247488, shape=(), dtype=float32)\n",
            "epoch: 3 step: 732 train batch loss: tf.Tensor(0.46252307, shape=(), dtype=float32)\n",
            "epoch: 3 step: 733 train batch loss: tf.Tensor(0.56160855, shape=(), dtype=float32)\n",
            "epoch: 3 step: 734 train batch loss: tf.Tensor(0.52264965, shape=(), dtype=float32)\n",
            "epoch: 3 step: 735 train batch loss: tf.Tensor(0.42981476, shape=(), dtype=float32)\n",
            "epoch: 3 step: 736 train batch loss: tf.Tensor(0.48785695, shape=(), dtype=float32)\n",
            "epoch: 3 step: 737 train batch loss: tf.Tensor(0.3893933, shape=(), dtype=float32)\n",
            "epoch: 3 step: 738 train batch loss: tf.Tensor(0.48212424, shape=(), dtype=float32)\n",
            "epoch: 3 step: 739 train batch loss: tf.Tensor(0.3473659, shape=(), dtype=float32)\n",
            "epoch: 3 step: 740 train batch loss: tf.Tensor(0.4780733, shape=(), dtype=float32)\n",
            "epoch: 3 step: 741 train batch loss: tf.Tensor(0.475329, shape=(), dtype=float32)\n",
            "epoch: 3 step: 742 train batch loss: tf.Tensor(0.4648327, shape=(), dtype=float32)\n",
            "epoch: 3 step: 743 train batch loss: tf.Tensor(0.42472443, shape=(), dtype=float32)\n",
            "epoch: 3 step: 744 train batch loss: tf.Tensor(0.39249927, shape=(), dtype=float32)\n",
            "epoch: 3 step: 745 train batch loss: tf.Tensor(0.34886166, shape=(), dtype=float32)\n",
            "epoch: 3 step: 746 train batch loss: tf.Tensor(0.51008034, shape=(), dtype=float32)\n",
            "epoch: 3 step: 747 train batch loss: tf.Tensor(0.45097852, shape=(), dtype=float32)\n",
            "epoch: 3 step: 748 train batch loss: tf.Tensor(0.41423088, shape=(), dtype=float32)\n",
            "epoch: 3 step: 749 train batch loss: tf.Tensor(0.45334992, shape=(), dtype=float32)\n",
            "epoch: 3 step: 750 train batch loss: tf.Tensor(0.32388496, shape=(), dtype=float32)\n",
            "epoch: 3 step: 751 train batch loss: tf.Tensor(0.41535217, shape=(), dtype=float32)\n",
            "epoch: 3 step: 752 train batch loss: tf.Tensor(0.4982853, shape=(), dtype=float32)\n",
            "epoch: 3 step: 753 train batch loss: tf.Tensor(0.38356376, shape=(), dtype=float32)\n",
            "epoch: 3 step: 754 train batch loss: tf.Tensor(0.4612395, shape=(), dtype=float32)\n",
            "epoch: 3 step: 755 train batch loss: tf.Tensor(0.42323932, shape=(), dtype=float32)\n",
            "epoch: 3 step: 756 train batch loss: tf.Tensor(0.35718125, shape=(), dtype=float32)\n",
            "epoch: 3 step: 757 train batch loss: tf.Tensor(0.50437117, shape=(), dtype=float32)\n",
            "epoch: 3 step: 758 train batch loss: tf.Tensor(0.47986394, shape=(), dtype=float32)\n",
            "epoch: 3 step: 759 train batch loss: tf.Tensor(0.4929553, shape=(), dtype=float32)\n",
            "epoch: 3 step: 760 train batch loss: tf.Tensor(0.39676028, shape=(), dtype=float32)\n",
            "epoch: 3 step: 761 train batch loss: tf.Tensor(0.44109136, shape=(), dtype=float32)\n",
            "epoch: 3 step: 762 train batch loss: tf.Tensor(0.47016987, shape=(), dtype=float32)\n",
            "epoch: 3 step: 763 train batch loss: tf.Tensor(0.4504045, shape=(), dtype=float32)\n",
            "epoch: 3 step: 764 train batch loss: tf.Tensor(0.5097299, shape=(), dtype=float32)\n",
            "epoch: 3 step: 765 train batch loss: tf.Tensor(0.43535537, shape=(), dtype=float32)\n",
            "epoch: 3 step: 766 train batch loss: tf.Tensor(0.43791333, shape=(), dtype=float32)\n",
            "epoch: 3 step: 767 train batch loss: tf.Tensor(0.4067338, shape=(), dtype=float32)\n",
            "epoch: 3 step: 768 train batch loss: tf.Tensor(0.40979117, shape=(), dtype=float32)\n",
            "epoch: 3 step: 769 train batch loss: tf.Tensor(0.5003385, shape=(), dtype=float32)\n",
            "epoch: 3 step: 770 train batch loss: tf.Tensor(0.31762636, shape=(), dtype=float32)\n",
            "epoch: 3 step: 771 train batch loss: tf.Tensor(0.43217722, shape=(), dtype=float32)\n",
            "epoch: 3 step: 772 train batch loss: tf.Tensor(0.44848347, shape=(), dtype=float32)\n",
            "epoch: 3 step: 773 train batch loss: tf.Tensor(0.34471184, shape=(), dtype=float32)\n",
            "epoch: 3 step: 774 train batch loss: tf.Tensor(0.37236315, shape=(), dtype=float32)\n",
            "epoch: 3 step: 775 train batch loss: tf.Tensor(0.4010418, shape=(), dtype=float32)\n",
            "epoch: 3 step: 776 train batch loss: tf.Tensor(0.47170314, shape=(), dtype=float32)\n",
            "epoch: 3 step: 777 train batch loss: tf.Tensor(0.47944656, shape=(), dtype=float32)\n",
            "epoch: 3 step: 778 train batch loss: tf.Tensor(0.514938, shape=(), dtype=float32)\n",
            "epoch: 3 step: 779 train batch loss: tf.Tensor(0.38077623, shape=(), dtype=float32)\n",
            "epoch: 3 step: 780 train batch loss: tf.Tensor(0.5622432, shape=(), dtype=float32)\n",
            "epoch: 3 step: 781 train batch loss: tf.Tensor(0.45079678, shape=(), dtype=float32)\n",
            "epoch: 3 step: 782 train batch loss: tf.Tensor(0.49353218, shape=(), dtype=float32)\n",
            "epoch: 3 step: 783 train batch loss: tf.Tensor(0.41950488, shape=(), dtype=float32)\n",
            "epoch: 3 step: 784 train batch loss: tf.Tensor(0.48980016, shape=(), dtype=float32)\n",
            "epoch: 3 step: 785 train batch loss: tf.Tensor(0.41957426, shape=(), dtype=float32)\n",
            "epoch: 3 step: 786 train batch loss: tf.Tensor(0.38284686, shape=(), dtype=float32)\n",
            "epoch: 3 step: 787 train batch loss: tf.Tensor(0.47292838, shape=(), dtype=float32)\n",
            "epoch: 3 step: 788 train batch loss: tf.Tensor(0.36577365, shape=(), dtype=float32)\n",
            "epoch: 3 step: 789 train batch loss: tf.Tensor(0.41140532, shape=(), dtype=float32)\n",
            "epoch: 3 step: 790 train batch loss: tf.Tensor(0.4150445, shape=(), dtype=float32)\n",
            "epoch: 3 step: 791 train batch loss: tf.Tensor(0.4378141, shape=(), dtype=float32)\n",
            "epoch: 3 step: 792 train batch loss: tf.Tensor(0.43380696, shape=(), dtype=float32)\n",
            "epoch: 3 step: 793 train batch loss: tf.Tensor(0.4513619, shape=(), dtype=float32)\n",
            "epoch: 3 step: 794 train batch loss: tf.Tensor(0.41283154, shape=(), dtype=float32)\n",
            "epoch: 3 step: 795 train batch loss: tf.Tensor(0.36960247, shape=(), dtype=float32)\n",
            "epoch: 3 step: 796 train batch loss: tf.Tensor(0.47353745, shape=(), dtype=float32)\n",
            "epoch: 3 step: 797 train batch loss: tf.Tensor(0.46885666, shape=(), dtype=float32)\n",
            "epoch: 3 step: 798 train batch loss: tf.Tensor(0.44464794, shape=(), dtype=float32)\n",
            "epoch: 3 step: 799 train batch loss: tf.Tensor(0.39422807, shape=(), dtype=float32)\n",
            "epoch: 3 step: 800 train batch loss: tf.Tensor(0.5014675, shape=(), dtype=float32)\n",
            "epoch: 3 step: 801 train batch loss: tf.Tensor(0.5079684, shape=(), dtype=float32)\n",
            "epoch: 3 step: 802 train batch loss: tf.Tensor(0.44164282, shape=(), dtype=float32)\n",
            "epoch: 3 step: 803 train batch loss: tf.Tensor(0.60793126, shape=(), dtype=float32)\n",
            "epoch: 3 step: 804 train batch loss: tf.Tensor(0.4829846, shape=(), dtype=float32)\n",
            "epoch: 3 step: 805 train batch loss: tf.Tensor(0.39597565, shape=(), dtype=float32)\n",
            "epoch: 3 step: 806 train batch loss: tf.Tensor(0.42743796, shape=(), dtype=float32)\n",
            "epoch: 3 step: 807 train batch loss: tf.Tensor(0.35904428, shape=(), dtype=float32)\n",
            "epoch: 3 step: 808 train batch loss: tf.Tensor(0.3806132, shape=(), dtype=float32)\n",
            "epoch: 3 step: 809 train batch loss: tf.Tensor(0.4296555, shape=(), dtype=float32)\n",
            "epoch: 3 step: 810 train batch loss: tf.Tensor(0.42916545, shape=(), dtype=float32)\n",
            "epoch: 3 step: 811 train batch loss: tf.Tensor(0.39448315, shape=(), dtype=float32)\n",
            "epoch: 3 step: 812 train batch loss: tf.Tensor(0.44691366, shape=(), dtype=float32)\n",
            "epoch: 3 step: 813 train batch loss: tf.Tensor(0.37035042, shape=(), dtype=float32)\n",
            "epoch: 3 step: 814 train batch loss: tf.Tensor(0.37714094, shape=(), dtype=float32)\n",
            "epoch: 3 step: 815 train batch loss: tf.Tensor(0.4986559, shape=(), dtype=float32)\n",
            "epoch: 3 step: 816 train batch loss: tf.Tensor(0.3896866, shape=(), dtype=float32)\n",
            "epoch: 3 step: 817 train batch loss: tf.Tensor(0.41182333, shape=(), dtype=float32)\n",
            "epoch: 3 step: 818 train batch loss: tf.Tensor(0.41284624, shape=(), dtype=float32)\n",
            "epoch: 3 step: 819 train batch loss: tf.Tensor(0.3901804, shape=(), dtype=float32)\n",
            "epoch: 3 step: 820 train batch loss: tf.Tensor(0.4194777, shape=(), dtype=float32)\n",
            "epoch: 3 step: 821 train batch loss: tf.Tensor(0.37502372, shape=(), dtype=float32)\n",
            "epoch: 3 step: 822 train batch loss: tf.Tensor(0.4302307, shape=(), dtype=float32)\n",
            "epoch: 3 step: 823 train batch loss: tf.Tensor(0.4724556, shape=(), dtype=float32)\n",
            "epoch: 3 step: 824 train batch loss: tf.Tensor(0.44194207, shape=(), dtype=float32)\n",
            "epoch: 3 step: 825 train batch loss: tf.Tensor(0.48499012, shape=(), dtype=float32)\n",
            "epoch: 3 step: 826 train batch loss: tf.Tensor(0.4670487, shape=(), dtype=float32)\n",
            "epoch: 3 step: 827 train batch loss: tf.Tensor(0.43836337, shape=(), dtype=float32)\n",
            "epoch: 3 step: 828 train batch loss: tf.Tensor(0.37624612, shape=(), dtype=float32)\n",
            "epoch: 3 step: 829 train batch loss: tf.Tensor(0.47897857, shape=(), dtype=float32)\n",
            "epoch: 3 step: 830 train batch loss: tf.Tensor(0.4046715, shape=(), dtype=float32)\n",
            "epoch: 3 step: 831 train batch loss: tf.Tensor(0.3512364, shape=(), dtype=float32)\n",
            "epoch: 3 step: 832 train batch loss: tf.Tensor(0.42159405, shape=(), dtype=float32)\n",
            "epoch: 3 step: 833 train batch loss: tf.Tensor(0.49971765, shape=(), dtype=float32)\n",
            "epoch: 3 step: 834 train batch loss: tf.Tensor(0.34035584, shape=(), dtype=float32)\n",
            "epoch: 3 step: 835 train batch loss: tf.Tensor(0.38326854, shape=(), dtype=float32)\n",
            "epoch: 3 step: 836 train batch loss: tf.Tensor(0.3897998, shape=(), dtype=float32)\n",
            "epoch: 3 step: 837 train batch loss: tf.Tensor(0.50299823, shape=(), dtype=float32)\n",
            "epoch: 3 step: 838 train batch loss: tf.Tensor(0.4179852, shape=(), dtype=float32)\n",
            "epoch: 3 step: 839 train batch loss: tf.Tensor(0.37630576, shape=(), dtype=float32)\n",
            "epoch: 3 step: 840 train batch loss: tf.Tensor(0.36971658, shape=(), dtype=float32)\n",
            "epoch: 3 step: 841 train batch loss: tf.Tensor(0.41418156, shape=(), dtype=float32)\n",
            "epoch: 3 step: 842 train batch loss: tf.Tensor(0.513898, shape=(), dtype=float32)\n",
            "epoch: 3 step: 843 train batch loss: tf.Tensor(0.5308377, shape=(), dtype=float32)\n",
            "epoch: 3 step: 844 train batch loss: tf.Tensor(0.38954526, shape=(), dtype=float32)\n",
            "epoch: 3 step: 845 train batch loss: tf.Tensor(0.42224115, shape=(), dtype=float32)\n",
            "epoch: 3 step: 846 train batch loss: tf.Tensor(0.35667783, shape=(), dtype=float32)\n",
            "epoch: 3 step: 847 train batch loss: tf.Tensor(0.46860442, shape=(), dtype=float32)\n",
            "epoch: 3 step: 848 train batch loss: tf.Tensor(0.48971543, shape=(), dtype=float32)\n",
            "epoch: 3 step: 849 train batch loss: tf.Tensor(0.40463075, shape=(), dtype=float32)\n",
            "epoch: 3 step: 850 train batch loss: tf.Tensor(0.38964498, shape=(), dtype=float32)\n",
            "epoch: 3 step: 851 train batch loss: tf.Tensor(0.3971614, shape=(), dtype=float32)\n",
            "epoch: 3 step: 852 train batch loss: tf.Tensor(0.5351088, shape=(), dtype=float32)\n",
            "epoch: 3 step: 853 train batch loss: tf.Tensor(0.36461458, shape=(), dtype=float32)\n",
            "epoch: 3 step: 854 train batch loss: tf.Tensor(0.45519185, shape=(), dtype=float32)\n",
            "epoch: 3 step: 855 train batch loss: tf.Tensor(0.37897658, shape=(), dtype=float32)\n",
            "epoch: 3 step: 856 train batch loss: tf.Tensor(0.5007388, shape=(), dtype=float32)\n",
            "epoch: 3 step: 857 train batch loss: tf.Tensor(0.34774187, shape=(), dtype=float32)\n",
            "epoch: 3 step: 858 train batch loss: tf.Tensor(0.5245131, shape=(), dtype=float32)\n",
            "epoch: 3 step: 859 train batch loss: tf.Tensor(0.36037558, shape=(), dtype=float32)\n",
            "epoch: 3 step: 860 train batch loss: tf.Tensor(0.3655609, shape=(), dtype=float32)\n",
            "epoch: 3 step: 861 train batch loss: tf.Tensor(0.4600602, shape=(), dtype=float32)\n",
            "epoch: 3 step: 862 train batch loss: tf.Tensor(0.46083847, shape=(), dtype=float32)\n",
            "epoch: 3 step: 863 train batch loss: tf.Tensor(0.42645913, shape=(), dtype=float32)\n",
            "epoch: 3 step: 864 train batch loss: tf.Tensor(0.36107117, shape=(), dtype=float32)\n",
            "epoch: 3 step: 865 train batch loss: tf.Tensor(0.37709743, shape=(), dtype=float32)\n",
            "epoch: 3 step: 866 train batch loss: tf.Tensor(0.40651006, shape=(), dtype=float32)\n",
            "epoch: 3 step: 867 train batch loss: tf.Tensor(0.38004956, shape=(), dtype=float32)\n",
            "epoch: 3 step: 868 train batch loss: tf.Tensor(0.34272343, shape=(), dtype=float32)\n",
            "epoch: 3 step: 869 train batch loss: tf.Tensor(0.39859125, shape=(), dtype=float32)\n",
            "epoch: 3 step: 870 train batch loss: tf.Tensor(0.3971035, shape=(), dtype=float32)\n",
            "epoch: 3 step: 871 train batch loss: tf.Tensor(0.41252455, shape=(), dtype=float32)\n",
            "epoch: 3 step: 872 train batch loss: tf.Tensor(0.32718924, shape=(), dtype=float32)\n",
            "epoch: 3 step: 873 train batch loss: tf.Tensor(0.44230455, shape=(), dtype=float32)\n",
            "epoch: 3 step: 874 train batch loss: tf.Tensor(0.46256787, shape=(), dtype=float32)\n",
            "epoch: 3 step: 875 train batch loss: tf.Tensor(0.36764523, shape=(), dtype=float32)\n",
            "epoch: 3 step: 876 train batch loss: tf.Tensor(0.5108323, shape=(), dtype=float32)\n",
            "epoch: 3 step: 877 train batch loss: tf.Tensor(0.34845096, shape=(), dtype=float32)\n",
            "epoch: 3 step: 878 train batch loss: tf.Tensor(0.5105132, shape=(), dtype=float32)\n",
            "epoch: 3 step: 879 train batch loss: tf.Tensor(0.44058958, shape=(), dtype=float32)\n",
            "epoch: 3 step: 880 train batch loss: tf.Tensor(0.45481092, shape=(), dtype=float32)\n",
            "epoch: 3 step: 881 train batch loss: tf.Tensor(0.39310992, shape=(), dtype=float32)\n",
            "epoch: 3 step: 882 train batch loss: tf.Tensor(0.35824832, shape=(), dtype=float32)\n",
            "epoch: 3 step: 883 train batch loss: tf.Tensor(0.38750046, shape=(), dtype=float32)\n",
            "epoch: 3 step: 884 train batch loss: tf.Tensor(0.4161635, shape=(), dtype=float32)\n",
            "epoch: 3 step: 885 train batch loss: tf.Tensor(0.46430105, shape=(), dtype=float32)\n",
            "epoch: 3 step: 886 train batch loss: tf.Tensor(0.48185986, shape=(), dtype=float32)\n",
            "epoch: 3 step: 887 train batch loss: tf.Tensor(0.46765566, shape=(), dtype=float32)\n",
            "epoch: 3 step: 888 train batch loss: tf.Tensor(0.41280094, shape=(), dtype=float32)\n",
            "epoch: 3 step: 889 train batch loss: tf.Tensor(0.31829336, shape=(), dtype=float32)\n",
            "epoch: 3 step: 890 train batch loss: tf.Tensor(0.5620857, shape=(), dtype=float32)\n",
            "epoch: 3 step: 891 train batch loss: tf.Tensor(0.36627862, shape=(), dtype=float32)\n",
            "epoch: 3 step: 892 train batch loss: tf.Tensor(0.39481422, shape=(), dtype=float32)\n",
            "epoch: 3 step: 893 train batch loss: tf.Tensor(0.44125146, shape=(), dtype=float32)\n",
            "epoch: 3 step: 894 train batch loss: tf.Tensor(0.40743083, shape=(), dtype=float32)\n",
            "epoch: 3 step: 895 train batch loss: tf.Tensor(0.48271418, shape=(), dtype=float32)\n",
            "epoch: 3 step: 896 train batch loss: tf.Tensor(0.37886357, shape=(), dtype=float32)\n",
            "epoch: 3 step: 897 train batch loss: tf.Tensor(0.43644887, shape=(), dtype=float32)\n",
            "epoch: 3 step: 898 train batch loss: tf.Tensor(0.46333688, shape=(), dtype=float32)\n",
            "epoch: 3 step: 899 train batch loss: tf.Tensor(0.483704, shape=(), dtype=float32)\n",
            "epoch: 3 step: 900 train batch loss: tf.Tensor(0.41011444, shape=(), dtype=float32)\n",
            "epoch: 3 step: 901 train batch loss: tf.Tensor(0.4949653, shape=(), dtype=float32)\n",
            "epoch: 3 step: 902 train batch loss: tf.Tensor(0.4999009, shape=(), dtype=float32)\n",
            "epoch: 3 step: 903 train batch loss: tf.Tensor(0.34260637, shape=(), dtype=float32)\n",
            "epoch: 3 step: 904 train batch loss: tf.Tensor(0.38830614, shape=(), dtype=float32)\n",
            "epoch: 3 step: 905 train batch loss: tf.Tensor(0.429401, shape=(), dtype=float32)\n",
            "epoch: 3 step: 906 train batch loss: tf.Tensor(0.39965227, shape=(), dtype=float32)\n",
            "epoch: 3 step: 907 train batch loss: tf.Tensor(0.42883453, shape=(), dtype=float32)\n",
            "epoch: 3 step: 908 train batch loss: tf.Tensor(0.47175494, shape=(), dtype=float32)\n",
            "epoch: 3 step: 909 train batch loss: tf.Tensor(0.47348738, shape=(), dtype=float32)\n",
            "epoch: 3 step: 910 train batch loss: tf.Tensor(0.51755303, shape=(), dtype=float32)\n",
            "epoch: 3 step: 911 train batch loss: tf.Tensor(0.48377204, shape=(), dtype=float32)\n",
            "epoch: 3 step: 912 train batch loss: tf.Tensor(0.4279487, shape=(), dtype=float32)\n",
            "epoch: 3 step: 913 train batch loss: tf.Tensor(0.43443853, shape=(), dtype=float32)\n",
            "epoch: 3 step: 914 train batch loss: tf.Tensor(0.47230512, shape=(), dtype=float32)\n",
            "epoch: 3 step: 915 train batch loss: tf.Tensor(0.38722906, shape=(), dtype=float32)\n",
            "epoch: 3 step: 916 train batch loss: tf.Tensor(0.39368922, shape=(), dtype=float32)\n",
            "epoch: 3 step: 917 train batch loss: tf.Tensor(0.51468325, shape=(), dtype=float32)\n",
            "epoch: 3 step: 918 train batch loss: tf.Tensor(0.37599388, shape=(), dtype=float32)\n",
            "epoch: 3 step: 919 train batch loss: tf.Tensor(0.3234105, shape=(), dtype=float32)\n",
            "epoch: 3 step: 920 train batch loss: tf.Tensor(0.39550382, shape=(), dtype=float32)\n",
            "epoch: 3 step: 921 train batch loss: tf.Tensor(0.3995601, shape=(), dtype=float32)\n",
            "epoch: 3 step: 922 train batch loss: tf.Tensor(0.47102016, shape=(), dtype=float32)\n",
            "epoch: 3 step: 923 train batch loss: tf.Tensor(0.43392372, shape=(), dtype=float32)\n",
            "epoch: 3 step: 924 train batch loss: tf.Tensor(0.4804996, shape=(), dtype=float32)\n",
            "epoch: 3 step: 925 train batch loss: tf.Tensor(0.39395064, shape=(), dtype=float32)\n",
            "epoch: 3 step: 926 train batch loss: tf.Tensor(0.41434988, shape=(), dtype=float32)\n",
            "epoch: 3 step: 927 train batch loss: tf.Tensor(0.40865487, shape=(), dtype=float32)\n",
            "epoch: 3 step: 928 train batch loss: tf.Tensor(0.40833822, shape=(), dtype=float32)\n",
            "epoch: 3 step: 929 train batch loss: tf.Tensor(0.41422257, shape=(), dtype=float32)\n",
            "epoch: 3 step: 930 train batch loss: tf.Tensor(0.4869063, shape=(), dtype=float32)\n",
            "epoch: 3 step: 931 train batch loss: tf.Tensor(0.33849165, shape=(), dtype=float32)\n",
            "epoch: 3 step: 932 train batch loss: tf.Tensor(0.4430657, shape=(), dtype=float32)\n",
            "epoch: 3 step: 933 train batch loss: tf.Tensor(0.3916436, shape=(), dtype=float32)\n",
            "epoch: 3 step: 934 train batch loss: tf.Tensor(0.5021886, shape=(), dtype=float32)\n",
            "epoch: 3 step: 935 train batch loss: tf.Tensor(0.40412813, shape=(), dtype=float32)\n",
            "epoch: 3 step: 936 train batch loss: tf.Tensor(0.3956161, shape=(), dtype=float32)\n",
            "epoch: 3 step: 937 train batch loss: tf.Tensor(0.36314175, shape=(), dtype=float32)\n",
            "epoch: 3 step: 938 train batch loss: tf.Tensor(0.40617776, shape=(), dtype=float32)\n",
            "epoch: 3 step: 939 train batch loss: tf.Tensor(0.36508134, shape=(), dtype=float32)\n",
            "epoch: 3 step: 940 train batch loss: tf.Tensor(0.41539803, shape=(), dtype=float32)\n",
            "epoch: 3 step: 941 train batch loss: tf.Tensor(0.45773655, shape=(), dtype=float32)\n",
            "epoch: 3 step: 942 train batch loss: tf.Tensor(0.49343824, shape=(), dtype=float32)\n",
            "epoch: 3 step: 943 train batch loss: tf.Tensor(0.3865635, shape=(), dtype=float32)\n",
            "epoch: 3 step: 944 train batch loss: tf.Tensor(0.4173947, shape=(), dtype=float32)\n",
            "epoch: 3 step: 945 train batch loss: tf.Tensor(0.48589653, shape=(), dtype=float32)\n",
            "epoch: 3 step: 946 train batch loss: tf.Tensor(0.39954656, shape=(), dtype=float32)\n",
            "epoch: 3 step: 947 train batch loss: tf.Tensor(0.47966456, shape=(), dtype=float32)\n",
            "epoch: 3 step: 948 train batch loss: tf.Tensor(0.3495204, shape=(), dtype=float32)\n",
            "epoch: 3 step: 949 train batch loss: tf.Tensor(0.40370965, shape=(), dtype=float32)\n",
            "epoch: 3 step: 950 train batch loss: tf.Tensor(0.5258987, shape=(), dtype=float32)\n",
            "epoch: 3 step: 951 train batch loss: tf.Tensor(0.4307464, shape=(), dtype=float32)\n",
            "epoch: 3 step: 952 train batch loss: tf.Tensor(0.4118988, shape=(), dtype=float32)\n",
            "epoch: 3 step: 953 train batch loss: tf.Tensor(0.40109387, shape=(), dtype=float32)\n",
            "epoch: 3 step: 954 train batch loss: tf.Tensor(0.33359596, shape=(), dtype=float32)\n",
            "epoch: 3 step: 955 train batch loss: tf.Tensor(0.5108263, shape=(), dtype=float32)\n",
            "epoch: 3 step: 956 train batch loss: tf.Tensor(0.42048678, shape=(), dtype=float32)\n",
            "epoch: 3 step: 957 train batch loss: tf.Tensor(0.37408856, shape=(), dtype=float32)\n",
            "epoch: 3 step: 958 train batch loss: tf.Tensor(0.43128315, shape=(), dtype=float32)\n",
            "epoch: 3 step: 959 train batch loss: tf.Tensor(0.36399412, shape=(), dtype=float32)\n",
            "epoch: 3 step: 960 train batch loss: tf.Tensor(0.46375036, shape=(), dtype=float32)\n",
            "epoch: 3 step: 961 train batch loss: tf.Tensor(0.344047, shape=(), dtype=float32)\n",
            "epoch: 3 step: 962 train batch loss: tf.Tensor(0.4803143, shape=(), dtype=float32)\n",
            "epoch: 3 step: 963 train batch loss: tf.Tensor(0.47340766, shape=(), dtype=float32)\n",
            "epoch: 3 step: 964 train batch loss: tf.Tensor(0.37786633, shape=(), dtype=float32)\n",
            "epoch: 3 step: 965 train batch loss: tf.Tensor(0.5015631, shape=(), dtype=float32)\n",
            "epoch: 3 step: 966 train batch loss: tf.Tensor(0.44346756, shape=(), dtype=float32)\n",
            "epoch: 3 step: 967 train batch loss: tf.Tensor(0.4564472, shape=(), dtype=float32)\n",
            "epoch: 3 step: 968 train batch loss: tf.Tensor(0.41160217, shape=(), dtype=float32)\n",
            "epoch: 3 step: 969 train batch loss: tf.Tensor(0.36885783, shape=(), dtype=float32)\n",
            "epoch: 3 step: 970 train batch loss: tf.Tensor(0.40809226, shape=(), dtype=float32)\n",
            "epoch: 3 step: 971 train batch loss: tf.Tensor(0.49998188, shape=(), dtype=float32)\n",
            "epoch: 3 step: 972 train batch loss: tf.Tensor(0.32635143, shape=(), dtype=float32)\n",
            "epoch: 3 step: 973 train batch loss: tf.Tensor(0.35960054, shape=(), dtype=float32)\n",
            "epoch: 3 step: 974 train batch loss: tf.Tensor(0.4268198, shape=(), dtype=float32)\n",
            "epoch: 3 step: 975 train batch loss: tf.Tensor(0.4711071, shape=(), dtype=float32)\n",
            "epoch: 3 step: 976 train batch loss: tf.Tensor(0.4088942, shape=(), dtype=float32)\n",
            "epoch: 3 step: 977 train batch loss: tf.Tensor(0.59266245, shape=(), dtype=float32)\n",
            "epoch: 3 step: 978 train batch loss: tf.Tensor(0.46962714, shape=(), dtype=float32)\n",
            "epoch: 3 step: 979 train batch loss: tf.Tensor(0.41980982, shape=(), dtype=float32)\n",
            "epoch: 3 step: 980 train batch loss: tf.Tensor(0.41717958, shape=(), dtype=float32)\n",
            "epoch: 3 step: 981 train batch loss: tf.Tensor(0.29558808, shape=(), dtype=float32)\n",
            "epoch: 3 step: 982 train batch loss: tf.Tensor(0.45202485, shape=(), dtype=float32)\n",
            "epoch: 3 step: 983 train batch loss: tf.Tensor(0.40814742, shape=(), dtype=float32)\n",
            "epoch: 3 step: 984 train batch loss: tf.Tensor(0.44063252, shape=(), dtype=float32)\n",
            "epoch: 3 step: 985 train batch loss: tf.Tensor(0.36958736, shape=(), dtype=float32)\n",
            "epoch: 3 step: 986 train batch loss: tf.Tensor(0.46455136, shape=(), dtype=float32)\n",
            "epoch: 3 step: 987 train batch loss: tf.Tensor(0.3785631, shape=(), dtype=float32)\n",
            "epoch: 3 step: 988 train batch loss: tf.Tensor(0.38864866, shape=(), dtype=float32)\n",
            "epoch: 3 step: 989 train batch loss: tf.Tensor(0.4156509, shape=(), dtype=float32)\n",
            "epoch: 3 step: 990 train batch loss: tf.Tensor(0.37636304, shape=(), dtype=float32)\n",
            "epoch: 3 step: 991 train batch loss: tf.Tensor(0.45128208, shape=(), dtype=float32)\n",
            "epoch: 3 step: 992 train batch loss: tf.Tensor(0.46855077, shape=(), dtype=float32)\n",
            "epoch: 3 step: 993 train batch loss: tf.Tensor(0.38741398, shape=(), dtype=float32)\n",
            "epoch: 3 step: 994 train batch loss: tf.Tensor(0.42025495, shape=(), dtype=float32)\n",
            "epoch: 3 step: 995 train batch loss: tf.Tensor(0.41914925, shape=(), dtype=float32)\n",
            "epoch: 3 step: 996 train batch loss: tf.Tensor(0.4850203, shape=(), dtype=float32)\n",
            "epoch: 3 step: 997 train batch loss: tf.Tensor(0.3785821, shape=(), dtype=float32)\n",
            "epoch: 3 step: 998 train batch loss: tf.Tensor(0.40218052, shape=(), dtype=float32)\n",
            "epoch: 3 step: 999 train batch loss: tf.Tensor(0.43728408, shape=(), dtype=float32)\n",
            "epoch: 3 step: 0 val batch loss: tf.Tensor(0.72194386, shape=(), dtype=float32)\n",
            "epoch: 3 step: 1 val batch loss: tf.Tensor(0.80580837, shape=(), dtype=float32)\n",
            "epoch: 3 step: 2 val batch loss: tf.Tensor(0.67685336, shape=(), dtype=float32)\n",
            "epoch: 3 step: 3 val batch loss: tf.Tensor(1.0829496, shape=(), dtype=float32)\n",
            "epoch: 3 step: 4 val batch loss: tf.Tensor(0.59959877, shape=(), dtype=float32)\n",
            "epoch: 3 step: 5 val batch loss: tf.Tensor(0.6136718, shape=(), dtype=float32)\n",
            "epoch: 3 step: 6 val batch loss: tf.Tensor(0.51461434, shape=(), dtype=float32)\n",
            "epoch: 3 step: 7 val batch loss: tf.Tensor(0.5664032, shape=(), dtype=float32)\n",
            "epoch: 3 step: 8 val batch loss: tf.Tensor(0.46662778, shape=(), dtype=float32)\n",
            "epoch: 3 step: 9 val batch loss: tf.Tensor(0.4809888, shape=(), dtype=float32)\n",
            "epoch: 3 step: 10 val batch loss: tf.Tensor(0.6371014, shape=(), dtype=float32)\n",
            "epoch: 3 step: 11 val batch loss: tf.Tensor(0.78611815, shape=(), dtype=float32)\n",
            "epoch: 3 step: 12 val batch loss: tf.Tensor(0.55883056, shape=(), dtype=float32)\n",
            "epoch: 3 step: 13 val batch loss: tf.Tensor(0.4696704, shape=(), dtype=float32)\n",
            "epoch: 3 step: 14 val batch loss: tf.Tensor(0.83392537, shape=(), dtype=float32)\n",
            "epoch: 3 step: 15 val batch loss: tf.Tensor(0.51073825, shape=(), dtype=float32)\n",
            "epoch: 3 step: 16 val batch loss: tf.Tensor(0.75030583, shape=(), dtype=float32)\n",
            "epoch: 3 step: 17 val batch loss: tf.Tensor(0.95394087, shape=(), dtype=float32)\n",
            "epoch: 3 step: 18 val batch loss: tf.Tensor(0.82490337, shape=(), dtype=float32)\n",
            "epoch: 3 step: 19 val batch loss: tf.Tensor(0.46003643, shape=(), dtype=float32)\n",
            "epoch: 3 step: 20 val batch loss: tf.Tensor(0.5210837, shape=(), dtype=float32)\n",
            "epoch: 3 step: 21 val batch loss: tf.Tensor(0.70548505, shape=(), dtype=float32)\n",
            "epoch: 3 step: 22 val batch loss: tf.Tensor(0.69284034, shape=(), dtype=float32)\n",
            "epoch: 3 step: 23 val batch loss: tf.Tensor(0.7018752, shape=(), dtype=float32)\n",
            "epoch: 3 step: 24 val batch loss: tf.Tensor(0.7438293, shape=(), dtype=float32)\n",
            "epoch: 3 step: 25 val batch loss: tf.Tensor(0.780321, shape=(), dtype=float32)\n",
            "epoch: 3 step: 26 val batch loss: tf.Tensor(0.51048267, shape=(), dtype=float32)\n",
            "epoch: 3 step: 27 val batch loss: tf.Tensor(0.8361634, shape=(), dtype=float32)\n",
            "epoch: 3 step: 28 val batch loss: tf.Tensor(0.7856499, shape=(), dtype=float32)\n",
            "epoch: 3 step: 29 val batch loss: tf.Tensor(0.5287928, shape=(), dtype=float32)\n",
            "epoch: 3 step: 30 val batch loss: tf.Tensor(0.7274608, shape=(), dtype=float32)\n",
            "epoch: 3 step: 31 val batch loss: tf.Tensor(0.64829427, shape=(), dtype=float32)\n",
            "epoch: 3 step: 32 val batch loss: tf.Tensor(0.71321166, shape=(), dtype=float32)\n",
            "epoch: 3 step: 33 val batch loss: tf.Tensor(0.6822693, shape=(), dtype=float32)\n",
            "epoch: 3 step: 34 val batch loss: tf.Tensor(0.7564104, shape=(), dtype=float32)\n",
            "epoch: 3 step: 35 val batch loss: tf.Tensor(0.5269158, shape=(), dtype=float32)\n",
            "epoch: 3 step: 36 val batch loss: tf.Tensor(0.58393633, shape=(), dtype=float32)\n",
            "epoch: 3 step: 37 val batch loss: tf.Tensor(0.5398416, shape=(), dtype=float32)\n",
            "epoch: 3 step: 38 val batch loss: tf.Tensor(0.67234147, shape=(), dtype=float32)\n",
            "epoch: 3 step: 39 val batch loss: tf.Tensor(0.51559657, shape=(), dtype=float32)\n",
            "epoch: 3 step: 40 val batch loss: tf.Tensor(0.68955755, shape=(), dtype=float32)\n",
            "epoch: 3 step: 41 val batch loss: tf.Tensor(0.5966649, shape=(), dtype=float32)\n",
            "epoch: 3 step: 42 val batch loss: tf.Tensor(0.7930899, shape=(), dtype=float32)\n",
            "epoch: 3 step: 43 val batch loss: tf.Tensor(0.5972133, shape=(), dtype=float32)\n",
            "epoch: 3 step: 44 val batch loss: tf.Tensor(0.62630934, shape=(), dtype=float32)\n",
            "epoch: 3 step: 45 val batch loss: tf.Tensor(0.7658497, shape=(), dtype=float32)\n",
            "epoch: 3 step: 46 val batch loss: tf.Tensor(0.83721983, shape=(), dtype=float32)\n",
            "epoch: 3 step: 47 val batch loss: tf.Tensor(0.6084688, shape=(), dtype=float32)\n",
            "epoch: 3 step: 48 val batch loss: tf.Tensor(0.58234084, shape=(), dtype=float32)\n",
            "epoch: 3 step: 49 val batch loss: tf.Tensor(0.60090953, shape=(), dtype=float32)\n",
            "epoch: 3 step: 50 val batch loss: tf.Tensor(0.65455407, shape=(), dtype=float32)\n",
            "epoch: 3 step: 51 val batch loss: tf.Tensor(0.6161659, shape=(), dtype=float32)\n",
            "epoch: 3 step: 52 val batch loss: tf.Tensor(0.727136, shape=(), dtype=float32)\n",
            "epoch: 3 step: 53 val batch loss: tf.Tensor(0.6930664, shape=(), dtype=float32)\n",
            "epoch: 3 step: 54 val batch loss: tf.Tensor(0.4713958, shape=(), dtype=float32)\n",
            "epoch: 3 step: 55 val batch loss: tf.Tensor(0.77506655, shape=(), dtype=float32)\n",
            "epoch: 3 step: 56 val batch loss: tf.Tensor(0.58647805, shape=(), dtype=float32)\n",
            "epoch: 3 step: 57 val batch loss: tf.Tensor(0.6490935, shape=(), dtype=float32)\n",
            "epoch: 3 step: 58 val batch loss: tf.Tensor(0.7375294, shape=(), dtype=float32)\n",
            "epoch: 3 step: 59 val batch loss: tf.Tensor(0.6747211, shape=(), dtype=float32)\n",
            "epoch: 3 step: 60 val batch loss: tf.Tensor(0.57207096, shape=(), dtype=float32)\n",
            "epoch: 3 step: 61 val batch loss: tf.Tensor(0.5972079, shape=(), dtype=float32)\n",
            "epoch: 3 step: 62 val batch loss: tf.Tensor(0.7550681, shape=(), dtype=float32)\n",
            "epoch: 3 step: 63 val batch loss: tf.Tensor(0.5521172, shape=(), dtype=float32)\n",
            "epoch: 3 step: 64 val batch loss: tf.Tensor(0.6137705, shape=(), dtype=float32)\n",
            "epoch: 3 step: 65 val batch loss: tf.Tensor(0.5863761, shape=(), dtype=float32)\n",
            "epoch: 3 step: 66 val batch loss: tf.Tensor(0.66975605, shape=(), dtype=float32)\n",
            "epoch: 3 step: 67 val batch loss: tf.Tensor(0.7683398, shape=(), dtype=float32)\n",
            "epoch: 3 step: 68 val batch loss: tf.Tensor(0.672818, shape=(), dtype=float32)\n",
            "epoch: 3 step: 69 val batch loss: tf.Tensor(0.54403305, shape=(), dtype=float32)\n",
            "epoch: 3 step: 70 val batch loss: tf.Tensor(0.80192435, shape=(), dtype=float32)\n",
            "epoch: 3 step: 71 val batch loss: tf.Tensor(0.6774956, shape=(), dtype=float32)\n",
            "epoch: 3 step: 72 val batch loss: tf.Tensor(0.6911664, shape=(), dtype=float32)\n",
            "epoch: 3 step: 73 val batch loss: tf.Tensor(0.76730514, shape=(), dtype=float32)\n",
            "epoch: 3 step: 74 val batch loss: tf.Tensor(0.71898293, shape=(), dtype=float32)\n",
            "epoch: 3 step: 75 val batch loss: tf.Tensor(0.60009557, shape=(), dtype=float32)\n",
            "epoch: 3 step: 76 val batch loss: tf.Tensor(0.75472665, shape=(), dtype=float32)\n",
            "epoch: 3 step: 77 val batch loss: tf.Tensor(0.6997179, shape=(), dtype=float32)\n",
            "epoch: 3 step: 78 val batch loss: tf.Tensor(0.59765357, shape=(), dtype=float32)\n",
            "epoch: 3 step: 79 val batch loss: tf.Tensor(0.7240312, shape=(), dtype=float32)\n",
            "epoch: 3 step: 80 val batch loss: tf.Tensor(0.5388456, shape=(), dtype=float32)\n",
            "epoch: 3 step: 81 val batch loss: tf.Tensor(0.6918606, shape=(), dtype=float32)\n",
            "epoch: 3 step: 82 val batch loss: tf.Tensor(0.66240734, shape=(), dtype=float32)\n",
            "epoch: 3 step: 83 val batch loss: tf.Tensor(0.6204623, shape=(), dtype=float32)\n",
            "epoch: 3 step: 84 val batch loss: tf.Tensor(0.5378679, shape=(), dtype=float32)\n",
            "epoch: 3 step: 85 val batch loss: tf.Tensor(0.6291999, shape=(), dtype=float32)\n",
            "epoch: 3 step: 86 val batch loss: tf.Tensor(0.6690125, shape=(), dtype=float32)\n",
            "epoch: 3 step: 87 val batch loss: tf.Tensor(0.6678829, shape=(), dtype=float32)\n",
            "epoch: 3 step: 88 val batch loss: tf.Tensor(0.6667366, shape=(), dtype=float32)\n",
            "epoch: 3 step: 89 val batch loss: tf.Tensor(0.5229251, shape=(), dtype=float32)\n",
            "epoch: 3 step: 90 val batch loss: tf.Tensor(0.87872285, shape=(), dtype=float32)\n",
            "epoch: 3 step: 91 val batch loss: tf.Tensor(0.7168649, shape=(), dtype=float32)\n",
            "epoch: 3 step: 92 val batch loss: tf.Tensor(0.68079835, shape=(), dtype=float32)\n",
            "epoch: 3 step: 93 val batch loss: tf.Tensor(0.64030993, shape=(), dtype=float32)\n",
            "epoch: 3 step: 94 val batch loss: tf.Tensor(0.48292762, shape=(), dtype=float32)\n",
            "epoch: 3 step: 95 val batch loss: tf.Tensor(0.5382698, shape=(), dtype=float32)\n",
            "epoch: 3 step: 96 val batch loss: tf.Tensor(0.46607813, shape=(), dtype=float32)\n",
            "epoch: 3 step: 97 val batch loss: tf.Tensor(0.7849751, shape=(), dtype=float32)\n",
            "epoch: 3 step: 98 val batch loss: tf.Tensor(0.8047215, shape=(), dtype=float32)\n",
            "epoch: 3 step: 99 val batch loss: tf.Tensor(0.6793741, shape=(), dtype=float32)\n",
            "epoch: 3 step: 100 val batch loss: tf.Tensor(0.8711149, shape=(), dtype=float32)\n",
            "epoch: 3 step: 101 val batch loss: tf.Tensor(0.66703975, shape=(), dtype=float32)\n",
            "epoch: 3 step: 102 val batch loss: tf.Tensor(0.7459351, shape=(), dtype=float32)\n",
            "epoch: 3 step: 103 val batch loss: tf.Tensor(0.62446624, shape=(), dtype=float32)\n",
            "epoch: 3 step: 104 val batch loss: tf.Tensor(0.70279104, shape=(), dtype=float32)\n",
            "epoch: 3 step: 105 val batch loss: tf.Tensor(0.4781263, shape=(), dtype=float32)\n",
            "epoch: 3 step: 106 val batch loss: tf.Tensor(0.66302854, shape=(), dtype=float32)\n",
            "epoch: 3 step: 107 val batch loss: tf.Tensor(0.5947728, shape=(), dtype=float32)\n",
            "epoch: 3 step: 108 val batch loss: tf.Tensor(0.64959764, shape=(), dtype=float32)\n",
            "epoch: 3 step: 109 val batch loss: tf.Tensor(0.66551894, shape=(), dtype=float32)\n",
            "epoch: 3 step: 110 val batch loss: tf.Tensor(0.55921704, shape=(), dtype=float32)\n",
            "epoch: 3 step: 111 val batch loss: tf.Tensor(0.5998715, shape=(), dtype=float32)\n",
            "epoch: 3 step: 112 val batch loss: tf.Tensor(0.72677195, shape=(), dtype=float32)\n",
            "epoch: 3 step: 113 val batch loss: tf.Tensor(0.67933893, shape=(), dtype=float32)\n",
            "epoch: 3 step: 114 val batch loss: tf.Tensor(0.6698183, shape=(), dtype=float32)\n",
            "epoch: 3 step: 115 val batch loss: tf.Tensor(0.57195395, shape=(), dtype=float32)\n",
            "epoch: 3 step: 116 val batch loss: tf.Tensor(0.7187517, shape=(), dtype=float32)\n",
            "epoch: 3 step: 117 val batch loss: tf.Tensor(0.52733177, shape=(), dtype=float32)\n",
            "epoch: 3 step: 118 val batch loss: tf.Tensor(0.5754031, shape=(), dtype=float32)\n",
            "epoch: 3 step: 119 val batch loss: tf.Tensor(0.71115875, shape=(), dtype=float32)\n",
            "epoch: 3 step: 120 val batch loss: tf.Tensor(0.5632298, shape=(), dtype=float32)\n",
            "epoch: 3 step: 121 val batch loss: tf.Tensor(0.92073345, shape=(), dtype=float32)\n",
            "epoch: 3 step: 122 val batch loss: tf.Tensor(0.60961145, shape=(), dtype=float32)\n",
            "epoch: 3 step: 123 val batch loss: tf.Tensor(0.6683723, shape=(), dtype=float32)\n",
            "epoch: 3 step: 124 val batch loss: tf.Tensor(0.6953448, shape=(), dtype=float32)\n",
            "epoch: 3 step: 125 val batch loss: tf.Tensor(0.6679452, shape=(), dtype=float32)\n",
            "epoch: 3 step: 126 val batch loss: tf.Tensor(0.75666493, shape=(), dtype=float32)\n",
            "epoch: 3 step: 127 val batch loss: tf.Tensor(0.70405424, shape=(), dtype=float32)\n",
            "epoch: 3 step: 128 val batch loss: tf.Tensor(0.6662472, shape=(), dtype=float32)\n",
            "epoch: 3 step: 129 val batch loss: tf.Tensor(0.6366939, shape=(), dtype=float32)\n",
            "epoch: 3 step: 130 val batch loss: tf.Tensor(0.61804223, shape=(), dtype=float32)\n",
            "epoch: 3 step: 131 val batch loss: tf.Tensor(0.70435673, shape=(), dtype=float32)\n",
            "epoch: 3 step: 132 val batch loss: tf.Tensor(0.5486264, shape=(), dtype=float32)\n",
            "epoch: 3 step: 133 val batch loss: tf.Tensor(0.5688049, shape=(), dtype=float32)\n",
            "epoch: 3 step: 134 val batch loss: tf.Tensor(0.6996653, shape=(), dtype=float32)\n",
            "epoch: 3 step: 135 val batch loss: tf.Tensor(0.53316796, shape=(), dtype=float32)\n",
            "epoch: 3 step: 136 val batch loss: tf.Tensor(0.5574576, shape=(), dtype=float32)\n",
            "epoch: 3 step: 137 val batch loss: tf.Tensor(0.5693685, shape=(), dtype=float32)\n",
            "epoch: 3 step: 138 val batch loss: tf.Tensor(0.55771554, shape=(), dtype=float32)\n",
            "epoch: 3 step: 139 val batch loss: tf.Tensor(0.70986015, shape=(), dtype=float32)\n",
            "epoch: 3 step: 140 val batch loss: tf.Tensor(0.5229996, shape=(), dtype=float32)\n",
            "epoch: 3 step: 141 val batch loss: tf.Tensor(0.563928, shape=(), dtype=float32)\n",
            "epoch: 3 step: 142 val batch loss: tf.Tensor(0.7501667, shape=(), dtype=float32)\n",
            "epoch: 3 step: 143 val batch loss: tf.Tensor(0.68764424, shape=(), dtype=float32)\n",
            "epoch: 3 step: 144 val batch loss: tf.Tensor(0.6812081, shape=(), dtype=float32)\n",
            "epoch: 3 step: 145 val batch loss: tf.Tensor(0.67710984, shape=(), dtype=float32)\n",
            "epoch: 3 step: 146 val batch loss: tf.Tensor(0.62193054, shape=(), dtype=float32)\n",
            "epoch: 3 step: 147 val batch loss: tf.Tensor(0.70879614, shape=(), dtype=float32)\n",
            "epoch: 3 step: 148 val batch loss: tf.Tensor(0.5973679, shape=(), dtype=float32)\n",
            "epoch: 3 step: 149 val batch loss: tf.Tensor(0.69596636, shape=(), dtype=float32)\n",
            "epoch: 3 step: 150 val batch loss: tf.Tensor(0.582448, shape=(), dtype=float32)\n",
            "epoch: 3 step: 151 val batch loss: tf.Tensor(0.71851027, shape=(), dtype=float32)\n",
            "epoch: 3 step: 152 val batch loss: tf.Tensor(0.6016515, shape=(), dtype=float32)\n",
            "epoch: 3 step: 153 val batch loss: tf.Tensor(0.5475187, shape=(), dtype=float32)\n",
            "epoch: 3 step: 154 val batch loss: tf.Tensor(0.5897932, shape=(), dtype=float32)\n",
            "epoch: 3 step: 155 val batch loss: tf.Tensor(0.7194639, shape=(), dtype=float32)\n",
            "epoch: 3 step: 156 val batch loss: tf.Tensor(0.54585904, shape=(), dtype=float32)\n",
            "epoch: 3 step: 157 val batch loss: tf.Tensor(0.58524555, shape=(), dtype=float32)\n",
            "epoch: 3 step: 158 val batch loss: tf.Tensor(0.60094285, shape=(), dtype=float32)\n",
            "epoch: 3 step: 159 val batch loss: tf.Tensor(0.5123683, shape=(), dtype=float32)\n",
            "epoch: 3 step: 160 val batch loss: tf.Tensor(0.64598215, shape=(), dtype=float32)\n",
            "epoch: 3 step: 161 val batch loss: tf.Tensor(0.4837695, shape=(), dtype=float32)\n",
            "epoch: 3 step: 162 val batch loss: tf.Tensor(0.7301294, shape=(), dtype=float32)\n",
            "epoch: 3 step: 163 val batch loss: tf.Tensor(0.66468483, shape=(), dtype=float32)\n",
            "epoch: 3 step: 164 val batch loss: tf.Tensor(0.83148015, shape=(), dtype=float32)\n",
            "epoch: 3 step: 165 val batch loss: tf.Tensor(0.72495157, shape=(), dtype=float32)\n",
            "epoch: 3 step: 166 val batch loss: tf.Tensor(0.6415287, shape=(), dtype=float32)\n",
            "epoch: 3 step: 167 val batch loss: tf.Tensor(0.6098622, shape=(), dtype=float32)\n",
            "epoch: 3 step: 168 val batch loss: tf.Tensor(0.52612793, shape=(), dtype=float32)\n",
            "epoch: 3 step: 169 val batch loss: tf.Tensor(0.5212109, shape=(), dtype=float32)\n",
            "epoch: 3 step: 170 val batch loss: tf.Tensor(0.64687574, shape=(), dtype=float32)\n",
            "epoch: 3 step: 171 val batch loss: tf.Tensor(0.66634715, shape=(), dtype=float32)\n",
            "epoch: 3 step: 172 val batch loss: tf.Tensor(0.506121, shape=(), dtype=float32)\n",
            "epoch: 3 step: 173 val batch loss: tf.Tensor(0.6175202, shape=(), dtype=float32)\n",
            "epoch: 3 step: 174 val batch loss: tf.Tensor(0.78917354, shape=(), dtype=float32)\n",
            "epoch: 3 step: 175 val batch loss: tf.Tensor(0.7364279, shape=(), dtype=float32)\n",
            "epoch: 3 step: 176 val batch loss: tf.Tensor(0.57086086, shape=(), dtype=float32)\n",
            "epoch: 3 step: 177 val batch loss: tf.Tensor(0.6087713, shape=(), dtype=float32)\n",
            "epoch: 3 step: 178 val batch loss: tf.Tensor(0.47729295, shape=(), dtype=float32)\n",
            "epoch: 3 step: 179 val batch loss: tf.Tensor(0.612584, shape=(), dtype=float32)\n",
            "epoch: 3 step: 180 val batch loss: tf.Tensor(0.7095183, shape=(), dtype=float32)\n",
            "epoch: 3 step: 181 val batch loss: tf.Tensor(0.5371108, shape=(), dtype=float32)\n",
            "epoch: 3 step: 182 val batch loss: tf.Tensor(0.63737047, shape=(), dtype=float32)\n",
            "epoch: 3 step: 183 val batch loss: tf.Tensor(0.61762154, shape=(), dtype=float32)\n",
            "epoch: 3 step: 184 val batch loss: tf.Tensor(0.53716356, shape=(), dtype=float32)\n",
            "epoch: 3 step: 185 val batch loss: tf.Tensor(0.6817985, shape=(), dtype=float32)\n",
            "epoch: 3 step: 186 val batch loss: tf.Tensor(0.7514029, shape=(), dtype=float32)\n",
            "epoch: 3 step: 187 val batch loss: tf.Tensor(0.72621995, shape=(), dtype=float32)\n",
            "epoch: 3 step: 188 val batch loss: tf.Tensor(0.587785, shape=(), dtype=float32)\n",
            "epoch: 3 step: 189 val batch loss: tf.Tensor(0.5238988, shape=(), dtype=float32)\n",
            "epoch: 3 step: 190 val batch loss: tf.Tensor(0.600361, shape=(), dtype=float32)\n",
            "epoch: 3 step: 191 val batch loss: tf.Tensor(0.63187844, shape=(), dtype=float32)\n",
            "epoch: 3 step: 192 val batch loss: tf.Tensor(0.6961093, shape=(), dtype=float32)\n",
            "epoch: 3 step: 193 val batch loss: tf.Tensor(0.53866154, shape=(), dtype=float32)\n",
            "epoch: 3 step: 194 val batch loss: tf.Tensor(0.5686369, shape=(), dtype=float32)\n",
            "epoch: 3 step: 195 val batch loss: tf.Tensor(0.61830807, shape=(), dtype=float32)\n",
            "epoch: 3 step: 196 val batch loss: tf.Tensor(0.6627264, shape=(), dtype=float32)\n",
            "epoch: 3 step: 197 val batch loss: tf.Tensor(0.6296295, shape=(), dtype=float32)\n",
            "epoch: 3 step: 198 val batch loss: tf.Tensor(0.5896534, shape=(), dtype=float32)\n",
            "epoch: 3 step: 199 val batch loss: tf.Tensor(0.5012525, shape=(), dtype=float32)\n",
            "epoch: 3 step: 200 val batch loss: tf.Tensor(0.5613071, shape=(), dtype=float32)\n",
            "epoch: 3 step: 201 val batch loss: tf.Tensor(0.71292174, shape=(), dtype=float32)\n",
            "epoch: 3 step: 202 val batch loss: tf.Tensor(0.57912594, shape=(), dtype=float32)\n",
            "epoch: 3 step: 203 val batch loss: tf.Tensor(0.8793227, shape=(), dtype=float32)\n",
            "epoch: 3 step: 204 val batch loss: tf.Tensor(0.5502667, shape=(), dtype=float32)\n",
            "epoch: 3 step: 205 val batch loss: tf.Tensor(0.6069812, shape=(), dtype=float32)\n",
            "epoch: 3 step: 206 val batch loss: tf.Tensor(0.49455127, shape=(), dtype=float32)\n",
            "epoch: 3 step: 207 val batch loss: tf.Tensor(0.5732878, shape=(), dtype=float32)\n",
            "epoch: 3 step: 208 val batch loss: tf.Tensor(0.51559824, shape=(), dtype=float32)\n",
            "epoch: 3 step: 209 val batch loss: tf.Tensor(0.4991971, shape=(), dtype=float32)\n",
            "epoch: 3 step: 210 val batch loss: tf.Tensor(0.60058725, shape=(), dtype=float32)\n",
            "epoch: 3 step: 211 val batch loss: tf.Tensor(0.55747664, shape=(), dtype=float32)\n",
            "epoch: 3 step: 212 val batch loss: tf.Tensor(0.6626311, shape=(), dtype=float32)\n",
            "epoch: 3 step: 213 val batch loss: tf.Tensor(0.7745126, shape=(), dtype=float32)\n",
            "epoch: 3 step: 214 val batch loss: tf.Tensor(0.6252951, shape=(), dtype=float32)\n",
            "epoch: 3 step: 215 val batch loss: tf.Tensor(0.5706492, shape=(), dtype=float32)\n",
            "epoch: 3 step: 216 val batch loss: tf.Tensor(0.5351638, shape=(), dtype=float32)\n",
            "epoch: 3 step: 217 val batch loss: tf.Tensor(0.575688, shape=(), dtype=float32)\n",
            "epoch: 3 step: 218 val batch loss: tf.Tensor(0.6424661, shape=(), dtype=float32)\n",
            "epoch: 3 step: 219 val batch loss: tf.Tensor(0.45294896, shape=(), dtype=float32)\n",
            "epoch: 3 step: 220 val batch loss: tf.Tensor(0.46964464, shape=(), dtype=float32)\n",
            "epoch: 3 step: 221 val batch loss: tf.Tensor(0.71624196, shape=(), dtype=float32)\n",
            "epoch: 3 step: 222 val batch loss: tf.Tensor(0.74353206, shape=(), dtype=float32)\n",
            "epoch: 3 step: 223 val batch loss: tf.Tensor(0.81641495, shape=(), dtype=float32)\n",
            "epoch: 3 step: 224 val batch loss: tf.Tensor(0.7289504, shape=(), dtype=float32)\n",
            "epoch: 3 step: 225 val batch loss: tf.Tensor(0.6578505, shape=(), dtype=float32)\n",
            "epoch: 3 step: 226 val batch loss: tf.Tensor(0.71070564, shape=(), dtype=float32)\n",
            "epoch: 3 step: 227 val batch loss: tf.Tensor(0.6192882, shape=(), dtype=float32)\n",
            "epoch: 3 step: 228 val batch loss: tf.Tensor(0.50398743, shape=(), dtype=float32)\n",
            "epoch: 3 step: 229 val batch loss: tf.Tensor(1.019378, shape=(), dtype=float32)\n",
            "epoch: 3 step: 230 val batch loss: tf.Tensor(0.75604355, shape=(), dtype=float32)\n",
            "epoch: 3 step: 231 val batch loss: tf.Tensor(0.48397532, shape=(), dtype=float32)\n",
            "epoch: 3 step: 232 val batch loss: tf.Tensor(0.7359621, shape=(), dtype=float32)\n",
            "epoch: 3 step: 233 val batch loss: tf.Tensor(0.59275186, shape=(), dtype=float32)\n",
            "epoch: 3 step: 234 val batch loss: tf.Tensor(0.70019776, shape=(), dtype=float32)\n",
            "epoch: 3 step: 235 val batch loss: tf.Tensor(0.5951021, shape=(), dtype=float32)\n",
            "epoch: 3 step: 236 val batch loss: tf.Tensor(0.735838, shape=(), dtype=float32)\n",
            "epoch: 3 step: 237 val batch loss: tf.Tensor(0.7582825, shape=(), dtype=float32)\n",
            "epoch: 3 step: 238 val batch loss: tf.Tensor(0.6895479, shape=(), dtype=float32)\n",
            "epoch: 3 step: 239 val batch loss: tf.Tensor(0.6545163, shape=(), dtype=float32)\n",
            "epoch: 3 step: 240 val batch loss: tf.Tensor(0.64906174, shape=(), dtype=float32)\n",
            "epoch: 3 step: 241 val batch loss: tf.Tensor(0.6450723, shape=(), dtype=float32)\n",
            "epoch: 3 step: 242 val batch loss: tf.Tensor(0.61510605, shape=(), dtype=float32)\n",
            "epoch: 3 step: 243 val batch loss: tf.Tensor(0.7415546, shape=(), dtype=float32)\n",
            "epoch: 3 step: 244 val batch loss: tf.Tensor(0.5818571, shape=(), dtype=float32)\n",
            "epoch: 3 step: 245 val batch loss: tf.Tensor(0.46956125, shape=(), dtype=float32)\n",
            "epoch: 3 step: 246 val batch loss: tf.Tensor(0.51802284, shape=(), dtype=float32)\n",
            "epoch: 3 step: 247 val batch loss: tf.Tensor(0.7404416, shape=(), dtype=float32)\n",
            "epoch: 3 step: 248 val batch loss: tf.Tensor(0.82409656, shape=(), dtype=float32)\n",
            "epoch: 3 step: 249 val batch loss: tf.Tensor(0.68401504, shape=(), dtype=float32)\n",
            "Epoch 3 training loss: 0.45923951268196106 val loss: 0.6462810039520264\n",
            "epoch: 4 step: 0 train batch loss: tf.Tensor(0.37823874, shape=(), dtype=float32)\n",
            "epoch: 4 step: 1 train batch loss: tf.Tensor(0.3809076, shape=(), dtype=float32)\n",
            "epoch: 4 step: 2 train batch loss: tf.Tensor(0.47096047, shape=(), dtype=float32)\n",
            "epoch: 4 step: 3 train batch loss: tf.Tensor(0.35219392, shape=(), dtype=float32)\n",
            "epoch: 4 step: 4 train batch loss: tf.Tensor(0.512317, shape=(), dtype=float32)\n",
            "epoch: 4 step: 5 train batch loss: tf.Tensor(0.34788185, shape=(), dtype=float32)\n",
            "epoch: 4 step: 6 train batch loss: tf.Tensor(0.46282107, shape=(), dtype=float32)\n",
            "epoch: 4 step: 7 train batch loss: tf.Tensor(0.40919676, shape=(), dtype=float32)\n",
            "epoch: 4 step: 8 train batch loss: tf.Tensor(0.42383057, shape=(), dtype=float32)\n",
            "epoch: 4 step: 9 train batch loss: tf.Tensor(0.4809758, shape=(), dtype=float32)\n",
            "epoch: 4 step: 10 train batch loss: tf.Tensor(0.4383892, shape=(), dtype=float32)\n",
            "epoch: 4 step: 11 train batch loss: tf.Tensor(0.45240894, shape=(), dtype=float32)\n",
            "epoch: 4 step: 12 train batch loss: tf.Tensor(0.43330294, shape=(), dtype=float32)\n",
            "epoch: 4 step: 13 train batch loss: tf.Tensor(0.41059223, shape=(), dtype=float32)\n",
            "epoch: 4 step: 14 train batch loss: tf.Tensor(0.41692877, shape=(), dtype=float32)\n",
            "epoch: 4 step: 15 train batch loss: tf.Tensor(0.365725, shape=(), dtype=float32)\n",
            "epoch: 4 step: 16 train batch loss: tf.Tensor(0.45032376, shape=(), dtype=float32)\n",
            "epoch: 4 step: 17 train batch loss: tf.Tensor(0.43022776, shape=(), dtype=float32)\n",
            "epoch: 4 step: 18 train batch loss: tf.Tensor(0.35974625, shape=(), dtype=float32)\n",
            "epoch: 4 step: 19 train batch loss: tf.Tensor(0.39505303, shape=(), dtype=float32)\n",
            "epoch: 4 step: 20 train batch loss: tf.Tensor(0.47639927, shape=(), dtype=float32)\n",
            "epoch: 4 step: 21 train batch loss: tf.Tensor(0.433105, shape=(), dtype=float32)\n",
            "epoch: 4 step: 22 train batch loss: tf.Tensor(0.38954633, shape=(), dtype=float32)\n",
            "epoch: 4 step: 23 train batch loss: tf.Tensor(0.37574273, shape=(), dtype=float32)\n",
            "epoch: 4 step: 24 train batch loss: tf.Tensor(0.4647282, shape=(), dtype=float32)\n",
            "epoch: 4 step: 25 train batch loss: tf.Tensor(0.38030538, shape=(), dtype=float32)\n",
            "epoch: 4 step: 26 train batch loss: tf.Tensor(0.4051796, shape=(), dtype=float32)\n",
            "epoch: 4 step: 27 train batch loss: tf.Tensor(0.32975706, shape=(), dtype=float32)\n",
            "epoch: 4 step: 28 train batch loss: tf.Tensor(0.43109864, shape=(), dtype=float32)\n",
            "epoch: 4 step: 29 train batch loss: tf.Tensor(0.37804452, shape=(), dtype=float32)\n",
            "epoch: 4 step: 30 train batch loss: tf.Tensor(0.4178838, shape=(), dtype=float32)\n",
            "epoch: 4 step: 31 train batch loss: tf.Tensor(0.44136232, shape=(), dtype=float32)\n",
            "epoch: 4 step: 32 train batch loss: tf.Tensor(0.3824154, shape=(), dtype=float32)\n",
            "epoch: 4 step: 33 train batch loss: tf.Tensor(0.33390674, shape=(), dtype=float32)\n",
            "epoch: 4 step: 34 train batch loss: tf.Tensor(0.3840254, shape=(), dtype=float32)\n",
            "epoch: 4 step: 35 train batch loss: tf.Tensor(0.3867014, shape=(), dtype=float32)\n",
            "epoch: 4 step: 36 train batch loss: tf.Tensor(0.3265759, shape=(), dtype=float32)\n",
            "epoch: 4 step: 37 train batch loss: tf.Tensor(0.3584346, shape=(), dtype=float32)\n",
            "epoch: 4 step: 38 train batch loss: tf.Tensor(0.39082652, shape=(), dtype=float32)\n",
            "epoch: 4 step: 39 train batch loss: tf.Tensor(0.46721262, shape=(), dtype=float32)\n",
            "epoch: 4 step: 40 train batch loss: tf.Tensor(0.622069, shape=(), dtype=float32)\n",
            "epoch: 4 step: 41 train batch loss: tf.Tensor(0.44008785, shape=(), dtype=float32)\n",
            "epoch: 4 step: 42 train batch loss: tf.Tensor(0.44734263, shape=(), dtype=float32)\n",
            "epoch: 4 step: 43 train batch loss: tf.Tensor(0.47545037, shape=(), dtype=float32)\n",
            "epoch: 4 step: 44 train batch loss: tf.Tensor(0.36039752, shape=(), dtype=float32)\n",
            "epoch: 4 step: 45 train batch loss: tf.Tensor(0.36379406, shape=(), dtype=float32)\n",
            "epoch: 4 step: 46 train batch loss: tf.Tensor(0.36904988, shape=(), dtype=float32)\n",
            "epoch: 4 step: 47 train batch loss: tf.Tensor(0.35277733, shape=(), dtype=float32)\n",
            "epoch: 4 step: 48 train batch loss: tf.Tensor(0.46585163, shape=(), dtype=float32)\n",
            "epoch: 4 step: 49 train batch loss: tf.Tensor(0.48558322, shape=(), dtype=float32)\n",
            "epoch: 4 step: 50 train batch loss: tf.Tensor(0.44783458, shape=(), dtype=float32)\n",
            "epoch: 4 step: 51 train batch loss: tf.Tensor(0.36679465, shape=(), dtype=float32)\n",
            "epoch: 4 step: 52 train batch loss: tf.Tensor(0.38375965, shape=(), dtype=float32)\n",
            "epoch: 4 step: 53 train batch loss: tf.Tensor(0.34576586, shape=(), dtype=float32)\n",
            "epoch: 4 step: 54 train batch loss: tf.Tensor(0.43038845, shape=(), dtype=float32)\n",
            "epoch: 4 step: 55 train batch loss: tf.Tensor(0.39938873, shape=(), dtype=float32)\n",
            "epoch: 4 step: 56 train batch loss: tf.Tensor(0.3763453, shape=(), dtype=float32)\n",
            "epoch: 4 step: 57 train batch loss: tf.Tensor(0.3148725, shape=(), dtype=float32)\n",
            "epoch: 4 step: 58 train batch loss: tf.Tensor(0.4498883, shape=(), dtype=float32)\n",
            "epoch: 4 step: 59 train batch loss: tf.Tensor(0.46444193, shape=(), dtype=float32)\n",
            "epoch: 4 step: 60 train batch loss: tf.Tensor(0.4033119, shape=(), dtype=float32)\n",
            "epoch: 4 step: 61 train batch loss: tf.Tensor(0.4419008, shape=(), dtype=float32)\n",
            "epoch: 4 step: 62 train batch loss: tf.Tensor(0.44071656, shape=(), dtype=float32)\n",
            "epoch: 4 step: 63 train batch loss: tf.Tensor(0.31916213, shape=(), dtype=float32)\n",
            "epoch: 4 step: 64 train batch loss: tf.Tensor(0.44226342, shape=(), dtype=float32)\n",
            "epoch: 4 step: 65 train batch loss: tf.Tensor(0.4167041, shape=(), dtype=float32)\n",
            "epoch: 4 step: 66 train batch loss: tf.Tensor(0.36817166, shape=(), dtype=float32)\n",
            "epoch: 4 step: 67 train batch loss: tf.Tensor(0.31552652, shape=(), dtype=float32)\n",
            "epoch: 4 step: 68 train batch loss: tf.Tensor(0.47921196, shape=(), dtype=float32)\n",
            "epoch: 4 step: 69 train batch loss: tf.Tensor(0.44747955, shape=(), dtype=float32)\n",
            "epoch: 4 step: 70 train batch loss: tf.Tensor(0.3917835, shape=(), dtype=float32)\n",
            "epoch: 4 step: 71 train batch loss: tf.Tensor(0.39639348, shape=(), dtype=float32)\n",
            "epoch: 4 step: 72 train batch loss: tf.Tensor(0.5374037, shape=(), dtype=float32)\n",
            "epoch: 4 step: 73 train batch loss: tf.Tensor(0.28336892, shape=(), dtype=float32)\n",
            "epoch: 4 step: 74 train batch loss: tf.Tensor(0.4886468, shape=(), dtype=float32)\n",
            "epoch: 4 step: 75 train batch loss: tf.Tensor(0.35875976, shape=(), dtype=float32)\n",
            "epoch: 4 step: 76 train batch loss: tf.Tensor(0.4286542, shape=(), dtype=float32)\n",
            "epoch: 4 step: 77 train batch loss: tf.Tensor(0.4371403, shape=(), dtype=float32)\n",
            "epoch: 4 step: 78 train batch loss: tf.Tensor(0.311235, shape=(), dtype=float32)\n",
            "epoch: 4 step: 79 train batch loss: tf.Tensor(0.33568054, shape=(), dtype=float32)\n",
            "epoch: 4 step: 80 train batch loss: tf.Tensor(0.36287186, shape=(), dtype=float32)\n",
            "epoch: 4 step: 81 train batch loss: tf.Tensor(0.39031193, shape=(), dtype=float32)\n",
            "epoch: 4 step: 82 train batch loss: tf.Tensor(0.43083566, shape=(), dtype=float32)\n",
            "epoch: 4 step: 83 train batch loss: tf.Tensor(0.41044816, shape=(), dtype=float32)\n",
            "epoch: 4 step: 84 train batch loss: tf.Tensor(0.4571081, shape=(), dtype=float32)\n",
            "epoch: 4 step: 85 train batch loss: tf.Tensor(0.42558497, shape=(), dtype=float32)\n",
            "epoch: 4 step: 86 train batch loss: tf.Tensor(0.44789815, shape=(), dtype=float32)\n",
            "epoch: 4 step: 87 train batch loss: tf.Tensor(0.40972495, shape=(), dtype=float32)\n",
            "epoch: 4 step: 88 train batch loss: tf.Tensor(0.364218, shape=(), dtype=float32)\n",
            "epoch: 4 step: 89 train batch loss: tf.Tensor(0.4031559, shape=(), dtype=float32)\n",
            "epoch: 4 step: 90 train batch loss: tf.Tensor(0.3674066, shape=(), dtype=float32)\n",
            "epoch: 4 step: 91 train batch loss: tf.Tensor(0.47519, shape=(), dtype=float32)\n",
            "epoch: 4 step: 92 train batch loss: tf.Tensor(0.54951924, shape=(), dtype=float32)\n",
            "epoch: 4 step: 93 train batch loss: tf.Tensor(0.33361658, shape=(), dtype=float32)\n",
            "epoch: 4 step: 94 train batch loss: tf.Tensor(0.39008906, shape=(), dtype=float32)\n",
            "epoch: 4 step: 95 train batch loss: tf.Tensor(0.3716331, shape=(), dtype=float32)\n",
            "epoch: 4 step: 96 train batch loss: tf.Tensor(0.3938127, shape=(), dtype=float32)\n",
            "epoch: 4 step: 97 train batch loss: tf.Tensor(0.35209623, shape=(), dtype=float32)\n",
            "epoch: 4 step: 98 train batch loss: tf.Tensor(0.33270484, shape=(), dtype=float32)\n",
            "epoch: 4 step: 99 train batch loss: tf.Tensor(0.43361157, shape=(), dtype=float32)\n",
            "epoch: 4 step: 100 train batch loss: tf.Tensor(0.38964173, shape=(), dtype=float32)\n",
            "epoch: 4 step: 101 train batch loss: tf.Tensor(0.37261793, shape=(), dtype=float32)\n",
            "epoch: 4 step: 102 train batch loss: tf.Tensor(0.46865082, shape=(), dtype=float32)\n",
            "epoch: 4 step: 103 train batch loss: tf.Tensor(0.4026162, shape=(), dtype=float32)\n",
            "epoch: 4 step: 104 train batch loss: tf.Tensor(0.44262776, shape=(), dtype=float32)\n",
            "epoch: 4 step: 105 train batch loss: tf.Tensor(0.384147, shape=(), dtype=float32)\n",
            "epoch: 4 step: 106 train batch loss: tf.Tensor(0.4006515, shape=(), dtype=float32)\n",
            "epoch: 4 step: 107 train batch loss: tf.Tensor(0.4507936, shape=(), dtype=float32)\n",
            "epoch: 4 step: 108 train batch loss: tf.Tensor(0.33872104, shape=(), dtype=float32)\n",
            "epoch: 4 step: 109 train batch loss: tf.Tensor(0.4111305, shape=(), dtype=float32)\n",
            "epoch: 4 step: 110 train batch loss: tf.Tensor(0.40572602, shape=(), dtype=float32)\n",
            "epoch: 4 step: 111 train batch loss: tf.Tensor(0.38013113, shape=(), dtype=float32)\n",
            "epoch: 4 step: 112 train batch loss: tf.Tensor(0.44815215, shape=(), dtype=float32)\n",
            "epoch: 4 step: 113 train batch loss: tf.Tensor(0.34202212, shape=(), dtype=float32)\n",
            "epoch: 4 step: 114 train batch loss: tf.Tensor(0.33917123, shape=(), dtype=float32)\n",
            "epoch: 4 step: 115 train batch loss: tf.Tensor(0.3870755, shape=(), dtype=float32)\n",
            "epoch: 4 step: 116 train batch loss: tf.Tensor(0.35288423, shape=(), dtype=float32)\n",
            "epoch: 4 step: 117 train batch loss: tf.Tensor(0.383167, shape=(), dtype=float32)\n",
            "epoch: 4 step: 118 train batch loss: tf.Tensor(0.4854628, shape=(), dtype=float32)\n",
            "epoch: 4 step: 119 train batch loss: tf.Tensor(0.37904745, shape=(), dtype=float32)\n",
            "epoch: 4 step: 120 train batch loss: tf.Tensor(0.32959536, shape=(), dtype=float32)\n",
            "epoch: 4 step: 121 train batch loss: tf.Tensor(0.35982046, shape=(), dtype=float32)\n",
            "epoch: 4 step: 122 train batch loss: tf.Tensor(0.3435038, shape=(), dtype=float32)\n",
            "epoch: 4 step: 123 train batch loss: tf.Tensor(0.414267, shape=(), dtype=float32)\n",
            "epoch: 4 step: 124 train batch loss: tf.Tensor(0.44897285, shape=(), dtype=float32)\n",
            "epoch: 4 step: 125 train batch loss: tf.Tensor(0.37224942, shape=(), dtype=float32)\n",
            "epoch: 4 step: 126 train batch loss: tf.Tensor(0.37672788, shape=(), dtype=float32)\n",
            "epoch: 4 step: 127 train batch loss: tf.Tensor(0.55296695, shape=(), dtype=float32)\n",
            "epoch: 4 step: 128 train batch loss: tf.Tensor(0.3972891, shape=(), dtype=float32)\n",
            "epoch: 4 step: 129 train batch loss: tf.Tensor(0.34621277, shape=(), dtype=float32)\n",
            "epoch: 4 step: 130 train batch loss: tf.Tensor(0.4023164, shape=(), dtype=float32)\n",
            "epoch: 4 step: 131 train batch loss: tf.Tensor(0.39910555, shape=(), dtype=float32)\n",
            "epoch: 4 step: 132 train batch loss: tf.Tensor(0.33923128, shape=(), dtype=float32)\n",
            "epoch: 4 step: 133 train batch loss: tf.Tensor(0.3220734, shape=(), dtype=float32)\n",
            "epoch: 4 step: 134 train batch loss: tf.Tensor(0.3402564, shape=(), dtype=float32)\n",
            "epoch: 4 step: 135 train batch loss: tf.Tensor(0.39280343, shape=(), dtype=float32)\n",
            "epoch: 4 step: 136 train batch loss: tf.Tensor(0.48433393, shape=(), dtype=float32)\n",
            "epoch: 4 step: 137 train batch loss: tf.Tensor(0.39262116, shape=(), dtype=float32)\n",
            "epoch: 4 step: 138 train batch loss: tf.Tensor(0.4074139, shape=(), dtype=float32)\n",
            "epoch: 4 step: 139 train batch loss: tf.Tensor(0.3074337, shape=(), dtype=float32)\n",
            "epoch: 4 step: 140 train batch loss: tf.Tensor(0.35646313, shape=(), dtype=float32)\n",
            "epoch: 4 step: 141 train batch loss: tf.Tensor(0.44307357, shape=(), dtype=float32)\n",
            "epoch: 4 step: 142 train batch loss: tf.Tensor(0.3912081, shape=(), dtype=float32)\n",
            "epoch: 4 step: 143 train batch loss: tf.Tensor(0.4611327, shape=(), dtype=float32)\n",
            "epoch: 4 step: 144 train batch loss: tf.Tensor(0.39088458, shape=(), dtype=float32)\n",
            "epoch: 4 step: 145 train batch loss: tf.Tensor(0.42307988, shape=(), dtype=float32)\n",
            "epoch: 4 step: 146 train batch loss: tf.Tensor(0.34065467, shape=(), dtype=float32)\n",
            "epoch: 4 step: 147 train batch loss: tf.Tensor(0.38810384, shape=(), dtype=float32)\n",
            "epoch: 4 step: 148 train batch loss: tf.Tensor(0.34608582, shape=(), dtype=float32)\n",
            "epoch: 4 step: 149 train batch loss: tf.Tensor(0.40664014, shape=(), dtype=float32)\n",
            "epoch: 4 step: 150 train batch loss: tf.Tensor(0.32517138, shape=(), dtype=float32)\n",
            "epoch: 4 step: 151 train batch loss: tf.Tensor(0.3507424, shape=(), dtype=float32)\n",
            "epoch: 4 step: 152 train batch loss: tf.Tensor(0.36191028, shape=(), dtype=float32)\n",
            "epoch: 4 step: 153 train batch loss: tf.Tensor(0.337258, shape=(), dtype=float32)\n",
            "epoch: 4 step: 154 train batch loss: tf.Tensor(0.40008205, shape=(), dtype=float32)\n",
            "epoch: 4 step: 155 train batch loss: tf.Tensor(0.36494607, shape=(), dtype=float32)\n",
            "epoch: 4 step: 156 train batch loss: tf.Tensor(0.32182088, shape=(), dtype=float32)\n",
            "epoch: 4 step: 157 train batch loss: tf.Tensor(0.3500321, shape=(), dtype=float32)\n",
            "epoch: 4 step: 158 train batch loss: tf.Tensor(0.4266309, shape=(), dtype=float32)\n",
            "epoch: 4 step: 159 train batch loss: tf.Tensor(0.4341228, shape=(), dtype=float32)\n",
            "epoch: 4 step: 160 train batch loss: tf.Tensor(0.45792955, shape=(), dtype=float32)\n",
            "epoch: 4 step: 161 train batch loss: tf.Tensor(0.37878364, shape=(), dtype=float32)\n",
            "epoch: 4 step: 162 train batch loss: tf.Tensor(0.38114712, shape=(), dtype=float32)\n",
            "epoch: 4 step: 163 train batch loss: tf.Tensor(0.38490364, shape=(), dtype=float32)\n",
            "epoch: 4 step: 164 train batch loss: tf.Tensor(0.40354306, shape=(), dtype=float32)\n",
            "epoch: 4 step: 165 train batch loss: tf.Tensor(0.42425984, shape=(), dtype=float32)\n",
            "epoch: 4 step: 166 train batch loss: tf.Tensor(0.38824606, shape=(), dtype=float32)\n",
            "epoch: 4 step: 167 train batch loss: tf.Tensor(0.33953387, shape=(), dtype=float32)\n",
            "epoch: 4 step: 168 train batch loss: tf.Tensor(0.38800973, shape=(), dtype=float32)\n",
            "epoch: 4 step: 169 train batch loss: tf.Tensor(0.36071324, shape=(), dtype=float32)\n",
            "epoch: 4 step: 170 train batch loss: tf.Tensor(0.42603016, shape=(), dtype=float32)\n",
            "epoch: 4 step: 171 train batch loss: tf.Tensor(0.3693087, shape=(), dtype=float32)\n",
            "epoch: 4 step: 172 train batch loss: tf.Tensor(0.35688248, shape=(), dtype=float32)\n",
            "epoch: 4 step: 173 train batch loss: tf.Tensor(0.37243038, shape=(), dtype=float32)\n",
            "epoch: 4 step: 174 train batch loss: tf.Tensor(0.317674, shape=(), dtype=float32)\n",
            "epoch: 4 step: 175 train batch loss: tf.Tensor(0.46703482, shape=(), dtype=float32)\n",
            "epoch: 4 step: 176 train batch loss: tf.Tensor(0.3847329, shape=(), dtype=float32)\n",
            "epoch: 4 step: 177 train batch loss: tf.Tensor(0.42729908, shape=(), dtype=float32)\n",
            "epoch: 4 step: 178 train batch loss: tf.Tensor(0.3498198, shape=(), dtype=float32)\n",
            "epoch: 4 step: 179 train batch loss: tf.Tensor(0.34123105, shape=(), dtype=float32)\n",
            "epoch: 4 step: 180 train batch loss: tf.Tensor(0.3710282, shape=(), dtype=float32)\n",
            "epoch: 4 step: 181 train batch loss: tf.Tensor(0.36194384, shape=(), dtype=float32)\n",
            "epoch: 4 step: 182 train batch loss: tf.Tensor(0.3272647, shape=(), dtype=float32)\n",
            "epoch: 4 step: 183 train batch loss: tf.Tensor(0.29090574, shape=(), dtype=float32)\n",
            "epoch: 4 step: 184 train batch loss: tf.Tensor(0.3370479, shape=(), dtype=float32)\n",
            "epoch: 4 step: 185 train batch loss: tf.Tensor(0.4074456, shape=(), dtype=float32)\n",
            "epoch: 4 step: 186 train batch loss: tf.Tensor(0.35481203, shape=(), dtype=float32)\n",
            "epoch: 4 step: 187 train batch loss: tf.Tensor(0.36647767, shape=(), dtype=float32)\n",
            "epoch: 4 step: 188 train batch loss: tf.Tensor(0.3595851, shape=(), dtype=float32)\n",
            "epoch: 4 step: 189 train batch loss: tf.Tensor(0.39377505, shape=(), dtype=float32)\n",
            "epoch: 4 step: 190 train batch loss: tf.Tensor(0.42908806, shape=(), dtype=float32)\n",
            "epoch: 4 step: 191 train batch loss: tf.Tensor(0.31372052, shape=(), dtype=float32)\n",
            "epoch: 4 step: 192 train batch loss: tf.Tensor(0.4188834, shape=(), dtype=float32)\n",
            "epoch: 4 step: 193 train batch loss: tf.Tensor(0.3913316, shape=(), dtype=float32)\n",
            "epoch: 4 step: 194 train batch loss: tf.Tensor(0.30659643, shape=(), dtype=float32)\n",
            "epoch: 4 step: 195 train batch loss: tf.Tensor(0.4521141, shape=(), dtype=float32)\n",
            "epoch: 4 step: 196 train batch loss: tf.Tensor(0.37330025, shape=(), dtype=float32)\n",
            "epoch: 4 step: 197 train batch loss: tf.Tensor(0.43028158, shape=(), dtype=float32)\n",
            "epoch: 4 step: 198 train batch loss: tf.Tensor(0.39704186, shape=(), dtype=float32)\n",
            "epoch: 4 step: 199 train batch loss: tf.Tensor(0.33246437, shape=(), dtype=float32)\n",
            "epoch: 4 step: 200 train batch loss: tf.Tensor(0.36448738, shape=(), dtype=float32)\n",
            "epoch: 4 step: 201 train batch loss: tf.Tensor(0.35120517, shape=(), dtype=float32)\n",
            "epoch: 4 step: 202 train batch loss: tf.Tensor(0.6448258, shape=(), dtype=float32)\n",
            "epoch: 4 step: 203 train batch loss: tf.Tensor(0.35033506, shape=(), dtype=float32)\n",
            "epoch: 4 step: 204 train batch loss: tf.Tensor(0.40632588, shape=(), dtype=float32)\n",
            "epoch: 4 step: 205 train batch loss: tf.Tensor(0.36673656, shape=(), dtype=float32)\n",
            "epoch: 4 step: 206 train batch loss: tf.Tensor(0.43545875, shape=(), dtype=float32)\n",
            "epoch: 4 step: 207 train batch loss: tf.Tensor(0.38875467, shape=(), dtype=float32)\n",
            "epoch: 4 step: 208 train batch loss: tf.Tensor(0.39212638, shape=(), dtype=float32)\n",
            "epoch: 4 step: 209 train batch loss: tf.Tensor(0.44305858, shape=(), dtype=float32)\n",
            "epoch: 4 step: 210 train batch loss: tf.Tensor(0.4194695, shape=(), dtype=float32)\n",
            "epoch: 4 step: 211 train batch loss: tf.Tensor(0.3536219, shape=(), dtype=float32)\n",
            "epoch: 4 step: 212 train batch loss: tf.Tensor(0.31505382, shape=(), dtype=float32)\n",
            "epoch: 4 step: 213 train batch loss: tf.Tensor(0.3717906, shape=(), dtype=float32)\n",
            "epoch: 4 step: 214 train batch loss: tf.Tensor(0.41621298, shape=(), dtype=float32)\n",
            "epoch: 4 step: 215 train batch loss: tf.Tensor(0.29595086, shape=(), dtype=float32)\n",
            "epoch: 4 step: 216 train batch loss: tf.Tensor(0.55979776, shape=(), dtype=float32)\n",
            "epoch: 4 step: 217 train batch loss: tf.Tensor(0.3267981, shape=(), dtype=float32)\n",
            "epoch: 4 step: 218 train batch loss: tf.Tensor(0.34402278, shape=(), dtype=float32)\n",
            "epoch: 4 step: 219 train batch loss: tf.Tensor(0.35304534, shape=(), dtype=float32)\n",
            "epoch: 4 step: 220 train batch loss: tf.Tensor(0.36037317, shape=(), dtype=float32)\n",
            "epoch: 4 step: 221 train batch loss: tf.Tensor(0.3407542, shape=(), dtype=float32)\n",
            "epoch: 4 step: 222 train batch loss: tf.Tensor(0.3533873, shape=(), dtype=float32)\n",
            "epoch: 4 step: 223 train batch loss: tf.Tensor(0.37041208, shape=(), dtype=float32)\n",
            "epoch: 4 step: 224 train batch loss: tf.Tensor(0.44220525, shape=(), dtype=float32)\n",
            "epoch: 4 step: 225 train batch loss: tf.Tensor(0.36671343, shape=(), dtype=float32)\n",
            "epoch: 4 step: 226 train batch loss: tf.Tensor(0.2837323, shape=(), dtype=float32)\n",
            "epoch: 4 step: 227 train batch loss: tf.Tensor(0.35071474, shape=(), dtype=float32)\n",
            "epoch: 4 step: 228 train batch loss: tf.Tensor(0.38126034, shape=(), dtype=float32)\n",
            "epoch: 4 step: 229 train batch loss: tf.Tensor(0.40145212, shape=(), dtype=float32)\n",
            "epoch: 4 step: 230 train batch loss: tf.Tensor(0.29235616, shape=(), dtype=float32)\n",
            "epoch: 4 step: 231 train batch loss: tf.Tensor(0.31390038, shape=(), dtype=float32)\n",
            "epoch: 4 step: 232 train batch loss: tf.Tensor(0.4136799, shape=(), dtype=float32)\n",
            "epoch: 4 step: 233 train batch loss: tf.Tensor(0.49849185, shape=(), dtype=float32)\n",
            "epoch: 4 step: 234 train batch loss: tf.Tensor(0.40231293, shape=(), dtype=float32)\n",
            "epoch: 4 step: 235 train batch loss: tf.Tensor(0.32203037, shape=(), dtype=float32)\n",
            "epoch: 4 step: 236 train batch loss: tf.Tensor(0.42175084, shape=(), dtype=float32)\n",
            "epoch: 4 step: 237 train batch loss: tf.Tensor(0.4509127, shape=(), dtype=float32)\n",
            "epoch: 4 step: 238 train batch loss: tf.Tensor(0.41509223, shape=(), dtype=float32)\n",
            "epoch: 4 step: 239 train batch loss: tf.Tensor(0.38516584, shape=(), dtype=float32)\n",
            "epoch: 4 step: 240 train batch loss: tf.Tensor(0.4211622, shape=(), dtype=float32)\n",
            "epoch: 4 step: 241 train batch loss: tf.Tensor(0.4738922, shape=(), dtype=float32)\n",
            "epoch: 4 step: 242 train batch loss: tf.Tensor(0.33218476, shape=(), dtype=float32)\n",
            "epoch: 4 step: 243 train batch loss: tf.Tensor(0.29949835, shape=(), dtype=float32)\n",
            "epoch: 4 step: 244 train batch loss: tf.Tensor(0.43048343, shape=(), dtype=float32)\n",
            "epoch: 4 step: 245 train batch loss: tf.Tensor(0.3667088, shape=(), dtype=float32)\n",
            "epoch: 4 step: 246 train batch loss: tf.Tensor(0.29470223, shape=(), dtype=float32)\n",
            "epoch: 4 step: 247 train batch loss: tf.Tensor(0.39548367, shape=(), dtype=float32)\n",
            "epoch: 4 step: 248 train batch loss: tf.Tensor(0.4308919, shape=(), dtype=float32)\n",
            "epoch: 4 step: 249 train batch loss: tf.Tensor(0.3406183, shape=(), dtype=float32)\n",
            "epoch: 4 step: 250 train batch loss: tf.Tensor(0.38920954, shape=(), dtype=float32)\n",
            "epoch: 4 step: 251 train batch loss: tf.Tensor(0.39597067, shape=(), dtype=float32)\n",
            "epoch: 4 step: 252 train batch loss: tf.Tensor(0.4988245, shape=(), dtype=float32)\n",
            "epoch: 4 step: 253 train batch loss: tf.Tensor(0.32448295, shape=(), dtype=float32)\n",
            "epoch: 4 step: 254 train batch loss: tf.Tensor(0.38046488, shape=(), dtype=float32)\n",
            "epoch: 4 step: 255 train batch loss: tf.Tensor(0.43356553, shape=(), dtype=float32)\n",
            "epoch: 4 step: 256 train batch loss: tf.Tensor(0.3383573, shape=(), dtype=float32)\n",
            "epoch: 4 step: 257 train batch loss: tf.Tensor(0.33883724, shape=(), dtype=float32)\n",
            "epoch: 4 step: 258 train batch loss: tf.Tensor(0.31425127, shape=(), dtype=float32)\n",
            "epoch: 4 step: 259 train batch loss: tf.Tensor(0.33541006, shape=(), dtype=float32)\n",
            "epoch: 4 step: 260 train batch loss: tf.Tensor(0.425775, shape=(), dtype=float32)\n",
            "epoch: 4 step: 261 train batch loss: tf.Tensor(0.36596853, shape=(), dtype=float32)\n",
            "epoch: 4 step: 262 train batch loss: tf.Tensor(0.47304478, shape=(), dtype=float32)\n",
            "epoch: 4 step: 263 train batch loss: tf.Tensor(0.40181798, shape=(), dtype=float32)\n",
            "epoch: 4 step: 264 train batch loss: tf.Tensor(0.40215787, shape=(), dtype=float32)\n",
            "epoch: 4 step: 265 train batch loss: tf.Tensor(0.3886189, shape=(), dtype=float32)\n",
            "epoch: 4 step: 266 train batch loss: tf.Tensor(0.39582145, shape=(), dtype=float32)\n",
            "epoch: 4 step: 267 train batch loss: tf.Tensor(0.33505148, shape=(), dtype=float32)\n",
            "epoch: 4 step: 268 train batch loss: tf.Tensor(0.33058518, shape=(), dtype=float32)\n",
            "epoch: 4 step: 269 train batch loss: tf.Tensor(0.36644152, shape=(), dtype=float32)\n",
            "epoch: 4 step: 270 train batch loss: tf.Tensor(0.37955612, shape=(), dtype=float32)\n",
            "epoch: 4 step: 271 train batch loss: tf.Tensor(0.41726065, shape=(), dtype=float32)\n",
            "epoch: 4 step: 272 train batch loss: tf.Tensor(0.32265624, shape=(), dtype=float32)\n",
            "epoch: 4 step: 273 train batch loss: tf.Tensor(0.34785238, shape=(), dtype=float32)\n",
            "epoch: 4 step: 274 train batch loss: tf.Tensor(0.5915766, shape=(), dtype=float32)\n",
            "epoch: 4 step: 275 train batch loss: tf.Tensor(0.325146, shape=(), dtype=float32)\n",
            "epoch: 4 step: 276 train batch loss: tf.Tensor(0.48412672, shape=(), dtype=float32)\n",
            "epoch: 4 step: 277 train batch loss: tf.Tensor(0.41189775, shape=(), dtype=float32)\n",
            "epoch: 4 step: 278 train batch loss: tf.Tensor(0.45466608, shape=(), dtype=float32)\n",
            "epoch: 4 step: 279 train batch loss: tf.Tensor(0.3485574, shape=(), dtype=float32)\n",
            "epoch: 4 step: 280 train batch loss: tf.Tensor(0.35743448, shape=(), dtype=float32)\n",
            "epoch: 4 step: 281 train batch loss: tf.Tensor(0.43844122, shape=(), dtype=float32)\n",
            "epoch: 4 step: 282 train batch loss: tf.Tensor(0.35763383, shape=(), dtype=float32)\n",
            "epoch: 4 step: 283 train batch loss: tf.Tensor(0.37189627, shape=(), dtype=float32)\n",
            "epoch: 4 step: 284 train batch loss: tf.Tensor(0.3409254, shape=(), dtype=float32)\n",
            "epoch: 4 step: 285 train batch loss: tf.Tensor(0.37658763, shape=(), dtype=float32)\n",
            "epoch: 4 step: 286 train batch loss: tf.Tensor(0.32337523, shape=(), dtype=float32)\n",
            "epoch: 4 step: 287 train batch loss: tf.Tensor(0.36303994, shape=(), dtype=float32)\n",
            "epoch: 4 step: 288 train batch loss: tf.Tensor(0.36357597, shape=(), dtype=float32)\n",
            "epoch: 4 step: 289 train batch loss: tf.Tensor(0.29271698, shape=(), dtype=float32)\n",
            "epoch: 4 step: 290 train batch loss: tf.Tensor(0.4302354, shape=(), dtype=float32)\n",
            "epoch: 4 step: 291 train batch loss: tf.Tensor(0.48816484, shape=(), dtype=float32)\n",
            "epoch: 4 step: 292 train batch loss: tf.Tensor(0.47248998, shape=(), dtype=float32)\n",
            "epoch: 4 step: 293 train batch loss: tf.Tensor(0.46171921, shape=(), dtype=float32)\n",
            "epoch: 4 step: 294 train batch loss: tf.Tensor(0.36175823, shape=(), dtype=float32)\n",
            "epoch: 4 step: 295 train batch loss: tf.Tensor(0.40442973, shape=(), dtype=float32)\n",
            "epoch: 4 step: 296 train batch loss: tf.Tensor(0.32561022, shape=(), dtype=float32)\n",
            "epoch: 4 step: 297 train batch loss: tf.Tensor(0.38458657, shape=(), dtype=float32)\n",
            "epoch: 4 step: 298 train batch loss: tf.Tensor(0.34832165, shape=(), dtype=float32)\n",
            "epoch: 4 step: 299 train batch loss: tf.Tensor(0.34830588, shape=(), dtype=float32)\n",
            "epoch: 4 step: 300 train batch loss: tf.Tensor(0.39282602, shape=(), dtype=float32)\n",
            "epoch: 4 step: 301 train batch loss: tf.Tensor(0.34421146, shape=(), dtype=float32)\n",
            "epoch: 4 step: 302 train batch loss: tf.Tensor(0.30132383, shape=(), dtype=float32)\n",
            "epoch: 4 step: 303 train batch loss: tf.Tensor(0.37450656, shape=(), dtype=float32)\n",
            "epoch: 4 step: 304 train batch loss: tf.Tensor(0.41686186, shape=(), dtype=float32)\n",
            "epoch: 4 step: 305 train batch loss: tf.Tensor(0.39596465, shape=(), dtype=float32)\n",
            "epoch: 4 step: 306 train batch loss: tf.Tensor(0.35053974, shape=(), dtype=float32)\n",
            "epoch: 4 step: 307 train batch loss: tf.Tensor(0.3333295, shape=(), dtype=float32)\n",
            "epoch: 4 step: 308 train batch loss: tf.Tensor(0.34777683, shape=(), dtype=float32)\n",
            "epoch: 4 step: 309 train batch loss: tf.Tensor(0.3756469, shape=(), dtype=float32)\n",
            "epoch: 4 step: 310 train batch loss: tf.Tensor(0.4250517, shape=(), dtype=float32)\n",
            "epoch: 4 step: 311 train batch loss: tf.Tensor(0.4178248, shape=(), dtype=float32)\n",
            "epoch: 4 step: 312 train batch loss: tf.Tensor(0.41562384, shape=(), dtype=float32)\n",
            "epoch: 4 step: 313 train batch loss: tf.Tensor(0.35378814, shape=(), dtype=float32)\n",
            "epoch: 4 step: 314 train batch loss: tf.Tensor(0.38235795, shape=(), dtype=float32)\n",
            "epoch: 4 step: 315 train batch loss: tf.Tensor(0.3422427, shape=(), dtype=float32)\n",
            "epoch: 4 step: 316 train batch loss: tf.Tensor(0.31517857, shape=(), dtype=float32)\n",
            "epoch: 4 step: 317 train batch loss: tf.Tensor(0.36570212, shape=(), dtype=float32)\n",
            "epoch: 4 step: 318 train batch loss: tf.Tensor(0.33529273, shape=(), dtype=float32)\n",
            "epoch: 4 step: 319 train batch loss: tf.Tensor(0.38333538, shape=(), dtype=float32)\n",
            "epoch: 4 step: 320 train batch loss: tf.Tensor(0.3517223, shape=(), dtype=float32)\n",
            "epoch: 4 step: 321 train batch loss: tf.Tensor(0.3622288, shape=(), dtype=float32)\n",
            "epoch: 4 step: 322 train batch loss: tf.Tensor(0.3438378, shape=(), dtype=float32)\n",
            "epoch: 4 step: 323 train batch loss: tf.Tensor(0.3745665, shape=(), dtype=float32)\n",
            "epoch: 4 step: 324 train batch loss: tf.Tensor(0.44496506, shape=(), dtype=float32)\n",
            "epoch: 4 step: 325 train batch loss: tf.Tensor(0.4474082, shape=(), dtype=float32)\n",
            "epoch: 4 step: 326 train batch loss: tf.Tensor(0.3584187, shape=(), dtype=float32)\n",
            "epoch: 4 step: 327 train batch loss: tf.Tensor(0.49021554, shape=(), dtype=float32)\n",
            "epoch: 4 step: 328 train batch loss: tf.Tensor(0.3332039, shape=(), dtype=float32)\n",
            "epoch: 4 step: 329 train batch loss: tf.Tensor(0.390258, shape=(), dtype=float32)\n",
            "epoch: 4 step: 330 train batch loss: tf.Tensor(0.3900867, shape=(), dtype=float32)\n",
            "epoch: 4 step: 331 train batch loss: tf.Tensor(0.32106513, shape=(), dtype=float32)\n",
            "epoch: 4 step: 332 train batch loss: tf.Tensor(0.38999668, shape=(), dtype=float32)\n",
            "epoch: 4 step: 333 train batch loss: tf.Tensor(0.4699758, shape=(), dtype=float32)\n",
            "epoch: 4 step: 334 train batch loss: tf.Tensor(0.35513335, shape=(), dtype=float32)\n",
            "epoch: 4 step: 335 train batch loss: tf.Tensor(0.3337004, shape=(), dtype=float32)\n",
            "epoch: 4 step: 336 train batch loss: tf.Tensor(0.50774914, shape=(), dtype=float32)\n",
            "epoch: 4 step: 337 train batch loss: tf.Tensor(0.29007405, shape=(), dtype=float32)\n",
            "epoch: 4 step: 338 train batch loss: tf.Tensor(0.3259607, shape=(), dtype=float32)\n",
            "epoch: 4 step: 339 train batch loss: tf.Tensor(0.35198003, shape=(), dtype=float32)\n",
            "epoch: 4 step: 340 train batch loss: tf.Tensor(0.36678022, shape=(), dtype=float32)\n",
            "epoch: 4 step: 341 train batch loss: tf.Tensor(0.3342572, shape=(), dtype=float32)\n",
            "epoch: 4 step: 342 train batch loss: tf.Tensor(0.31839335, shape=(), dtype=float32)\n",
            "epoch: 4 step: 343 train batch loss: tf.Tensor(0.279981, shape=(), dtype=float32)\n",
            "epoch: 4 step: 344 train batch loss: tf.Tensor(0.3516619, shape=(), dtype=float32)\n",
            "epoch: 4 step: 345 train batch loss: tf.Tensor(0.44251913, shape=(), dtype=float32)\n",
            "epoch: 4 step: 346 train batch loss: tf.Tensor(0.3654702, shape=(), dtype=float32)\n",
            "epoch: 4 step: 347 train batch loss: tf.Tensor(0.3995372, shape=(), dtype=float32)\n",
            "epoch: 4 step: 348 train batch loss: tf.Tensor(0.30730814, shape=(), dtype=float32)\n",
            "epoch: 4 step: 349 train batch loss: tf.Tensor(0.32165965, shape=(), dtype=float32)\n",
            "epoch: 4 step: 350 train batch loss: tf.Tensor(0.40609425, shape=(), dtype=float32)\n",
            "epoch: 4 step: 351 train batch loss: tf.Tensor(0.27830505, shape=(), dtype=float32)\n",
            "epoch: 4 step: 352 train batch loss: tf.Tensor(0.36587173, shape=(), dtype=float32)\n",
            "epoch: 4 step: 353 train batch loss: tf.Tensor(0.44416714, shape=(), dtype=float32)\n",
            "epoch: 4 step: 354 train batch loss: tf.Tensor(0.31175265, shape=(), dtype=float32)\n",
            "epoch: 4 step: 355 train batch loss: tf.Tensor(0.3305506, shape=(), dtype=float32)\n",
            "epoch: 4 step: 356 train batch loss: tf.Tensor(0.38010404, shape=(), dtype=float32)\n",
            "epoch: 4 step: 357 train batch loss: tf.Tensor(0.3878259, shape=(), dtype=float32)\n",
            "epoch: 4 step: 358 train batch loss: tf.Tensor(0.41829315, shape=(), dtype=float32)\n",
            "epoch: 4 step: 359 train batch loss: tf.Tensor(0.36341998, shape=(), dtype=float32)\n",
            "epoch: 4 step: 360 train batch loss: tf.Tensor(0.43284422, shape=(), dtype=float32)\n",
            "epoch: 4 step: 361 train batch loss: tf.Tensor(0.37023, shape=(), dtype=float32)\n",
            "epoch: 4 step: 362 train batch loss: tf.Tensor(0.34333622, shape=(), dtype=float32)\n",
            "epoch: 4 step: 363 train batch loss: tf.Tensor(0.2939758, shape=(), dtype=float32)\n",
            "epoch: 4 step: 364 train batch loss: tf.Tensor(0.32737237, shape=(), dtype=float32)\n",
            "epoch: 4 step: 365 train batch loss: tf.Tensor(0.3300579, shape=(), dtype=float32)\n",
            "epoch: 4 step: 366 train batch loss: tf.Tensor(0.39068574, shape=(), dtype=float32)\n",
            "epoch: 4 step: 367 train batch loss: tf.Tensor(0.32426056, shape=(), dtype=float32)\n",
            "epoch: 4 step: 368 train batch loss: tf.Tensor(0.3538468, shape=(), dtype=float32)\n",
            "epoch: 4 step: 369 train batch loss: tf.Tensor(0.38787675, shape=(), dtype=float32)\n",
            "epoch: 4 step: 370 train batch loss: tf.Tensor(0.33973247, shape=(), dtype=float32)\n",
            "epoch: 4 step: 371 train batch loss: tf.Tensor(0.30912167, shape=(), dtype=float32)\n",
            "epoch: 4 step: 372 train batch loss: tf.Tensor(0.36780563, shape=(), dtype=float32)\n",
            "epoch: 4 step: 373 train batch loss: tf.Tensor(0.36714652, shape=(), dtype=float32)\n",
            "epoch: 4 step: 374 train batch loss: tf.Tensor(0.34476, shape=(), dtype=float32)\n",
            "epoch: 4 step: 375 train batch loss: tf.Tensor(0.33771566, shape=(), dtype=float32)\n",
            "epoch: 4 step: 376 train batch loss: tf.Tensor(0.379422, shape=(), dtype=float32)\n",
            "epoch: 4 step: 377 train batch loss: tf.Tensor(0.39784226, shape=(), dtype=float32)\n",
            "epoch: 4 step: 378 train batch loss: tf.Tensor(0.3537108, shape=(), dtype=float32)\n",
            "epoch: 4 step: 379 train batch loss: tf.Tensor(0.40137133, shape=(), dtype=float32)\n",
            "epoch: 4 step: 380 train batch loss: tf.Tensor(0.33155233, shape=(), dtype=float32)\n",
            "epoch: 4 step: 381 train batch loss: tf.Tensor(0.44694513, shape=(), dtype=float32)\n",
            "epoch: 4 step: 382 train batch loss: tf.Tensor(0.34751844, shape=(), dtype=float32)\n",
            "epoch: 4 step: 383 train batch loss: tf.Tensor(0.4525249, shape=(), dtype=float32)\n",
            "epoch: 4 step: 384 train batch loss: tf.Tensor(0.33449194, shape=(), dtype=float32)\n",
            "epoch: 4 step: 385 train batch loss: tf.Tensor(0.3651169, shape=(), dtype=float32)\n",
            "epoch: 4 step: 386 train batch loss: tf.Tensor(0.4047748, shape=(), dtype=float32)\n",
            "epoch: 4 step: 387 train batch loss: tf.Tensor(0.26720282, shape=(), dtype=float32)\n",
            "epoch: 4 step: 388 train batch loss: tf.Tensor(0.35302982, shape=(), dtype=float32)\n",
            "epoch: 4 step: 389 train batch loss: tf.Tensor(0.35100514, shape=(), dtype=float32)\n",
            "epoch: 4 step: 390 train batch loss: tf.Tensor(0.4484562, shape=(), dtype=float32)\n",
            "epoch: 4 step: 391 train batch loss: tf.Tensor(0.39035583, shape=(), dtype=float32)\n",
            "epoch: 4 step: 392 train batch loss: tf.Tensor(0.44477963, shape=(), dtype=float32)\n",
            "epoch: 4 step: 393 train batch loss: tf.Tensor(0.3741148, shape=(), dtype=float32)\n",
            "epoch: 4 step: 394 train batch loss: tf.Tensor(0.39971378, shape=(), dtype=float32)\n",
            "epoch: 4 step: 395 train batch loss: tf.Tensor(0.42510644, shape=(), dtype=float32)\n",
            "epoch: 4 step: 396 train batch loss: tf.Tensor(0.36392936, shape=(), dtype=float32)\n",
            "epoch: 4 step: 397 train batch loss: tf.Tensor(0.37900764, shape=(), dtype=float32)\n",
            "epoch: 4 step: 398 train batch loss: tf.Tensor(0.31555927, shape=(), dtype=float32)\n",
            "epoch: 4 step: 399 train batch loss: tf.Tensor(0.43334183, shape=(), dtype=float32)\n",
            "epoch: 4 step: 400 train batch loss: tf.Tensor(0.35237762, shape=(), dtype=float32)\n",
            "epoch: 4 step: 401 train batch loss: tf.Tensor(0.32230338, shape=(), dtype=float32)\n",
            "epoch: 4 step: 402 train batch loss: tf.Tensor(0.42180744, shape=(), dtype=float32)\n",
            "epoch: 4 step: 403 train batch loss: tf.Tensor(0.322642, shape=(), dtype=float32)\n",
            "epoch: 4 step: 404 train batch loss: tf.Tensor(0.36634356, shape=(), dtype=float32)\n",
            "epoch: 4 step: 405 train batch loss: tf.Tensor(0.31005922, shape=(), dtype=float32)\n",
            "epoch: 4 step: 406 train batch loss: tf.Tensor(0.27305022, shape=(), dtype=float32)\n",
            "epoch: 4 step: 407 train batch loss: tf.Tensor(0.28839892, shape=(), dtype=float32)\n",
            "epoch: 4 step: 408 train batch loss: tf.Tensor(0.33440897, shape=(), dtype=float32)\n",
            "epoch: 4 step: 409 train batch loss: tf.Tensor(0.38978228, shape=(), dtype=float32)\n",
            "epoch: 4 step: 410 train batch loss: tf.Tensor(0.33819, shape=(), dtype=float32)\n",
            "epoch: 4 step: 411 train batch loss: tf.Tensor(0.39990032, shape=(), dtype=float32)\n",
            "epoch: 4 step: 412 train batch loss: tf.Tensor(0.35244006, shape=(), dtype=float32)\n",
            "epoch: 4 step: 413 train batch loss: tf.Tensor(0.33992207, shape=(), dtype=float32)\n",
            "epoch: 4 step: 414 train batch loss: tf.Tensor(0.44684187, shape=(), dtype=float32)\n",
            "epoch: 4 step: 415 train batch loss: tf.Tensor(0.3517507, shape=(), dtype=float32)\n",
            "epoch: 4 step: 416 train batch loss: tf.Tensor(0.33975983, shape=(), dtype=float32)\n",
            "epoch: 4 step: 417 train batch loss: tf.Tensor(0.3808729, shape=(), dtype=float32)\n",
            "epoch: 4 step: 418 train batch loss: tf.Tensor(0.32507437, shape=(), dtype=float32)\n",
            "epoch: 4 step: 419 train batch loss: tf.Tensor(0.3534348, shape=(), dtype=float32)\n",
            "epoch: 4 step: 420 train batch loss: tf.Tensor(0.37973163, shape=(), dtype=float32)\n",
            "epoch: 4 step: 421 train batch loss: tf.Tensor(0.35238966, shape=(), dtype=float32)\n",
            "epoch: 4 step: 422 train batch loss: tf.Tensor(0.3160044, shape=(), dtype=float32)\n",
            "epoch: 4 step: 423 train batch loss: tf.Tensor(0.32291895, shape=(), dtype=float32)\n",
            "epoch: 4 step: 424 train batch loss: tf.Tensor(0.3884899, shape=(), dtype=float32)\n",
            "epoch: 4 step: 425 train batch loss: tf.Tensor(0.33902287, shape=(), dtype=float32)\n",
            "epoch: 4 step: 426 train batch loss: tf.Tensor(0.3973372, shape=(), dtype=float32)\n",
            "epoch: 4 step: 427 train batch loss: tf.Tensor(0.3202942, shape=(), dtype=float32)\n",
            "epoch: 4 step: 428 train batch loss: tf.Tensor(0.3024219, shape=(), dtype=float32)\n",
            "epoch: 4 step: 429 train batch loss: tf.Tensor(0.3003113, shape=(), dtype=float32)\n",
            "epoch: 4 step: 430 train batch loss: tf.Tensor(0.3828555, shape=(), dtype=float32)\n",
            "epoch: 4 step: 431 train batch loss: tf.Tensor(0.3774517, shape=(), dtype=float32)\n",
            "epoch: 4 step: 432 train batch loss: tf.Tensor(0.39783472, shape=(), dtype=float32)\n",
            "epoch: 4 step: 433 train batch loss: tf.Tensor(0.41498154, shape=(), dtype=float32)\n",
            "epoch: 4 step: 434 train batch loss: tf.Tensor(0.3617173, shape=(), dtype=float32)\n",
            "epoch: 4 step: 435 train batch loss: tf.Tensor(0.44429427, shape=(), dtype=float32)\n",
            "epoch: 4 step: 436 train batch loss: tf.Tensor(0.34603262, shape=(), dtype=float32)\n",
            "epoch: 4 step: 437 train batch loss: tf.Tensor(0.53035545, shape=(), dtype=float32)\n",
            "epoch: 4 step: 438 train batch loss: tf.Tensor(0.32890713, shape=(), dtype=float32)\n",
            "epoch: 4 step: 439 train batch loss: tf.Tensor(0.38667983, shape=(), dtype=float32)\n",
            "epoch: 4 step: 440 train batch loss: tf.Tensor(0.33754417, shape=(), dtype=float32)\n",
            "epoch: 4 step: 441 train batch loss: tf.Tensor(0.47438064, shape=(), dtype=float32)\n",
            "epoch: 4 step: 442 train batch loss: tf.Tensor(0.46256775, shape=(), dtype=float32)\n",
            "epoch: 4 step: 443 train batch loss: tf.Tensor(0.29381743, shape=(), dtype=float32)\n",
            "epoch: 4 step: 444 train batch loss: tf.Tensor(0.37444907, shape=(), dtype=float32)\n",
            "epoch: 4 step: 445 train batch loss: tf.Tensor(0.44768986, shape=(), dtype=float32)\n",
            "epoch: 4 step: 446 train batch loss: tf.Tensor(0.33642063, shape=(), dtype=float32)\n",
            "epoch: 4 step: 447 train batch loss: tf.Tensor(0.38456514, shape=(), dtype=float32)\n",
            "epoch: 4 step: 448 train batch loss: tf.Tensor(0.28927264, shape=(), dtype=float32)\n",
            "epoch: 4 step: 449 train batch loss: tf.Tensor(0.42790365, shape=(), dtype=float32)\n",
            "epoch: 4 step: 450 train batch loss: tf.Tensor(0.32582352, shape=(), dtype=float32)\n",
            "epoch: 4 step: 451 train batch loss: tf.Tensor(0.42744985, shape=(), dtype=float32)\n",
            "epoch: 4 step: 452 train batch loss: tf.Tensor(0.3527267, shape=(), dtype=float32)\n",
            "epoch: 4 step: 453 train batch loss: tf.Tensor(0.357362, shape=(), dtype=float32)\n",
            "epoch: 4 step: 454 train batch loss: tf.Tensor(0.3693755, shape=(), dtype=float32)\n",
            "epoch: 4 step: 455 train batch loss: tf.Tensor(0.40979892, shape=(), dtype=float32)\n",
            "epoch: 4 step: 456 train batch loss: tf.Tensor(0.303975, shape=(), dtype=float32)\n",
            "epoch: 4 step: 457 train batch loss: tf.Tensor(0.45452017, shape=(), dtype=float32)\n",
            "epoch: 4 step: 458 train batch loss: tf.Tensor(0.4156398, shape=(), dtype=float32)\n",
            "epoch: 4 step: 459 train batch loss: tf.Tensor(0.38666773, shape=(), dtype=float32)\n",
            "epoch: 4 step: 460 train batch loss: tf.Tensor(0.31612986, shape=(), dtype=float32)\n",
            "epoch: 4 step: 461 train batch loss: tf.Tensor(0.35863882, shape=(), dtype=float32)\n",
            "epoch: 4 step: 462 train batch loss: tf.Tensor(0.38519633, shape=(), dtype=float32)\n",
            "epoch: 4 step: 463 train batch loss: tf.Tensor(0.373035, shape=(), dtype=float32)\n",
            "epoch: 4 step: 464 train batch loss: tf.Tensor(0.37135246, shape=(), dtype=float32)\n",
            "epoch: 4 step: 465 train batch loss: tf.Tensor(0.34062034, shape=(), dtype=float32)\n",
            "epoch: 4 step: 466 train batch loss: tf.Tensor(0.35575095, shape=(), dtype=float32)\n",
            "epoch: 4 step: 467 train batch loss: tf.Tensor(0.30281037, shape=(), dtype=float32)\n",
            "epoch: 4 step: 468 train batch loss: tf.Tensor(0.41491276, shape=(), dtype=float32)\n",
            "epoch: 4 step: 469 train batch loss: tf.Tensor(0.41212875, shape=(), dtype=float32)\n",
            "epoch: 4 step: 470 train batch loss: tf.Tensor(0.38934958, shape=(), dtype=float32)\n",
            "epoch: 4 step: 471 train batch loss: tf.Tensor(0.37669945, shape=(), dtype=float32)\n",
            "epoch: 4 step: 472 train batch loss: tf.Tensor(0.39444143, shape=(), dtype=float32)\n",
            "epoch: 4 step: 473 train batch loss: tf.Tensor(0.35448605, shape=(), dtype=float32)\n",
            "epoch: 4 step: 474 train batch loss: tf.Tensor(0.38490015, shape=(), dtype=float32)\n",
            "epoch: 4 step: 475 train batch loss: tf.Tensor(0.35460383, shape=(), dtype=float32)\n",
            "epoch: 4 step: 476 train batch loss: tf.Tensor(0.36658695, shape=(), dtype=float32)\n",
            "epoch: 4 step: 477 train batch loss: tf.Tensor(0.33367118, shape=(), dtype=float32)\n",
            "epoch: 4 step: 478 train batch loss: tf.Tensor(0.36879843, shape=(), dtype=float32)\n",
            "epoch: 4 step: 479 train batch loss: tf.Tensor(0.3748602, shape=(), dtype=float32)\n",
            "epoch: 4 step: 480 train batch loss: tf.Tensor(0.46233553, shape=(), dtype=float32)\n",
            "epoch: 4 step: 481 train batch loss: tf.Tensor(0.27898645, shape=(), dtype=float32)\n",
            "epoch: 4 step: 482 train batch loss: tf.Tensor(0.36394924, shape=(), dtype=float32)\n",
            "epoch: 4 step: 483 train batch loss: tf.Tensor(0.36917958, shape=(), dtype=float32)\n",
            "epoch: 4 step: 484 train batch loss: tf.Tensor(0.29971442, shape=(), dtype=float32)\n",
            "epoch: 4 step: 485 train batch loss: tf.Tensor(0.4249831, shape=(), dtype=float32)\n",
            "epoch: 4 step: 486 train batch loss: tf.Tensor(0.31964684, shape=(), dtype=float32)\n",
            "epoch: 4 step: 487 train batch loss: tf.Tensor(0.30976233, shape=(), dtype=float32)\n",
            "epoch: 4 step: 488 train batch loss: tf.Tensor(0.36170325, shape=(), dtype=float32)\n",
            "epoch: 4 step: 489 train batch loss: tf.Tensor(0.30515733, shape=(), dtype=float32)\n",
            "epoch: 4 step: 490 train batch loss: tf.Tensor(0.33839205, shape=(), dtype=float32)\n",
            "epoch: 4 step: 491 train batch loss: tf.Tensor(0.3402846, shape=(), dtype=float32)\n",
            "epoch: 4 step: 492 train batch loss: tf.Tensor(0.45323372, shape=(), dtype=float32)\n",
            "epoch: 4 step: 493 train batch loss: tf.Tensor(0.32820377, shape=(), dtype=float32)\n",
            "epoch: 4 step: 494 train batch loss: tf.Tensor(0.32804328, shape=(), dtype=float32)\n",
            "epoch: 4 step: 495 train batch loss: tf.Tensor(0.26501566, shape=(), dtype=float32)\n",
            "epoch: 4 step: 496 train batch loss: tf.Tensor(0.40959173, shape=(), dtype=float32)\n",
            "epoch: 4 step: 497 train batch loss: tf.Tensor(0.40658253, shape=(), dtype=float32)\n",
            "epoch: 4 step: 498 train batch loss: tf.Tensor(0.34025508, shape=(), dtype=float32)\n",
            "epoch: 4 step: 499 train batch loss: tf.Tensor(0.34546673, shape=(), dtype=float32)\n",
            "epoch: 4 step: 500 train batch loss: tf.Tensor(0.41274443, shape=(), dtype=float32)\n",
            "epoch: 4 step: 501 train batch loss: tf.Tensor(0.29955053, shape=(), dtype=float32)\n",
            "epoch: 4 step: 502 train batch loss: tf.Tensor(0.39167613, shape=(), dtype=float32)\n",
            "epoch: 4 step: 503 train batch loss: tf.Tensor(0.3333448, shape=(), dtype=float32)\n",
            "epoch: 4 step: 504 train batch loss: tf.Tensor(0.30275238, shape=(), dtype=float32)\n",
            "epoch: 4 step: 505 train batch loss: tf.Tensor(0.33026198, shape=(), dtype=float32)\n",
            "epoch: 4 step: 506 train batch loss: tf.Tensor(0.34835738, shape=(), dtype=float32)\n",
            "epoch: 4 step: 507 train batch loss: tf.Tensor(0.25978208, shape=(), dtype=float32)\n",
            "epoch: 4 step: 508 train batch loss: tf.Tensor(0.40664458, shape=(), dtype=float32)\n",
            "epoch: 4 step: 509 train batch loss: tf.Tensor(0.36269343, shape=(), dtype=float32)\n",
            "epoch: 4 step: 510 train batch loss: tf.Tensor(0.35169566, shape=(), dtype=float32)\n",
            "epoch: 4 step: 511 train batch loss: tf.Tensor(0.3687833, shape=(), dtype=float32)\n",
            "epoch: 4 step: 512 train batch loss: tf.Tensor(0.3694165, shape=(), dtype=float32)\n",
            "epoch: 4 step: 513 train batch loss: tf.Tensor(0.4301218, shape=(), dtype=float32)\n",
            "epoch: 4 step: 514 train batch loss: tf.Tensor(0.40705538, shape=(), dtype=float32)\n",
            "epoch: 4 step: 515 train batch loss: tf.Tensor(0.3581768, shape=(), dtype=float32)\n",
            "epoch: 4 step: 516 train batch loss: tf.Tensor(0.27705258, shape=(), dtype=float32)\n",
            "epoch: 4 step: 517 train batch loss: tf.Tensor(0.27376252, shape=(), dtype=float32)\n",
            "epoch: 4 step: 518 train batch loss: tf.Tensor(0.39846116, shape=(), dtype=float32)\n",
            "epoch: 4 step: 519 train batch loss: tf.Tensor(0.32778686, shape=(), dtype=float32)\n",
            "epoch: 4 step: 520 train batch loss: tf.Tensor(0.37637287, shape=(), dtype=float32)\n",
            "epoch: 4 step: 521 train batch loss: tf.Tensor(0.34537506, shape=(), dtype=float32)\n",
            "epoch: 4 step: 522 train batch loss: tf.Tensor(0.40115827, shape=(), dtype=float32)\n",
            "epoch: 4 step: 523 train batch loss: tf.Tensor(0.33694276, shape=(), dtype=float32)\n",
            "epoch: 4 step: 524 train batch loss: tf.Tensor(0.42562237, shape=(), dtype=float32)\n",
            "epoch: 4 step: 525 train batch loss: tf.Tensor(0.33856806, shape=(), dtype=float32)\n",
            "epoch: 4 step: 526 train batch loss: tf.Tensor(0.361863, shape=(), dtype=float32)\n",
            "epoch: 4 step: 527 train batch loss: tf.Tensor(0.27493504, shape=(), dtype=float32)\n",
            "epoch: 4 step: 528 train batch loss: tf.Tensor(0.30040404, shape=(), dtype=float32)\n",
            "epoch: 4 step: 529 train batch loss: tf.Tensor(0.3576206, shape=(), dtype=float32)\n",
            "epoch: 4 step: 530 train batch loss: tf.Tensor(0.34947646, shape=(), dtype=float32)\n",
            "epoch: 4 step: 531 train batch loss: tf.Tensor(0.41605553, shape=(), dtype=float32)\n",
            "epoch: 4 step: 532 train batch loss: tf.Tensor(0.3296731, shape=(), dtype=float32)\n",
            "epoch: 4 step: 533 train batch loss: tf.Tensor(0.4153205, shape=(), dtype=float32)\n",
            "epoch: 4 step: 534 train batch loss: tf.Tensor(0.32581943, shape=(), dtype=float32)\n",
            "epoch: 4 step: 535 train batch loss: tf.Tensor(0.26683918, shape=(), dtype=float32)\n",
            "epoch: 4 step: 536 train batch loss: tf.Tensor(0.2956937, shape=(), dtype=float32)\n",
            "epoch: 4 step: 537 train batch loss: tf.Tensor(0.3599626, shape=(), dtype=float32)\n",
            "epoch: 4 step: 538 train batch loss: tf.Tensor(0.29418972, shape=(), dtype=float32)\n",
            "epoch: 4 step: 539 train batch loss: tf.Tensor(0.38226423, shape=(), dtype=float32)\n",
            "epoch: 4 step: 540 train batch loss: tf.Tensor(0.38945654, shape=(), dtype=float32)\n",
            "epoch: 4 step: 541 train batch loss: tf.Tensor(0.31400287, shape=(), dtype=float32)\n",
            "epoch: 4 step: 542 train batch loss: tf.Tensor(0.3801633, shape=(), dtype=float32)\n",
            "epoch: 4 step: 543 train batch loss: tf.Tensor(0.4725575, shape=(), dtype=float32)\n",
            "epoch: 4 step: 544 train batch loss: tf.Tensor(0.3649867, shape=(), dtype=float32)\n",
            "epoch: 4 step: 545 train batch loss: tf.Tensor(0.35714084, shape=(), dtype=float32)\n",
            "epoch: 4 step: 546 train batch loss: tf.Tensor(0.28998935, shape=(), dtype=float32)\n",
            "epoch: 4 step: 547 train batch loss: tf.Tensor(0.2816028, shape=(), dtype=float32)\n",
            "epoch: 4 step: 548 train batch loss: tf.Tensor(0.36737603, shape=(), dtype=float32)\n",
            "epoch: 4 step: 549 train batch loss: tf.Tensor(0.31532747, shape=(), dtype=float32)\n",
            "epoch: 4 step: 550 train batch loss: tf.Tensor(0.3236331, shape=(), dtype=float32)\n",
            "epoch: 4 step: 551 train batch loss: tf.Tensor(0.35426575, shape=(), dtype=float32)\n",
            "epoch: 4 step: 552 train batch loss: tf.Tensor(0.32418856, shape=(), dtype=float32)\n",
            "epoch: 4 step: 553 train batch loss: tf.Tensor(0.42366904, shape=(), dtype=float32)\n",
            "epoch: 4 step: 554 train batch loss: tf.Tensor(0.4626797, shape=(), dtype=float32)\n",
            "epoch: 4 step: 555 train batch loss: tf.Tensor(0.33491433, shape=(), dtype=float32)\n",
            "epoch: 4 step: 556 train batch loss: tf.Tensor(0.38501972, shape=(), dtype=float32)\n",
            "epoch: 4 step: 557 train batch loss: tf.Tensor(0.33263797, shape=(), dtype=float32)\n",
            "epoch: 4 step: 558 train batch loss: tf.Tensor(0.3031107, shape=(), dtype=float32)\n",
            "epoch: 4 step: 559 train batch loss: tf.Tensor(0.3518611, shape=(), dtype=float32)\n",
            "epoch: 4 step: 560 train batch loss: tf.Tensor(0.2885318, shape=(), dtype=float32)\n",
            "epoch: 4 step: 561 train batch loss: tf.Tensor(0.33474514, shape=(), dtype=float32)\n",
            "epoch: 4 step: 562 train batch loss: tf.Tensor(0.3574446, shape=(), dtype=float32)\n",
            "epoch: 4 step: 563 train batch loss: tf.Tensor(0.33669958, shape=(), dtype=float32)\n",
            "epoch: 4 step: 564 train batch loss: tf.Tensor(0.34906316, shape=(), dtype=float32)\n",
            "epoch: 4 step: 565 train batch loss: tf.Tensor(0.45695084, shape=(), dtype=float32)\n",
            "epoch: 4 step: 566 train batch loss: tf.Tensor(0.34726936, shape=(), dtype=float32)\n",
            "epoch: 4 step: 567 train batch loss: tf.Tensor(0.2916297, shape=(), dtype=float32)\n",
            "epoch: 4 step: 568 train batch loss: tf.Tensor(0.3312059, shape=(), dtype=float32)\n",
            "epoch: 4 step: 569 train batch loss: tf.Tensor(0.36780313, shape=(), dtype=float32)\n",
            "epoch: 4 step: 570 train batch loss: tf.Tensor(0.35017103, shape=(), dtype=float32)\n",
            "epoch: 4 step: 571 train batch loss: tf.Tensor(0.35413536, shape=(), dtype=float32)\n",
            "epoch: 4 step: 572 train batch loss: tf.Tensor(0.41528875, shape=(), dtype=float32)\n",
            "epoch: 4 step: 573 train batch loss: tf.Tensor(0.30928326, shape=(), dtype=float32)\n",
            "epoch: 4 step: 574 train batch loss: tf.Tensor(0.34102094, shape=(), dtype=float32)\n",
            "epoch: 4 step: 575 train batch loss: tf.Tensor(0.33343616, shape=(), dtype=float32)\n",
            "epoch: 4 step: 576 train batch loss: tf.Tensor(0.39092204, shape=(), dtype=float32)\n",
            "epoch: 4 step: 577 train batch loss: tf.Tensor(0.31510097, shape=(), dtype=float32)\n",
            "epoch: 4 step: 578 train batch loss: tf.Tensor(0.37930724, shape=(), dtype=float32)\n",
            "epoch: 4 step: 579 train batch loss: tf.Tensor(0.45821825, shape=(), dtype=float32)\n",
            "epoch: 4 step: 580 train batch loss: tf.Tensor(0.33664817, shape=(), dtype=float32)\n",
            "epoch: 4 step: 581 train batch loss: tf.Tensor(0.3371306, shape=(), dtype=float32)\n",
            "epoch: 4 step: 582 train batch loss: tf.Tensor(0.3764812, shape=(), dtype=float32)\n",
            "epoch: 4 step: 583 train batch loss: tf.Tensor(0.31143662, shape=(), dtype=float32)\n",
            "epoch: 4 step: 584 train batch loss: tf.Tensor(0.3474341, shape=(), dtype=float32)\n",
            "epoch: 4 step: 585 train batch loss: tf.Tensor(0.37640873, shape=(), dtype=float32)\n",
            "epoch: 4 step: 586 train batch loss: tf.Tensor(0.3308639, shape=(), dtype=float32)\n",
            "epoch: 4 step: 587 train batch loss: tf.Tensor(0.30433422, shape=(), dtype=float32)\n",
            "epoch: 4 step: 588 train batch loss: tf.Tensor(0.31570047, shape=(), dtype=float32)\n",
            "epoch: 4 step: 589 train batch loss: tf.Tensor(0.32687536, shape=(), dtype=float32)\n",
            "epoch: 4 step: 590 train batch loss: tf.Tensor(0.34071046, shape=(), dtype=float32)\n",
            "epoch: 4 step: 591 train batch loss: tf.Tensor(0.3437758, shape=(), dtype=float32)\n",
            "epoch: 4 step: 592 train batch loss: tf.Tensor(0.46881762, shape=(), dtype=float32)\n",
            "epoch: 4 step: 593 train batch loss: tf.Tensor(0.33985096, shape=(), dtype=float32)\n",
            "epoch: 4 step: 594 train batch loss: tf.Tensor(0.2989251, shape=(), dtype=float32)\n",
            "epoch: 4 step: 595 train batch loss: tf.Tensor(0.35683468, shape=(), dtype=float32)\n",
            "epoch: 4 step: 596 train batch loss: tf.Tensor(0.36146617, shape=(), dtype=float32)\n",
            "epoch: 4 step: 597 train batch loss: tf.Tensor(0.31075704, shape=(), dtype=float32)\n",
            "epoch: 4 step: 598 train batch loss: tf.Tensor(0.41937298, shape=(), dtype=float32)\n",
            "epoch: 4 step: 599 train batch loss: tf.Tensor(0.26392227, shape=(), dtype=float32)\n",
            "epoch: 4 step: 600 train batch loss: tf.Tensor(0.42398992, shape=(), dtype=float32)\n",
            "epoch: 4 step: 601 train batch loss: tf.Tensor(0.37065288, shape=(), dtype=float32)\n",
            "epoch: 4 step: 602 train batch loss: tf.Tensor(0.4318018, shape=(), dtype=float32)\n",
            "epoch: 4 step: 603 train batch loss: tf.Tensor(0.40485388, shape=(), dtype=float32)\n",
            "epoch: 4 step: 604 train batch loss: tf.Tensor(0.3563251, shape=(), dtype=float32)\n",
            "epoch: 4 step: 605 train batch loss: tf.Tensor(0.38452914, shape=(), dtype=float32)\n",
            "epoch: 4 step: 606 train batch loss: tf.Tensor(0.31702903, shape=(), dtype=float32)\n",
            "epoch: 4 step: 607 train batch loss: tf.Tensor(0.43571815, shape=(), dtype=float32)\n",
            "epoch: 4 step: 608 train batch loss: tf.Tensor(0.36350623, shape=(), dtype=float32)\n",
            "epoch: 4 step: 609 train batch loss: tf.Tensor(0.3771435, shape=(), dtype=float32)\n",
            "epoch: 4 step: 610 train batch loss: tf.Tensor(0.39586028, shape=(), dtype=float32)\n",
            "epoch: 4 step: 611 train batch loss: tf.Tensor(0.34493935, shape=(), dtype=float32)\n",
            "epoch: 4 step: 612 train batch loss: tf.Tensor(0.39798945, shape=(), dtype=float32)\n",
            "epoch: 4 step: 613 train batch loss: tf.Tensor(0.29917747, shape=(), dtype=float32)\n",
            "epoch: 4 step: 614 train batch loss: tf.Tensor(0.3323159, shape=(), dtype=float32)\n",
            "epoch: 4 step: 615 train batch loss: tf.Tensor(0.37564987, shape=(), dtype=float32)\n",
            "epoch: 4 step: 616 train batch loss: tf.Tensor(0.38160735, shape=(), dtype=float32)\n",
            "epoch: 4 step: 617 train batch loss: tf.Tensor(0.29552457, shape=(), dtype=float32)\n",
            "epoch: 4 step: 618 train batch loss: tf.Tensor(0.2856081, shape=(), dtype=float32)\n",
            "epoch: 4 step: 619 train batch loss: tf.Tensor(0.4027423, shape=(), dtype=float32)\n",
            "epoch: 4 step: 620 train batch loss: tf.Tensor(0.32463935, shape=(), dtype=float32)\n",
            "epoch: 4 step: 621 train batch loss: tf.Tensor(0.2968549, shape=(), dtype=float32)\n",
            "epoch: 4 step: 622 train batch loss: tf.Tensor(0.38221443, shape=(), dtype=float32)\n",
            "epoch: 4 step: 623 train batch loss: tf.Tensor(0.33274385, shape=(), dtype=float32)\n",
            "epoch: 4 step: 624 train batch loss: tf.Tensor(0.33608422, shape=(), dtype=float32)\n",
            "epoch: 4 step: 625 train batch loss: tf.Tensor(0.29824564, shape=(), dtype=float32)\n",
            "epoch: 4 step: 626 train batch loss: tf.Tensor(0.290172, shape=(), dtype=float32)\n",
            "epoch: 4 step: 627 train batch loss: tf.Tensor(0.4145464, shape=(), dtype=float32)\n",
            "epoch: 4 step: 628 train batch loss: tf.Tensor(0.3533925, shape=(), dtype=float32)\n",
            "epoch: 4 step: 629 train batch loss: tf.Tensor(0.36140764, shape=(), dtype=float32)\n",
            "epoch: 4 step: 630 train batch loss: tf.Tensor(0.33633095, shape=(), dtype=float32)\n",
            "epoch: 4 step: 631 train batch loss: tf.Tensor(0.38499624, shape=(), dtype=float32)\n",
            "epoch: 4 step: 632 train batch loss: tf.Tensor(0.3020484, shape=(), dtype=float32)\n",
            "epoch: 4 step: 633 train batch loss: tf.Tensor(0.3060488, shape=(), dtype=float32)\n",
            "epoch: 4 step: 634 train batch loss: tf.Tensor(0.34550616, shape=(), dtype=float32)\n",
            "epoch: 4 step: 635 train batch loss: tf.Tensor(0.408221, shape=(), dtype=float32)\n",
            "epoch: 4 step: 636 train batch loss: tf.Tensor(0.37229276, shape=(), dtype=float32)\n",
            "epoch: 4 step: 637 train batch loss: tf.Tensor(0.3012524, shape=(), dtype=float32)\n",
            "epoch: 4 step: 638 train batch loss: tf.Tensor(0.4203086, shape=(), dtype=float32)\n",
            "epoch: 4 step: 639 train batch loss: tf.Tensor(0.33810705, shape=(), dtype=float32)\n",
            "epoch: 4 step: 640 train batch loss: tf.Tensor(0.3894021, shape=(), dtype=float32)\n",
            "epoch: 4 step: 641 train batch loss: tf.Tensor(0.5018803, shape=(), dtype=float32)\n",
            "epoch: 4 step: 642 train batch loss: tf.Tensor(0.43097496, shape=(), dtype=float32)\n",
            "epoch: 4 step: 643 train batch loss: tf.Tensor(0.36732092, shape=(), dtype=float32)\n",
            "epoch: 4 step: 644 train batch loss: tf.Tensor(0.3451472, shape=(), dtype=float32)\n",
            "epoch: 4 step: 645 train batch loss: tf.Tensor(0.3515498, shape=(), dtype=float32)\n",
            "epoch: 4 step: 646 train batch loss: tf.Tensor(0.31412458, shape=(), dtype=float32)\n",
            "epoch: 4 step: 647 train batch loss: tf.Tensor(0.30433187, shape=(), dtype=float32)\n",
            "epoch: 4 step: 648 train batch loss: tf.Tensor(0.31358695, shape=(), dtype=float32)\n",
            "epoch: 4 step: 649 train batch loss: tf.Tensor(0.31240097, shape=(), dtype=float32)\n",
            "epoch: 4 step: 650 train batch loss: tf.Tensor(0.37557882, shape=(), dtype=float32)\n",
            "epoch: 4 step: 651 train batch loss: tf.Tensor(0.46547192, shape=(), dtype=float32)\n",
            "epoch: 4 step: 652 train batch loss: tf.Tensor(0.39472765, shape=(), dtype=float32)\n",
            "epoch: 4 step: 653 train batch loss: tf.Tensor(0.34740764, shape=(), dtype=float32)\n",
            "epoch: 4 step: 654 train batch loss: tf.Tensor(0.25712788, shape=(), dtype=float32)\n",
            "epoch: 4 step: 655 train batch loss: tf.Tensor(0.43109208, shape=(), dtype=float32)\n",
            "epoch: 4 step: 656 train batch loss: tf.Tensor(0.40931678, shape=(), dtype=float32)\n",
            "epoch: 4 step: 657 train batch loss: tf.Tensor(0.27908048, shape=(), dtype=float32)\n",
            "epoch: 4 step: 658 train batch loss: tf.Tensor(0.34320253, shape=(), dtype=float32)\n",
            "epoch: 4 step: 659 train batch loss: tf.Tensor(0.3491456, shape=(), dtype=float32)\n",
            "epoch: 4 step: 660 train batch loss: tf.Tensor(0.39610624, shape=(), dtype=float32)\n",
            "epoch: 4 step: 661 train batch loss: tf.Tensor(0.32065436, shape=(), dtype=float32)\n",
            "epoch: 4 step: 662 train batch loss: tf.Tensor(0.36522758, shape=(), dtype=float32)\n",
            "epoch: 4 step: 663 train batch loss: tf.Tensor(0.33637565, shape=(), dtype=float32)\n",
            "epoch: 4 step: 664 train batch loss: tf.Tensor(0.33961153, shape=(), dtype=float32)\n",
            "epoch: 4 step: 665 train batch loss: tf.Tensor(0.37907633, shape=(), dtype=float32)\n",
            "epoch: 4 step: 666 train batch loss: tf.Tensor(0.31129116, shape=(), dtype=float32)\n",
            "epoch: 4 step: 667 train batch loss: tf.Tensor(0.34859055, shape=(), dtype=float32)\n",
            "epoch: 4 step: 668 train batch loss: tf.Tensor(0.3663537, shape=(), dtype=float32)\n",
            "epoch: 4 step: 669 train batch loss: tf.Tensor(0.2921825, shape=(), dtype=float32)\n",
            "epoch: 4 step: 670 train batch loss: tf.Tensor(0.37311056, shape=(), dtype=float32)\n",
            "epoch: 4 step: 671 train batch loss: tf.Tensor(0.3153682, shape=(), dtype=float32)\n",
            "epoch: 4 step: 672 train batch loss: tf.Tensor(0.47385597, shape=(), dtype=float32)\n",
            "epoch: 4 step: 673 train batch loss: tf.Tensor(0.2797107, shape=(), dtype=float32)\n",
            "epoch: 4 step: 674 train batch loss: tf.Tensor(0.33183914, shape=(), dtype=float32)\n",
            "epoch: 4 step: 675 train batch loss: tf.Tensor(0.3945145, shape=(), dtype=float32)\n",
            "epoch: 4 step: 676 train batch loss: tf.Tensor(0.24544081, shape=(), dtype=float32)\n",
            "epoch: 4 step: 677 train batch loss: tf.Tensor(0.43028355, shape=(), dtype=float32)\n",
            "epoch: 4 step: 678 train batch loss: tf.Tensor(0.32412514, shape=(), dtype=float32)\n",
            "epoch: 4 step: 679 train batch loss: tf.Tensor(0.31280383, shape=(), dtype=float32)\n",
            "epoch: 4 step: 680 train batch loss: tf.Tensor(0.34770906, shape=(), dtype=float32)\n",
            "epoch: 4 step: 681 train batch loss: tf.Tensor(0.4142085, shape=(), dtype=float32)\n",
            "epoch: 4 step: 682 train batch loss: tf.Tensor(0.34289002, shape=(), dtype=float32)\n",
            "epoch: 4 step: 683 train batch loss: tf.Tensor(0.3448792, shape=(), dtype=float32)\n",
            "epoch: 4 step: 684 train batch loss: tf.Tensor(0.35890228, shape=(), dtype=float32)\n",
            "epoch: 4 step: 685 train batch loss: tf.Tensor(0.38649544, shape=(), dtype=float32)\n",
            "epoch: 4 step: 686 train batch loss: tf.Tensor(0.31883723, shape=(), dtype=float32)\n",
            "epoch: 4 step: 687 train batch loss: tf.Tensor(0.32594123, shape=(), dtype=float32)\n",
            "epoch: 4 step: 688 train batch loss: tf.Tensor(0.33832702, shape=(), dtype=float32)\n",
            "epoch: 4 step: 689 train batch loss: tf.Tensor(0.36567602, shape=(), dtype=float32)\n",
            "epoch: 4 step: 690 train batch loss: tf.Tensor(0.33313254, shape=(), dtype=float32)\n",
            "epoch: 4 step: 691 train batch loss: tf.Tensor(0.25700134, shape=(), dtype=float32)\n",
            "epoch: 4 step: 692 train batch loss: tf.Tensor(0.31946024, shape=(), dtype=float32)\n",
            "epoch: 4 step: 693 train batch loss: tf.Tensor(0.23514763, shape=(), dtype=float32)\n",
            "epoch: 4 step: 694 train batch loss: tf.Tensor(0.27267742, shape=(), dtype=float32)\n",
            "epoch: 4 step: 695 train batch loss: tf.Tensor(0.33652002, shape=(), dtype=float32)\n",
            "epoch: 4 step: 696 train batch loss: tf.Tensor(0.3271948, shape=(), dtype=float32)\n",
            "epoch: 4 step: 697 train batch loss: tf.Tensor(0.3838665, shape=(), dtype=float32)\n",
            "epoch: 4 step: 698 train batch loss: tf.Tensor(0.34354958, shape=(), dtype=float32)\n",
            "epoch: 4 step: 699 train batch loss: tf.Tensor(0.37069598, shape=(), dtype=float32)\n",
            "epoch: 4 step: 700 train batch loss: tf.Tensor(0.31189424, shape=(), dtype=float32)\n",
            "epoch: 4 step: 701 train batch loss: tf.Tensor(0.35362056, shape=(), dtype=float32)\n",
            "epoch: 4 step: 702 train batch loss: tf.Tensor(0.35404524, shape=(), dtype=float32)\n",
            "epoch: 4 step: 703 train batch loss: tf.Tensor(0.32016724, shape=(), dtype=float32)\n",
            "epoch: 4 step: 704 train batch loss: tf.Tensor(0.39537638, shape=(), dtype=float32)\n",
            "epoch: 4 step: 705 train batch loss: tf.Tensor(0.35623068, shape=(), dtype=float32)\n",
            "epoch: 4 step: 706 train batch loss: tf.Tensor(0.32651174, shape=(), dtype=float32)\n",
            "epoch: 4 step: 707 train batch loss: tf.Tensor(0.26834822, shape=(), dtype=float32)\n",
            "epoch: 4 step: 708 train batch loss: tf.Tensor(0.27127224, shape=(), dtype=float32)\n",
            "epoch: 4 step: 709 train batch loss: tf.Tensor(0.32962665, shape=(), dtype=float32)\n",
            "epoch: 4 step: 710 train batch loss: tf.Tensor(0.3724891, shape=(), dtype=float32)\n",
            "epoch: 4 step: 711 train batch loss: tf.Tensor(0.31428275, shape=(), dtype=float32)\n",
            "epoch: 4 step: 712 train batch loss: tf.Tensor(0.36272806, shape=(), dtype=float32)\n",
            "epoch: 4 step: 713 train batch loss: tf.Tensor(0.38984224, shape=(), dtype=float32)\n",
            "epoch: 4 step: 714 train batch loss: tf.Tensor(0.37806082, shape=(), dtype=float32)\n",
            "epoch: 4 step: 715 train batch loss: tf.Tensor(0.34704247, shape=(), dtype=float32)\n",
            "epoch: 4 step: 716 train batch loss: tf.Tensor(0.336602, shape=(), dtype=float32)\n",
            "epoch: 4 step: 717 train batch loss: tf.Tensor(0.35393506, shape=(), dtype=float32)\n",
            "epoch: 4 step: 718 train batch loss: tf.Tensor(0.39405215, shape=(), dtype=float32)\n",
            "epoch: 4 step: 719 train batch loss: tf.Tensor(0.47309846, shape=(), dtype=float32)\n",
            "epoch: 4 step: 720 train batch loss: tf.Tensor(0.35835862, shape=(), dtype=float32)\n",
            "epoch: 4 step: 721 train batch loss: tf.Tensor(0.2951221, shape=(), dtype=float32)\n",
            "epoch: 4 step: 722 train batch loss: tf.Tensor(0.40561038, shape=(), dtype=float32)\n",
            "epoch: 4 step: 723 train batch loss: tf.Tensor(0.35494933, shape=(), dtype=float32)\n",
            "epoch: 4 step: 724 train batch loss: tf.Tensor(0.31555176, shape=(), dtype=float32)\n",
            "epoch: 4 step: 725 train batch loss: tf.Tensor(0.2976101, shape=(), dtype=float32)\n",
            "epoch: 4 step: 726 train batch loss: tf.Tensor(0.28624505, shape=(), dtype=float32)\n",
            "epoch: 4 step: 727 train batch loss: tf.Tensor(0.37202403, shape=(), dtype=float32)\n",
            "epoch: 4 step: 728 train batch loss: tf.Tensor(0.3796639, shape=(), dtype=float32)\n",
            "epoch: 4 step: 729 train batch loss: tf.Tensor(0.35296434, shape=(), dtype=float32)\n",
            "epoch: 4 step: 730 train batch loss: tf.Tensor(0.3789559, shape=(), dtype=float32)\n",
            "epoch: 4 step: 731 train batch loss: tf.Tensor(0.28991038, shape=(), dtype=float32)\n",
            "epoch: 4 step: 732 train batch loss: tf.Tensor(0.3632881, shape=(), dtype=float32)\n",
            "epoch: 4 step: 733 train batch loss: tf.Tensor(0.42818314, shape=(), dtype=float32)\n",
            "epoch: 4 step: 734 train batch loss: tf.Tensor(0.40359777, shape=(), dtype=float32)\n",
            "epoch: 4 step: 735 train batch loss: tf.Tensor(0.33860472, shape=(), dtype=float32)\n",
            "epoch: 4 step: 736 train batch loss: tf.Tensor(0.38570863, shape=(), dtype=float32)\n",
            "epoch: 4 step: 737 train batch loss: tf.Tensor(0.32020837, shape=(), dtype=float32)\n",
            "epoch: 4 step: 738 train batch loss: tf.Tensor(0.39744788, shape=(), dtype=float32)\n",
            "epoch: 4 step: 739 train batch loss: tf.Tensor(0.2858298, shape=(), dtype=float32)\n",
            "epoch: 4 step: 740 train batch loss: tf.Tensor(0.39000323, shape=(), dtype=float32)\n",
            "epoch: 4 step: 741 train batch loss: tf.Tensor(0.388352, shape=(), dtype=float32)\n",
            "epoch: 4 step: 742 train batch loss: tf.Tensor(0.39084798, shape=(), dtype=float32)\n",
            "epoch: 4 step: 743 train batch loss: tf.Tensor(0.32572585, shape=(), dtype=float32)\n",
            "epoch: 4 step: 744 train batch loss: tf.Tensor(0.3066805, shape=(), dtype=float32)\n",
            "epoch: 4 step: 745 train batch loss: tf.Tensor(0.28234094, shape=(), dtype=float32)\n",
            "epoch: 4 step: 746 train batch loss: tf.Tensor(0.42228207, shape=(), dtype=float32)\n",
            "epoch: 4 step: 747 train batch loss: tf.Tensor(0.3494399, shape=(), dtype=float32)\n",
            "epoch: 4 step: 748 train batch loss: tf.Tensor(0.34466073, shape=(), dtype=float32)\n",
            "epoch: 4 step: 749 train batch loss: tf.Tensor(0.36983684, shape=(), dtype=float32)\n",
            "epoch: 4 step: 750 train batch loss: tf.Tensor(0.25069848, shape=(), dtype=float32)\n",
            "epoch: 4 step: 751 train batch loss: tf.Tensor(0.3267154, shape=(), dtype=float32)\n",
            "epoch: 4 step: 752 train batch loss: tf.Tensor(0.39651486, shape=(), dtype=float32)\n",
            "epoch: 4 step: 753 train batch loss: tf.Tensor(0.3120784, shape=(), dtype=float32)\n",
            "epoch: 4 step: 754 train batch loss: tf.Tensor(0.36859295, shape=(), dtype=float32)\n",
            "epoch: 4 step: 755 train batch loss: tf.Tensor(0.31657153, shape=(), dtype=float32)\n",
            "epoch: 4 step: 756 train batch loss: tf.Tensor(0.26918632, shape=(), dtype=float32)\n",
            "epoch: 4 step: 757 train batch loss: tf.Tensor(0.3955047, shape=(), dtype=float32)\n",
            "epoch: 4 step: 758 train batch loss: tf.Tensor(0.38572174, shape=(), dtype=float32)\n",
            "epoch: 4 step: 759 train batch loss: tf.Tensor(0.37533647, shape=(), dtype=float32)\n",
            "epoch: 4 step: 760 train batch loss: tf.Tensor(0.31469086, shape=(), dtype=float32)\n",
            "epoch: 4 step: 761 train batch loss: tf.Tensor(0.32932553, shape=(), dtype=float32)\n",
            "epoch: 4 step: 762 train batch loss: tf.Tensor(0.35898054, shape=(), dtype=float32)\n",
            "epoch: 4 step: 763 train batch loss: tf.Tensor(0.35687256, shape=(), dtype=float32)\n",
            "epoch: 4 step: 764 train batch loss: tf.Tensor(0.4127883, shape=(), dtype=float32)\n",
            "epoch: 4 step: 765 train batch loss: tf.Tensor(0.3487801, shape=(), dtype=float32)\n",
            "epoch: 4 step: 766 train batch loss: tf.Tensor(0.34308082, shape=(), dtype=float32)\n",
            "epoch: 4 step: 767 train batch loss: tf.Tensor(0.3224614, shape=(), dtype=float32)\n",
            "epoch: 4 step: 768 train batch loss: tf.Tensor(0.327231, shape=(), dtype=float32)\n",
            "epoch: 4 step: 769 train batch loss: tf.Tensor(0.41764036, shape=(), dtype=float32)\n",
            "epoch: 4 step: 770 train batch loss: tf.Tensor(0.24028762, shape=(), dtype=float32)\n",
            "epoch: 4 step: 771 train batch loss: tf.Tensor(0.3477403, shape=(), dtype=float32)\n",
            "epoch: 4 step: 772 train batch loss: tf.Tensor(0.35912868, shape=(), dtype=float32)\n",
            "epoch: 4 step: 773 train batch loss: tf.Tensor(0.27302915, shape=(), dtype=float32)\n",
            "epoch: 4 step: 774 train batch loss: tf.Tensor(0.31791747, shape=(), dtype=float32)\n",
            "epoch: 4 step: 775 train batch loss: tf.Tensor(0.3073539, shape=(), dtype=float32)\n",
            "epoch: 4 step: 776 train batch loss: tf.Tensor(0.39251938, shape=(), dtype=float32)\n",
            "epoch: 4 step: 777 train batch loss: tf.Tensor(0.37337726, shape=(), dtype=float32)\n",
            "epoch: 4 step: 778 train batch loss: tf.Tensor(0.402344, shape=(), dtype=float32)\n",
            "epoch: 4 step: 779 train batch loss: tf.Tensor(0.28425062, shape=(), dtype=float32)\n",
            "epoch: 4 step: 780 train batch loss: tf.Tensor(0.463228, shape=(), dtype=float32)\n",
            "epoch: 4 step: 781 train batch loss: tf.Tensor(0.3273489, shape=(), dtype=float32)\n",
            "epoch: 4 step: 782 train batch loss: tf.Tensor(0.37745792, shape=(), dtype=float32)\n",
            "epoch: 4 step: 783 train batch loss: tf.Tensor(0.31793252, shape=(), dtype=float32)\n",
            "epoch: 4 step: 784 train batch loss: tf.Tensor(0.39025253, shape=(), dtype=float32)\n",
            "epoch: 4 step: 785 train batch loss: tf.Tensor(0.33953387, shape=(), dtype=float32)\n",
            "epoch: 4 step: 786 train batch loss: tf.Tensor(0.29757053, shape=(), dtype=float32)\n",
            "epoch: 4 step: 787 train batch loss: tf.Tensor(0.3615494, shape=(), dtype=float32)\n",
            "epoch: 4 step: 788 train batch loss: tf.Tensor(0.30010572, shape=(), dtype=float32)\n",
            "epoch: 4 step: 789 train batch loss: tf.Tensor(0.30990735, shape=(), dtype=float32)\n",
            "epoch: 4 step: 790 train batch loss: tf.Tensor(0.3454467, shape=(), dtype=float32)\n",
            "epoch: 4 step: 791 train batch loss: tf.Tensor(0.34200025, shape=(), dtype=float32)\n",
            "epoch: 4 step: 792 train batch loss: tf.Tensor(0.3486442, shape=(), dtype=float32)\n",
            "epoch: 4 step: 793 train batch loss: tf.Tensor(0.3611357, shape=(), dtype=float32)\n",
            "epoch: 4 step: 794 train batch loss: tf.Tensor(0.3140157, shape=(), dtype=float32)\n",
            "epoch: 4 step: 795 train batch loss: tf.Tensor(0.27129707, shape=(), dtype=float32)\n",
            "epoch: 4 step: 796 train batch loss: tf.Tensor(0.37137005, shape=(), dtype=float32)\n",
            "epoch: 4 step: 797 train batch loss: tf.Tensor(0.3615503, shape=(), dtype=float32)\n",
            "epoch: 4 step: 798 train batch loss: tf.Tensor(0.36135787, shape=(), dtype=float32)\n",
            "epoch: 4 step: 799 train batch loss: tf.Tensor(0.29566476, shape=(), dtype=float32)\n",
            "epoch: 4 step: 800 train batch loss: tf.Tensor(0.37790027, shape=(), dtype=float32)\n",
            "epoch: 4 step: 801 train batch loss: tf.Tensor(0.4009726, shape=(), dtype=float32)\n",
            "epoch: 4 step: 802 train batch loss: tf.Tensor(0.33182985, shape=(), dtype=float32)\n",
            "epoch: 4 step: 803 train batch loss: tf.Tensor(0.50256217, shape=(), dtype=float32)\n",
            "epoch: 4 step: 804 train batch loss: tf.Tensor(0.38488802, shape=(), dtype=float32)\n",
            "epoch: 4 step: 805 train batch loss: tf.Tensor(0.30875927, shape=(), dtype=float32)\n",
            "epoch: 4 step: 806 train batch loss: tf.Tensor(0.330026, shape=(), dtype=float32)\n",
            "epoch: 4 step: 807 train batch loss: tf.Tensor(0.30290076, shape=(), dtype=float32)\n",
            "epoch: 4 step: 808 train batch loss: tf.Tensor(0.3046171, shape=(), dtype=float32)\n",
            "epoch: 4 step: 809 train batch loss: tf.Tensor(0.33494192, shape=(), dtype=float32)\n",
            "epoch: 4 step: 810 train batch loss: tf.Tensor(0.35712206, shape=(), dtype=float32)\n",
            "epoch: 4 step: 811 train batch loss: tf.Tensor(0.31268418, shape=(), dtype=float32)\n",
            "epoch: 4 step: 812 train batch loss: tf.Tensor(0.33856508, shape=(), dtype=float32)\n",
            "epoch: 4 step: 813 train batch loss: tf.Tensor(0.3072582, shape=(), dtype=float32)\n",
            "epoch: 4 step: 814 train batch loss: tf.Tensor(0.31786424, shape=(), dtype=float32)\n",
            "epoch: 4 step: 815 train batch loss: tf.Tensor(0.3899833, shape=(), dtype=float32)\n",
            "epoch: 4 step: 816 train batch loss: tf.Tensor(0.33579162, shape=(), dtype=float32)\n",
            "epoch: 4 step: 817 train batch loss: tf.Tensor(0.3391921, shape=(), dtype=float32)\n",
            "epoch: 4 step: 818 train batch loss: tf.Tensor(0.3466984, shape=(), dtype=float32)\n",
            "epoch: 4 step: 819 train batch loss: tf.Tensor(0.311149, shape=(), dtype=float32)\n",
            "epoch: 4 step: 820 train batch loss: tf.Tensor(0.31379682, shape=(), dtype=float32)\n",
            "epoch: 4 step: 821 train batch loss: tf.Tensor(0.2940232, shape=(), dtype=float32)\n",
            "epoch: 4 step: 822 train batch loss: tf.Tensor(0.32585692, shape=(), dtype=float32)\n",
            "epoch: 4 step: 823 train batch loss: tf.Tensor(0.38319358, shape=(), dtype=float32)\n",
            "epoch: 4 step: 824 train batch loss: tf.Tensor(0.33452743, shape=(), dtype=float32)\n",
            "epoch: 4 step: 825 train batch loss: tf.Tensor(0.3787766, shape=(), dtype=float32)\n",
            "epoch: 4 step: 826 train batch loss: tf.Tensor(0.34943944, shape=(), dtype=float32)\n",
            "epoch: 4 step: 827 train batch loss: tf.Tensor(0.3518221, shape=(), dtype=float32)\n",
            "epoch: 4 step: 828 train batch loss: tf.Tensor(0.29996154, shape=(), dtype=float32)\n",
            "epoch: 4 step: 829 train batch loss: tf.Tensor(0.38640445, shape=(), dtype=float32)\n",
            "epoch: 4 step: 830 train batch loss: tf.Tensor(0.3254715, shape=(), dtype=float32)\n",
            "epoch: 4 step: 831 train batch loss: tf.Tensor(0.27728885, shape=(), dtype=float32)\n",
            "epoch: 4 step: 832 train batch loss: tf.Tensor(0.33621246, shape=(), dtype=float32)\n",
            "epoch: 4 step: 833 train batch loss: tf.Tensor(0.41642064, shape=(), dtype=float32)\n",
            "epoch: 4 step: 834 train batch loss: tf.Tensor(0.2740764, shape=(), dtype=float32)\n",
            "epoch: 4 step: 835 train batch loss: tf.Tensor(0.29124016, shape=(), dtype=float32)\n",
            "epoch: 4 step: 836 train batch loss: tf.Tensor(0.32395536, shape=(), dtype=float32)\n",
            "epoch: 4 step: 837 train batch loss: tf.Tensor(0.37463915, shape=(), dtype=float32)\n",
            "epoch: 4 step: 838 train batch loss: tf.Tensor(0.3277966, shape=(), dtype=float32)\n",
            "epoch: 4 step: 839 train batch loss: tf.Tensor(0.30441064, shape=(), dtype=float32)\n",
            "epoch: 4 step: 840 train batch loss: tf.Tensor(0.2860142, shape=(), dtype=float32)\n",
            "epoch: 4 step: 841 train batch loss: tf.Tensor(0.33043286, shape=(), dtype=float32)\n",
            "epoch: 4 step: 842 train batch loss: tf.Tensor(0.39943844, shape=(), dtype=float32)\n",
            "epoch: 4 step: 843 train batch loss: tf.Tensor(0.41420382, shape=(), dtype=float32)\n",
            "epoch: 4 step: 844 train batch loss: tf.Tensor(0.31156173, shape=(), dtype=float32)\n",
            "epoch: 4 step: 845 train batch loss: tf.Tensor(0.34423122, shape=(), dtype=float32)\n",
            "epoch: 4 step: 846 train batch loss: tf.Tensor(0.28503758, shape=(), dtype=float32)\n",
            "epoch: 4 step: 847 train batch loss: tf.Tensor(0.39879346, shape=(), dtype=float32)\n",
            "epoch: 4 step: 848 train batch loss: tf.Tensor(0.41594666, shape=(), dtype=float32)\n",
            "epoch: 4 step: 849 train batch loss: tf.Tensor(0.3198928, shape=(), dtype=float32)\n",
            "epoch: 4 step: 850 train batch loss: tf.Tensor(0.30874348, shape=(), dtype=float32)\n",
            "epoch: 4 step: 851 train batch loss: tf.Tensor(0.34215787, shape=(), dtype=float32)\n",
            "epoch: 4 step: 852 train batch loss: tf.Tensor(0.42045504, shape=(), dtype=float32)\n",
            "epoch: 4 step: 853 train batch loss: tf.Tensor(0.28269196, shape=(), dtype=float32)\n",
            "epoch: 4 step: 854 train batch loss: tf.Tensor(0.35273263, shape=(), dtype=float32)\n",
            "epoch: 4 step: 855 train batch loss: tf.Tensor(0.28725725, shape=(), dtype=float32)\n",
            "epoch: 4 step: 856 train batch loss: tf.Tensor(0.39534912, shape=(), dtype=float32)\n",
            "epoch: 4 step: 857 train batch loss: tf.Tensor(0.26980636, shape=(), dtype=float32)\n",
            "epoch: 4 step: 858 train batch loss: tf.Tensor(0.40935397, shape=(), dtype=float32)\n",
            "epoch: 4 step: 859 train batch loss: tf.Tensor(0.28547588, shape=(), dtype=float32)\n",
            "epoch: 4 step: 860 train batch loss: tf.Tensor(0.27818933, shape=(), dtype=float32)\n",
            "epoch: 4 step: 861 train batch loss: tf.Tensor(0.36482948, shape=(), dtype=float32)\n",
            "epoch: 4 step: 862 train batch loss: tf.Tensor(0.38870814, shape=(), dtype=float32)\n",
            "epoch: 4 step: 863 train batch loss: tf.Tensor(0.33988047, shape=(), dtype=float32)\n",
            "epoch: 4 step: 864 train batch loss: tf.Tensor(0.2776242, shape=(), dtype=float32)\n",
            "epoch: 4 step: 865 train batch loss: tf.Tensor(0.2976821, shape=(), dtype=float32)\n",
            "epoch: 4 step: 866 train batch loss: tf.Tensor(0.3243664, shape=(), dtype=float32)\n",
            "epoch: 4 step: 867 train batch loss: tf.Tensor(0.32118335, shape=(), dtype=float32)\n",
            "epoch: 4 step: 868 train batch loss: tf.Tensor(0.2836392, shape=(), dtype=float32)\n",
            "epoch: 4 step: 869 train batch loss: tf.Tensor(0.30985293, shape=(), dtype=float32)\n",
            "epoch: 4 step: 870 train batch loss: tf.Tensor(0.34134644, shape=(), dtype=float32)\n",
            "epoch: 4 step: 871 train batch loss: tf.Tensor(0.318878, shape=(), dtype=float32)\n",
            "epoch: 4 step: 872 train batch loss: tf.Tensor(0.25412273, shape=(), dtype=float32)\n",
            "epoch: 4 step: 873 train batch loss: tf.Tensor(0.36472192, shape=(), dtype=float32)\n",
            "epoch: 4 step: 874 train batch loss: tf.Tensor(0.37823954, shape=(), dtype=float32)\n",
            "epoch: 4 step: 875 train batch loss: tf.Tensor(0.30769548, shape=(), dtype=float32)\n",
            "epoch: 4 step: 876 train batch loss: tf.Tensor(0.41517115, shape=(), dtype=float32)\n",
            "epoch: 4 step: 877 train batch loss: tf.Tensor(0.30146492, shape=(), dtype=float32)\n",
            "epoch: 4 step: 878 train batch loss: tf.Tensor(0.4060215, shape=(), dtype=float32)\n",
            "epoch: 4 step: 879 train batch loss: tf.Tensor(0.34164992, shape=(), dtype=float32)\n",
            "epoch: 4 step: 880 train batch loss: tf.Tensor(0.36115727, shape=(), dtype=float32)\n",
            "epoch: 4 step: 881 train batch loss: tf.Tensor(0.30582958, shape=(), dtype=float32)\n",
            "epoch: 4 step: 882 train batch loss: tf.Tensor(0.2663458, shape=(), dtype=float32)\n",
            "epoch: 4 step: 883 train batch loss: tf.Tensor(0.3229715, shape=(), dtype=float32)\n",
            "epoch: 4 step: 884 train batch loss: tf.Tensor(0.32878876, shape=(), dtype=float32)\n",
            "epoch: 4 step: 885 train batch loss: tf.Tensor(0.38146204, shape=(), dtype=float32)\n",
            "epoch: 4 step: 886 train batch loss: tf.Tensor(0.358669, shape=(), dtype=float32)\n",
            "epoch: 4 step: 887 train batch loss: tf.Tensor(0.38193122, shape=(), dtype=float32)\n",
            "epoch: 4 step: 888 train batch loss: tf.Tensor(0.3392952, shape=(), dtype=float32)\n",
            "epoch: 4 step: 889 train batch loss: tf.Tensor(0.240949, shape=(), dtype=float32)\n",
            "epoch: 4 step: 890 train batch loss: tf.Tensor(0.44098765, shape=(), dtype=float32)\n",
            "epoch: 4 step: 891 train batch loss: tf.Tensor(0.28493258, shape=(), dtype=float32)\n",
            "epoch: 4 step: 892 train batch loss: tf.Tensor(0.30087632, shape=(), dtype=float32)\n",
            "epoch: 4 step: 893 train batch loss: tf.Tensor(0.36356476, shape=(), dtype=float32)\n",
            "epoch: 4 step: 894 train batch loss: tf.Tensor(0.3358101, shape=(), dtype=float32)\n",
            "epoch: 4 step: 895 train batch loss: tf.Tensor(0.37638405, shape=(), dtype=float32)\n",
            "epoch: 4 step: 896 train batch loss: tf.Tensor(0.28860068, shape=(), dtype=float32)\n",
            "epoch: 4 step: 897 train batch loss: tf.Tensor(0.36526757, shape=(), dtype=float32)\n",
            "epoch: 4 step: 898 train batch loss: tf.Tensor(0.34559596, shape=(), dtype=float32)\n",
            "epoch: 4 step: 899 train batch loss: tf.Tensor(0.3848153, shape=(), dtype=float32)\n",
            "epoch: 4 step: 900 train batch loss: tf.Tensor(0.33347228, shape=(), dtype=float32)\n",
            "epoch: 4 step: 901 train batch loss: tf.Tensor(0.40007314, shape=(), dtype=float32)\n",
            "epoch: 4 step: 902 train batch loss: tf.Tensor(0.40446836, shape=(), dtype=float32)\n",
            "epoch: 4 step: 903 train batch loss: tf.Tensor(0.26177743, shape=(), dtype=float32)\n",
            "epoch: 4 step: 904 train batch loss: tf.Tensor(0.31944403, shape=(), dtype=float32)\n",
            "epoch: 4 step: 905 train batch loss: tf.Tensor(0.33004266, shape=(), dtype=float32)\n",
            "epoch: 4 step: 906 train batch loss: tf.Tensor(0.33319232, shape=(), dtype=float32)\n",
            "epoch: 4 step: 907 train batch loss: tf.Tensor(0.3329728, shape=(), dtype=float32)\n",
            "epoch: 4 step: 908 train batch loss: tf.Tensor(0.38041064, shape=(), dtype=float32)\n",
            "epoch: 4 step: 909 train batch loss: tf.Tensor(0.360885, shape=(), dtype=float32)\n",
            "epoch: 4 step: 910 train batch loss: tf.Tensor(0.41398793, shape=(), dtype=float32)\n",
            "epoch: 4 step: 911 train batch loss: tf.Tensor(0.40092716, shape=(), dtype=float32)\n",
            "epoch: 4 step: 912 train batch loss: tf.Tensor(0.353294, shape=(), dtype=float32)\n",
            "epoch: 4 step: 913 train batch loss: tf.Tensor(0.3431086, shape=(), dtype=float32)\n",
            "epoch: 4 step: 914 train batch loss: tf.Tensor(0.37580383, shape=(), dtype=float32)\n",
            "epoch: 4 step: 915 train batch loss: tf.Tensor(0.3262343, shape=(), dtype=float32)\n",
            "epoch: 4 step: 916 train batch loss: tf.Tensor(0.30780753, shape=(), dtype=float32)\n",
            "epoch: 4 step: 917 train batch loss: tf.Tensor(0.39673418, shape=(), dtype=float32)\n",
            "epoch: 4 step: 918 train batch loss: tf.Tensor(0.29270333, shape=(), dtype=float32)\n",
            "epoch: 4 step: 919 train batch loss: tf.Tensor(0.25543082, shape=(), dtype=float32)\n",
            "epoch: 4 step: 920 train batch loss: tf.Tensor(0.32077464, shape=(), dtype=float32)\n",
            "epoch: 4 step: 921 train batch loss: tf.Tensor(0.32692242, shape=(), dtype=float32)\n",
            "epoch: 4 step: 922 train batch loss: tf.Tensor(0.38275844, shape=(), dtype=float32)\n",
            "epoch: 4 step: 923 train batch loss: tf.Tensor(0.33047092, shape=(), dtype=float32)\n",
            "epoch: 4 step: 924 train batch loss: tf.Tensor(0.38866097, shape=(), dtype=float32)\n",
            "epoch: 4 step: 925 train batch loss: tf.Tensor(0.32596838, shape=(), dtype=float32)\n",
            "epoch: 4 step: 926 train batch loss: tf.Tensor(0.35726008, shape=(), dtype=float32)\n",
            "epoch: 4 step: 927 train batch loss: tf.Tensor(0.31058615, shape=(), dtype=float32)\n",
            "epoch: 4 step: 928 train batch loss: tf.Tensor(0.32393262, shape=(), dtype=float32)\n",
            "epoch: 4 step: 929 train batch loss: tf.Tensor(0.3355118, shape=(), dtype=float32)\n",
            "epoch: 4 step: 930 train batch loss: tf.Tensor(0.38721275, shape=(), dtype=float32)\n",
            "epoch: 4 step: 931 train batch loss: tf.Tensor(0.27402216, shape=(), dtype=float32)\n",
            "epoch: 4 step: 932 train batch loss: tf.Tensor(0.34746224, shape=(), dtype=float32)\n",
            "epoch: 4 step: 933 train batch loss: tf.Tensor(0.29739967, shape=(), dtype=float32)\n",
            "epoch: 4 step: 934 train batch loss: tf.Tensor(0.38381526, shape=(), dtype=float32)\n",
            "epoch: 4 step: 935 train batch loss: tf.Tensor(0.3154177, shape=(), dtype=float32)\n",
            "epoch: 4 step: 936 train batch loss: tf.Tensor(0.30312628, shape=(), dtype=float32)\n",
            "epoch: 4 step: 937 train batch loss: tf.Tensor(0.28529578, shape=(), dtype=float32)\n",
            "epoch: 4 step: 938 train batch loss: tf.Tensor(0.3213736, shape=(), dtype=float32)\n",
            "epoch: 4 step: 939 train batch loss: tf.Tensor(0.28907, shape=(), dtype=float32)\n",
            "epoch: 4 step: 940 train batch loss: tf.Tensor(0.32177368, shape=(), dtype=float32)\n",
            "epoch: 4 step: 941 train batch loss: tf.Tensor(0.35854015, shape=(), dtype=float32)\n",
            "epoch: 4 step: 942 train batch loss: tf.Tensor(0.39621526, shape=(), dtype=float32)\n",
            "epoch: 4 step: 943 train batch loss: tf.Tensor(0.2989202, shape=(), dtype=float32)\n",
            "epoch: 4 step: 944 train batch loss: tf.Tensor(0.31484094, shape=(), dtype=float32)\n",
            "epoch: 4 step: 945 train batch loss: tf.Tensor(0.39628455, shape=(), dtype=float32)\n",
            "epoch: 4 step: 946 train batch loss: tf.Tensor(0.31506017, shape=(), dtype=float32)\n",
            "epoch: 4 step: 947 train batch loss: tf.Tensor(0.4001697, shape=(), dtype=float32)\n",
            "epoch: 4 step: 948 train batch loss: tf.Tensor(0.30015558, shape=(), dtype=float32)\n",
            "epoch: 4 step: 949 train batch loss: tf.Tensor(0.31607297, shape=(), dtype=float32)\n",
            "epoch: 4 step: 950 train batch loss: tf.Tensor(0.41031665, shape=(), dtype=float32)\n",
            "epoch: 4 step: 951 train batch loss: tf.Tensor(0.3377693, shape=(), dtype=float32)\n",
            "epoch: 4 step: 952 train batch loss: tf.Tensor(0.31150484, shape=(), dtype=float32)\n",
            "epoch: 4 step: 953 train batch loss: tf.Tensor(0.34137282, shape=(), dtype=float32)\n",
            "epoch: 4 step: 954 train batch loss: tf.Tensor(0.25254682, shape=(), dtype=float32)\n",
            "epoch: 4 step: 955 train batch loss: tf.Tensor(0.40038604, shape=(), dtype=float32)\n",
            "epoch: 4 step: 956 train batch loss: tf.Tensor(0.32209897, shape=(), dtype=float32)\n",
            "epoch: 4 step: 957 train batch loss: tf.Tensor(0.31428927, shape=(), dtype=float32)\n",
            "epoch: 4 step: 958 train batch loss: tf.Tensor(0.34718934, shape=(), dtype=float32)\n",
            "epoch: 4 step: 959 train batch loss: tf.Tensor(0.29964375, shape=(), dtype=float32)\n",
            "epoch: 4 step: 960 train batch loss: tf.Tensor(0.37643212, shape=(), dtype=float32)\n",
            "epoch: 4 step: 961 train batch loss: tf.Tensor(0.2925579, shape=(), dtype=float32)\n",
            "epoch: 4 step: 962 train batch loss: tf.Tensor(0.36540097, shape=(), dtype=float32)\n",
            "epoch: 4 step: 963 train batch loss: tf.Tensor(0.34605733, shape=(), dtype=float32)\n",
            "epoch: 4 step: 964 train batch loss: tf.Tensor(0.31202823, shape=(), dtype=float32)\n",
            "epoch: 4 step: 965 train batch loss: tf.Tensor(0.37946153, shape=(), dtype=float32)\n",
            "epoch: 4 step: 966 train batch loss: tf.Tensor(0.36316684, shape=(), dtype=float32)\n",
            "epoch: 4 step: 967 train batch loss: tf.Tensor(0.347686, shape=(), dtype=float32)\n",
            "epoch: 4 step: 968 train batch loss: tf.Tensor(0.33004275, shape=(), dtype=float32)\n",
            "epoch: 4 step: 969 train batch loss: tf.Tensor(0.27968866, shape=(), dtype=float32)\n",
            "epoch: 4 step: 970 train batch loss: tf.Tensor(0.2928891, shape=(), dtype=float32)\n",
            "epoch: 4 step: 971 train batch loss: tf.Tensor(0.42218938, shape=(), dtype=float32)\n",
            "epoch: 4 step: 972 train batch loss: tf.Tensor(0.25147575, shape=(), dtype=float32)\n",
            "epoch: 4 step: 973 train batch loss: tf.Tensor(0.2903567, shape=(), dtype=float32)\n",
            "epoch: 4 step: 974 train batch loss: tf.Tensor(0.33556113, shape=(), dtype=float32)\n",
            "epoch: 4 step: 975 train batch loss: tf.Tensor(0.34998015, shape=(), dtype=float32)\n",
            "epoch: 4 step: 976 train batch loss: tf.Tensor(0.31425396, shape=(), dtype=float32)\n",
            "epoch: 4 step: 977 train batch loss: tf.Tensor(0.47559863, shape=(), dtype=float32)\n",
            "epoch: 4 step: 978 train batch loss: tf.Tensor(0.37576008, shape=(), dtype=float32)\n",
            "epoch: 4 step: 979 train batch loss: tf.Tensor(0.32335025, shape=(), dtype=float32)\n",
            "epoch: 4 step: 980 train batch loss: tf.Tensor(0.3350251, shape=(), dtype=float32)\n",
            "epoch: 4 step: 981 train batch loss: tf.Tensor(0.21947575, shape=(), dtype=float32)\n",
            "epoch: 4 step: 982 train batch loss: tf.Tensor(0.36371547, shape=(), dtype=float32)\n",
            "epoch: 4 step: 983 train batch loss: tf.Tensor(0.31923273, shape=(), dtype=float32)\n",
            "epoch: 4 step: 984 train batch loss: tf.Tensor(0.3522704, shape=(), dtype=float32)\n",
            "epoch: 4 step: 985 train batch loss: tf.Tensor(0.29795086, shape=(), dtype=float32)\n",
            "epoch: 4 step: 986 train batch loss: tf.Tensor(0.38404566, shape=(), dtype=float32)\n",
            "epoch: 4 step: 987 train batch loss: tf.Tensor(0.3081407, shape=(), dtype=float32)\n",
            "epoch: 4 step: 988 train batch loss: tf.Tensor(0.3323338, shape=(), dtype=float32)\n",
            "epoch: 4 step: 989 train batch loss: tf.Tensor(0.33179393, shape=(), dtype=float32)\n",
            "epoch: 4 step: 990 train batch loss: tf.Tensor(0.29476398, shape=(), dtype=float32)\n",
            "epoch: 4 step: 991 train batch loss: tf.Tensor(0.3595118, shape=(), dtype=float32)\n",
            "epoch: 4 step: 992 train batch loss: tf.Tensor(0.38378316, shape=(), dtype=float32)\n",
            "epoch: 4 step: 993 train batch loss: tf.Tensor(0.30676278, shape=(), dtype=float32)\n",
            "epoch: 4 step: 994 train batch loss: tf.Tensor(0.33800927, shape=(), dtype=float32)\n",
            "epoch: 4 step: 995 train batch loss: tf.Tensor(0.33249092, shape=(), dtype=float32)\n",
            "epoch: 4 step: 996 train batch loss: tf.Tensor(0.39132974, shape=(), dtype=float32)\n",
            "epoch: 4 step: 997 train batch loss: tf.Tensor(0.27841017, shape=(), dtype=float32)\n",
            "epoch: 4 step: 998 train batch loss: tf.Tensor(0.34052798, shape=(), dtype=float32)\n",
            "epoch: 4 step: 999 train batch loss: tf.Tensor(0.34376848, shape=(), dtype=float32)\n",
            "epoch: 4 step: 0 val batch loss: tf.Tensor(0.6695134, shape=(), dtype=float32)\n",
            "epoch: 4 step: 1 val batch loss: tf.Tensor(0.77244014, shape=(), dtype=float32)\n",
            "epoch: 4 step: 2 val batch loss: tf.Tensor(0.6684332, shape=(), dtype=float32)\n",
            "epoch: 4 step: 3 val batch loss: tf.Tensor(1.0979434, shape=(), dtype=float32)\n",
            "epoch: 4 step: 4 val batch loss: tf.Tensor(0.6130758, shape=(), dtype=float32)\n",
            "epoch: 4 step: 5 val batch loss: tf.Tensor(0.6122422, shape=(), dtype=float32)\n",
            "epoch: 4 step: 6 val batch loss: tf.Tensor(0.47191507, shape=(), dtype=float32)\n",
            "epoch: 4 step: 7 val batch loss: tf.Tensor(0.5343791, shape=(), dtype=float32)\n",
            "epoch: 4 step: 8 val batch loss: tf.Tensor(0.46391553, shape=(), dtype=float32)\n",
            "epoch: 4 step: 9 val batch loss: tf.Tensor(0.44896108, shape=(), dtype=float32)\n",
            "epoch: 4 step: 10 val batch loss: tf.Tensor(0.631467, shape=(), dtype=float32)\n",
            "epoch: 4 step: 11 val batch loss: tf.Tensor(0.7698024, shape=(), dtype=float32)\n",
            "epoch: 4 step: 12 val batch loss: tf.Tensor(0.5464901, shape=(), dtype=float32)\n",
            "epoch: 4 step: 13 val batch loss: tf.Tensor(0.44996324, shape=(), dtype=float32)\n",
            "epoch: 4 step: 14 val batch loss: tf.Tensor(0.8287605, shape=(), dtype=float32)\n",
            "epoch: 4 step: 15 val batch loss: tf.Tensor(0.4844622, shape=(), dtype=float32)\n",
            "epoch: 4 step: 16 val batch loss: tf.Tensor(0.7311336, shape=(), dtype=float32)\n",
            "epoch: 4 step: 17 val batch loss: tf.Tensor(0.9490363, shape=(), dtype=float32)\n",
            "epoch: 4 step: 18 val batch loss: tf.Tensor(0.8300355, shape=(), dtype=float32)\n",
            "epoch: 4 step: 19 val batch loss: tf.Tensor(0.4191373, shape=(), dtype=float32)\n",
            "epoch: 4 step: 20 val batch loss: tf.Tensor(0.52212757, shape=(), dtype=float32)\n",
            "epoch: 4 step: 21 val batch loss: tf.Tensor(0.6678409, shape=(), dtype=float32)\n",
            "epoch: 4 step: 22 val batch loss: tf.Tensor(0.6872649, shape=(), dtype=float32)\n",
            "epoch: 4 step: 23 val batch loss: tf.Tensor(0.68236375, shape=(), dtype=float32)\n",
            "epoch: 4 step: 24 val batch loss: tf.Tensor(0.7469613, shape=(), dtype=float32)\n",
            "epoch: 4 step: 25 val batch loss: tf.Tensor(0.7810437, shape=(), dtype=float32)\n",
            "epoch: 4 step: 26 val batch loss: tf.Tensor(0.4911027, shape=(), dtype=float32)\n",
            "epoch: 4 step: 27 val batch loss: tf.Tensor(0.85487664, shape=(), dtype=float32)\n",
            "epoch: 4 step: 28 val batch loss: tf.Tensor(0.79197776, shape=(), dtype=float32)\n",
            "epoch: 4 step: 29 val batch loss: tf.Tensor(0.51426965, shape=(), dtype=float32)\n",
            "epoch: 4 step: 30 val batch loss: tf.Tensor(0.70385563, shape=(), dtype=float32)\n",
            "epoch: 4 step: 31 val batch loss: tf.Tensor(0.61677575, shape=(), dtype=float32)\n",
            "epoch: 4 step: 32 val batch loss: tf.Tensor(0.7102274, shape=(), dtype=float32)\n",
            "epoch: 4 step: 33 val batch loss: tf.Tensor(0.67471457, shape=(), dtype=float32)\n",
            "epoch: 4 step: 34 val batch loss: tf.Tensor(0.7549597, shape=(), dtype=float32)\n",
            "epoch: 4 step: 35 val batch loss: tf.Tensor(0.4914754, shape=(), dtype=float32)\n",
            "epoch: 4 step: 36 val batch loss: tf.Tensor(0.561166, shape=(), dtype=float32)\n",
            "epoch: 4 step: 37 val batch loss: tf.Tensor(0.53438216, shape=(), dtype=float32)\n",
            "epoch: 4 step: 38 val batch loss: tf.Tensor(0.6700882, shape=(), dtype=float32)\n",
            "epoch: 4 step: 39 val batch loss: tf.Tensor(0.49449563, shape=(), dtype=float32)\n",
            "epoch: 4 step: 40 val batch loss: tf.Tensor(0.69855356, shape=(), dtype=float32)\n",
            "epoch: 4 step: 41 val batch loss: tf.Tensor(0.5823999, shape=(), dtype=float32)\n",
            "epoch: 4 step: 42 val batch loss: tf.Tensor(0.78141654, shape=(), dtype=float32)\n",
            "epoch: 4 step: 43 val batch loss: tf.Tensor(0.5969086, shape=(), dtype=float32)\n",
            "epoch: 4 step: 44 val batch loss: tf.Tensor(0.6160432, shape=(), dtype=float32)\n",
            "epoch: 4 step: 45 val batch loss: tf.Tensor(0.7305132, shape=(), dtype=float32)\n",
            "epoch: 4 step: 46 val batch loss: tf.Tensor(0.8421205, shape=(), dtype=float32)\n",
            "epoch: 4 step: 47 val batch loss: tf.Tensor(0.59840095, shape=(), dtype=float32)\n",
            "epoch: 4 step: 48 val batch loss: tf.Tensor(0.5660792, shape=(), dtype=float32)\n",
            "epoch: 4 step: 49 val batch loss: tf.Tensor(0.6012902, shape=(), dtype=float32)\n",
            "epoch: 4 step: 50 val batch loss: tf.Tensor(0.63253635, shape=(), dtype=float32)\n",
            "epoch: 4 step: 51 val batch loss: tf.Tensor(0.5962988, shape=(), dtype=float32)\n",
            "epoch: 4 step: 52 val batch loss: tf.Tensor(0.7080987, shape=(), dtype=float32)\n",
            "epoch: 4 step: 53 val batch loss: tf.Tensor(0.6943467, shape=(), dtype=float32)\n",
            "epoch: 4 step: 54 val batch loss: tf.Tensor(0.43751273, shape=(), dtype=float32)\n",
            "epoch: 4 step: 55 val batch loss: tf.Tensor(0.77625495, shape=(), dtype=float32)\n",
            "epoch: 4 step: 56 val batch loss: tf.Tensor(0.5446657, shape=(), dtype=float32)\n",
            "epoch: 4 step: 57 val batch loss: tf.Tensor(0.62688017, shape=(), dtype=float32)\n",
            "epoch: 4 step: 58 val batch loss: tf.Tensor(0.70937514, shape=(), dtype=float32)\n",
            "epoch: 4 step: 59 val batch loss: tf.Tensor(0.6712681, shape=(), dtype=float32)\n",
            "epoch: 4 step: 60 val batch loss: tf.Tensor(0.5691988, shape=(), dtype=float32)\n",
            "epoch: 4 step: 61 val batch loss: tf.Tensor(0.5849452, shape=(), dtype=float32)\n",
            "epoch: 4 step: 62 val batch loss: tf.Tensor(0.74789476, shape=(), dtype=float32)\n",
            "epoch: 4 step: 63 val batch loss: tf.Tensor(0.55866873, shape=(), dtype=float32)\n",
            "epoch: 4 step: 64 val batch loss: tf.Tensor(0.5952156, shape=(), dtype=float32)\n",
            "epoch: 4 step: 65 val batch loss: tf.Tensor(0.55286014, shape=(), dtype=float32)\n",
            "epoch: 4 step: 66 val batch loss: tf.Tensor(0.6656245, shape=(), dtype=float32)\n",
            "epoch: 4 step: 67 val batch loss: tf.Tensor(0.79134834, shape=(), dtype=float32)\n",
            "epoch: 4 step: 68 val batch loss: tf.Tensor(0.6684977, shape=(), dtype=float32)\n",
            "epoch: 4 step: 69 val batch loss: tf.Tensor(0.5191096, shape=(), dtype=float32)\n",
            "epoch: 4 step: 70 val batch loss: tf.Tensor(0.80735093, shape=(), dtype=float32)\n",
            "epoch: 4 step: 71 val batch loss: tf.Tensor(0.664039, shape=(), dtype=float32)\n",
            "epoch: 4 step: 72 val batch loss: tf.Tensor(0.6717113, shape=(), dtype=float32)\n",
            "epoch: 4 step: 73 val batch loss: tf.Tensor(0.7437614, shape=(), dtype=float32)\n",
            "epoch: 4 step: 74 val batch loss: tf.Tensor(0.6939585, shape=(), dtype=float32)\n",
            "epoch: 4 step: 75 val batch loss: tf.Tensor(0.6028182, shape=(), dtype=float32)\n",
            "epoch: 4 step: 76 val batch loss: tf.Tensor(0.7299709, shape=(), dtype=float32)\n",
            "epoch: 4 step: 77 val batch loss: tf.Tensor(0.69064385, shape=(), dtype=float32)\n",
            "epoch: 4 step: 78 val batch loss: tf.Tensor(0.57757074, shape=(), dtype=float32)\n",
            "epoch: 4 step: 79 val batch loss: tf.Tensor(0.69389546, shape=(), dtype=float32)\n",
            "epoch: 4 step: 80 val batch loss: tf.Tensor(0.51014537, shape=(), dtype=float32)\n",
            "epoch: 4 step: 81 val batch loss: tf.Tensor(0.66464055, shape=(), dtype=float32)\n",
            "epoch: 4 step: 82 val batch loss: tf.Tensor(0.6487791, shape=(), dtype=float32)\n",
            "epoch: 4 step: 83 val batch loss: tf.Tensor(0.6024326, shape=(), dtype=float32)\n",
            "epoch: 4 step: 84 val batch loss: tf.Tensor(0.518552, shape=(), dtype=float32)\n",
            "epoch: 4 step: 85 val batch loss: tf.Tensor(0.61494726, shape=(), dtype=float32)\n",
            "epoch: 4 step: 86 val batch loss: tf.Tensor(0.6767896, shape=(), dtype=float32)\n",
            "epoch: 4 step: 87 val batch loss: tf.Tensor(0.6448177, shape=(), dtype=float32)\n",
            "epoch: 4 step: 88 val batch loss: tf.Tensor(0.6396516, shape=(), dtype=float32)\n",
            "epoch: 4 step: 89 val batch loss: tf.Tensor(0.51109356, shape=(), dtype=float32)\n",
            "epoch: 4 step: 90 val batch loss: tf.Tensor(0.8890206, shape=(), dtype=float32)\n",
            "epoch: 4 step: 91 val batch loss: tf.Tensor(0.70742005, shape=(), dtype=float32)\n",
            "epoch: 4 step: 92 val batch loss: tf.Tensor(0.6554083, shape=(), dtype=float32)\n",
            "epoch: 4 step: 93 val batch loss: tf.Tensor(0.61284006, shape=(), dtype=float32)\n",
            "epoch: 4 step: 94 val batch loss: tf.Tensor(0.461574, shape=(), dtype=float32)\n",
            "epoch: 4 step: 95 val batch loss: tf.Tensor(0.52883834, shape=(), dtype=float32)\n",
            "epoch: 4 step: 96 val batch loss: tf.Tensor(0.4571031, shape=(), dtype=float32)\n",
            "epoch: 4 step: 97 val batch loss: tf.Tensor(0.79204166, shape=(), dtype=float32)\n",
            "epoch: 4 step: 98 val batch loss: tf.Tensor(0.75611943, shape=(), dtype=float32)\n",
            "epoch: 4 step: 99 val batch loss: tf.Tensor(0.6673008, shape=(), dtype=float32)\n",
            "epoch: 4 step: 100 val batch loss: tf.Tensor(0.8943828, shape=(), dtype=float32)\n",
            "epoch: 4 step: 101 val batch loss: tf.Tensor(0.6635523, shape=(), dtype=float32)\n",
            "epoch: 4 step: 102 val batch loss: tf.Tensor(0.7400811, shape=(), dtype=float32)\n",
            "epoch: 4 step: 103 val batch loss: tf.Tensor(0.60986596, shape=(), dtype=float32)\n",
            "epoch: 4 step: 104 val batch loss: tf.Tensor(0.6742801, shape=(), dtype=float32)\n",
            "epoch: 4 step: 105 val batch loss: tf.Tensor(0.4640239, shape=(), dtype=float32)\n",
            "epoch: 4 step: 106 val batch loss: tf.Tensor(0.6688485, shape=(), dtype=float32)\n",
            "epoch: 4 step: 107 val batch loss: tf.Tensor(0.5752317, shape=(), dtype=float32)\n",
            "epoch: 4 step: 108 val batch loss: tf.Tensor(0.63622385, shape=(), dtype=float32)\n",
            "epoch: 4 step: 109 val batch loss: tf.Tensor(0.6545747, shape=(), dtype=float32)\n",
            "epoch: 4 step: 110 val batch loss: tf.Tensor(0.53115356, shape=(), dtype=float32)\n",
            "epoch: 4 step: 111 val batch loss: tf.Tensor(0.590616, shape=(), dtype=float32)\n",
            "epoch: 4 step: 112 val batch loss: tf.Tensor(0.6819728, shape=(), dtype=float32)\n",
            "epoch: 4 step: 113 val batch loss: tf.Tensor(0.64416856, shape=(), dtype=float32)\n",
            "epoch: 4 step: 114 val batch loss: tf.Tensor(0.62824786, shape=(), dtype=float32)\n",
            "epoch: 4 step: 115 val batch loss: tf.Tensor(0.5653132, shape=(), dtype=float32)\n",
            "epoch: 4 step: 116 val batch loss: tf.Tensor(0.6871883, shape=(), dtype=float32)\n",
            "epoch: 4 step: 117 val batch loss: tf.Tensor(0.51053065, shape=(), dtype=float32)\n",
            "epoch: 4 step: 118 val batch loss: tf.Tensor(0.580835, shape=(), dtype=float32)\n",
            "epoch: 4 step: 119 val batch loss: tf.Tensor(0.68952954, shape=(), dtype=float32)\n",
            "epoch: 4 step: 120 val batch loss: tf.Tensor(0.5456845, shape=(), dtype=float32)\n",
            "epoch: 4 step: 121 val batch loss: tf.Tensor(0.9019049, shape=(), dtype=float32)\n",
            "epoch: 4 step: 122 val batch loss: tf.Tensor(0.619298, shape=(), dtype=float32)\n",
            "epoch: 4 step: 123 val batch loss: tf.Tensor(0.64128315, shape=(), dtype=float32)\n",
            "epoch: 4 step: 124 val batch loss: tf.Tensor(0.6923589, shape=(), dtype=float32)\n",
            "epoch: 4 step: 125 val batch loss: tf.Tensor(0.6286904, shape=(), dtype=float32)\n",
            "epoch: 4 step: 126 val batch loss: tf.Tensor(0.7535396, shape=(), dtype=float32)\n",
            "epoch: 4 step: 127 val batch loss: tf.Tensor(0.7113733, shape=(), dtype=float32)\n",
            "epoch: 4 step: 128 val batch loss: tf.Tensor(0.6468357, shape=(), dtype=float32)\n",
            "epoch: 4 step: 129 val batch loss: tf.Tensor(0.63234967, shape=(), dtype=float32)\n",
            "epoch: 4 step: 130 val batch loss: tf.Tensor(0.5681965, shape=(), dtype=float32)\n",
            "epoch: 4 step: 131 val batch loss: tf.Tensor(0.67181444, shape=(), dtype=float32)\n",
            "epoch: 4 step: 132 val batch loss: tf.Tensor(0.53268206, shape=(), dtype=float32)\n",
            "epoch: 4 step: 133 val batch loss: tf.Tensor(0.55220795, shape=(), dtype=float32)\n",
            "epoch: 4 step: 134 val batch loss: tf.Tensor(0.68025374, shape=(), dtype=float32)\n",
            "epoch: 4 step: 135 val batch loss: tf.Tensor(0.5190387, shape=(), dtype=float32)\n",
            "epoch: 4 step: 136 val batch loss: tf.Tensor(0.56314147, shape=(), dtype=float32)\n",
            "epoch: 4 step: 137 val batch loss: tf.Tensor(0.5801958, shape=(), dtype=float32)\n",
            "epoch: 4 step: 138 val batch loss: tf.Tensor(0.5321175, shape=(), dtype=float32)\n",
            "epoch: 4 step: 139 val batch loss: tf.Tensor(0.69036025, shape=(), dtype=float32)\n",
            "epoch: 4 step: 140 val batch loss: tf.Tensor(0.54208213, shape=(), dtype=float32)\n",
            "epoch: 4 step: 141 val batch loss: tf.Tensor(0.5515348, shape=(), dtype=float32)\n",
            "epoch: 4 step: 142 val batch loss: tf.Tensor(0.7465219, shape=(), dtype=float32)\n",
            "epoch: 4 step: 143 val batch loss: tf.Tensor(0.66863126, shape=(), dtype=float32)\n",
            "epoch: 4 step: 144 val batch loss: tf.Tensor(0.69587815, shape=(), dtype=float32)\n",
            "epoch: 4 step: 145 val batch loss: tf.Tensor(0.6638889, shape=(), dtype=float32)\n",
            "epoch: 4 step: 146 val batch loss: tf.Tensor(0.5975899, shape=(), dtype=float32)\n",
            "epoch: 4 step: 147 val batch loss: tf.Tensor(0.69779235, shape=(), dtype=float32)\n",
            "epoch: 4 step: 148 val batch loss: tf.Tensor(0.60337716, shape=(), dtype=float32)\n",
            "epoch: 4 step: 149 val batch loss: tf.Tensor(0.6880678, shape=(), dtype=float32)\n",
            "epoch: 4 step: 150 val batch loss: tf.Tensor(0.5884438, shape=(), dtype=float32)\n",
            "epoch: 4 step: 151 val batch loss: tf.Tensor(0.7419865, shape=(), dtype=float32)\n",
            "epoch: 4 step: 152 val batch loss: tf.Tensor(0.61049974, shape=(), dtype=float32)\n",
            "epoch: 4 step: 153 val batch loss: tf.Tensor(0.5118584, shape=(), dtype=float32)\n",
            "epoch: 4 step: 154 val batch loss: tf.Tensor(0.5745082, shape=(), dtype=float32)\n",
            "epoch: 4 step: 155 val batch loss: tf.Tensor(0.7103345, shape=(), dtype=float32)\n",
            "epoch: 4 step: 156 val batch loss: tf.Tensor(0.54416764, shape=(), dtype=float32)\n",
            "epoch: 4 step: 157 val batch loss: tf.Tensor(0.5637809, shape=(), dtype=float32)\n",
            "epoch: 4 step: 158 val batch loss: tf.Tensor(0.5605582, shape=(), dtype=float32)\n",
            "epoch: 4 step: 159 val batch loss: tf.Tensor(0.4986443, shape=(), dtype=float32)\n",
            "epoch: 4 step: 160 val batch loss: tf.Tensor(0.6417108, shape=(), dtype=float32)\n",
            "epoch: 4 step: 161 val batch loss: tf.Tensor(0.46770257, shape=(), dtype=float32)\n",
            "epoch: 4 step: 162 val batch loss: tf.Tensor(0.7334572, shape=(), dtype=float32)\n",
            "epoch: 4 step: 163 val batch loss: tf.Tensor(0.6611318, shape=(), dtype=float32)\n",
            "epoch: 4 step: 164 val batch loss: tf.Tensor(0.8236054, shape=(), dtype=float32)\n",
            "epoch: 4 step: 165 val batch loss: tf.Tensor(0.71445775, shape=(), dtype=float32)\n",
            "epoch: 4 step: 166 val batch loss: tf.Tensor(0.6156488, shape=(), dtype=float32)\n",
            "epoch: 4 step: 167 val batch loss: tf.Tensor(0.57994705, shape=(), dtype=float32)\n",
            "epoch: 4 step: 168 val batch loss: tf.Tensor(0.53242016, shape=(), dtype=float32)\n",
            "epoch: 4 step: 169 val batch loss: tf.Tensor(0.47319287, shape=(), dtype=float32)\n",
            "epoch: 4 step: 170 val batch loss: tf.Tensor(0.6252726, shape=(), dtype=float32)\n",
            "epoch: 4 step: 171 val batch loss: tf.Tensor(0.6378294, shape=(), dtype=float32)\n",
            "epoch: 4 step: 172 val batch loss: tf.Tensor(0.4961029, shape=(), dtype=float32)\n",
            "epoch: 4 step: 173 val batch loss: tf.Tensor(0.59412026, shape=(), dtype=float32)\n",
            "epoch: 4 step: 174 val batch loss: tf.Tensor(0.7606029, shape=(), dtype=float32)\n",
            "epoch: 4 step: 175 val batch loss: tf.Tensor(0.7388139, shape=(), dtype=float32)\n",
            "epoch: 4 step: 176 val batch loss: tf.Tensor(0.53826517, shape=(), dtype=float32)\n",
            "epoch: 4 step: 177 val batch loss: tf.Tensor(0.58079183, shape=(), dtype=float32)\n",
            "epoch: 4 step: 178 val batch loss: tf.Tensor(0.46211568, shape=(), dtype=float32)\n",
            "epoch: 4 step: 179 val batch loss: tf.Tensor(0.6072155, shape=(), dtype=float32)\n",
            "epoch: 4 step: 180 val batch loss: tf.Tensor(0.70923555, shape=(), dtype=float32)\n",
            "epoch: 4 step: 181 val batch loss: tf.Tensor(0.50582826, shape=(), dtype=float32)\n",
            "epoch: 4 step: 182 val batch loss: tf.Tensor(0.6330398, shape=(), dtype=float32)\n",
            "epoch: 4 step: 183 val batch loss: tf.Tensor(0.57235444, shape=(), dtype=float32)\n",
            "epoch: 4 step: 184 val batch loss: tf.Tensor(0.5259856, shape=(), dtype=float32)\n",
            "epoch: 4 step: 185 val batch loss: tf.Tensor(0.6717816, shape=(), dtype=float32)\n",
            "epoch: 4 step: 186 val batch loss: tf.Tensor(0.763215, shape=(), dtype=float32)\n",
            "epoch: 4 step: 187 val batch loss: tf.Tensor(0.71730316, shape=(), dtype=float32)\n",
            "epoch: 4 step: 188 val batch loss: tf.Tensor(0.56583536, shape=(), dtype=float32)\n",
            "epoch: 4 step: 189 val batch loss: tf.Tensor(0.5180448, shape=(), dtype=float32)\n",
            "epoch: 4 step: 190 val batch loss: tf.Tensor(0.5959811, shape=(), dtype=float32)\n",
            "epoch: 4 step: 191 val batch loss: tf.Tensor(0.6375251, shape=(), dtype=float32)\n",
            "epoch: 4 step: 192 val batch loss: tf.Tensor(0.6730287, shape=(), dtype=float32)\n",
            "epoch: 4 step: 193 val batch loss: tf.Tensor(0.5311576, shape=(), dtype=float32)\n",
            "epoch: 4 step: 194 val batch loss: tf.Tensor(0.5680608, shape=(), dtype=float32)\n",
            "epoch: 4 step: 195 val batch loss: tf.Tensor(0.6207078, shape=(), dtype=float32)\n",
            "epoch: 4 step: 196 val batch loss: tf.Tensor(0.63416946, shape=(), dtype=float32)\n",
            "epoch: 4 step: 197 val batch loss: tf.Tensor(0.62408787, shape=(), dtype=float32)\n",
            "epoch: 4 step: 198 val batch loss: tf.Tensor(0.5544662, shape=(), dtype=float32)\n",
            "epoch: 4 step: 199 val batch loss: tf.Tensor(0.4725891, shape=(), dtype=float32)\n",
            "epoch: 4 step: 200 val batch loss: tf.Tensor(0.5382301, shape=(), dtype=float32)\n",
            "epoch: 4 step: 201 val batch loss: tf.Tensor(0.7134424, shape=(), dtype=float32)\n",
            "epoch: 4 step: 202 val batch loss: tf.Tensor(0.55360425, shape=(), dtype=float32)\n",
            "epoch: 4 step: 203 val batch loss: tf.Tensor(0.8903901, shape=(), dtype=float32)\n",
            "epoch: 4 step: 204 val batch loss: tf.Tensor(0.52172816, shape=(), dtype=float32)\n",
            "epoch: 4 step: 205 val batch loss: tf.Tensor(0.5977914, shape=(), dtype=float32)\n",
            "epoch: 4 step: 206 val batch loss: tf.Tensor(0.47407407, shape=(), dtype=float32)\n",
            "epoch: 4 step: 207 val batch loss: tf.Tensor(0.5641916, shape=(), dtype=float32)\n",
            "epoch: 4 step: 208 val batch loss: tf.Tensor(0.49947244, shape=(), dtype=float32)\n",
            "epoch: 4 step: 209 val batch loss: tf.Tensor(0.4715448, shape=(), dtype=float32)\n",
            "epoch: 4 step: 210 val batch loss: tf.Tensor(0.57515985, shape=(), dtype=float32)\n",
            "epoch: 4 step: 211 val batch loss: tf.Tensor(0.52850014, shape=(), dtype=float32)\n",
            "epoch: 4 step: 212 val batch loss: tf.Tensor(0.648533, shape=(), dtype=float32)\n",
            "epoch: 4 step: 213 val batch loss: tf.Tensor(0.74612737, shape=(), dtype=float32)\n",
            "epoch: 4 step: 214 val batch loss: tf.Tensor(0.6323075, shape=(), dtype=float32)\n",
            "epoch: 4 step: 215 val batch loss: tf.Tensor(0.55940837, shape=(), dtype=float32)\n",
            "epoch: 4 step: 216 val batch loss: tf.Tensor(0.52258956, shape=(), dtype=float32)\n",
            "epoch: 4 step: 217 val batch loss: tf.Tensor(0.5611516, shape=(), dtype=float32)\n",
            "epoch: 4 step: 218 val batch loss: tf.Tensor(0.627257, shape=(), dtype=float32)\n",
            "epoch: 4 step: 219 val batch loss: tf.Tensor(0.4173704, shape=(), dtype=float32)\n",
            "epoch: 4 step: 220 val batch loss: tf.Tensor(0.43385416, shape=(), dtype=float32)\n",
            "epoch: 4 step: 221 val batch loss: tf.Tensor(0.7105371, shape=(), dtype=float32)\n",
            "epoch: 4 step: 222 val batch loss: tf.Tensor(0.74716854, shape=(), dtype=float32)\n",
            "epoch: 4 step: 223 val batch loss: tf.Tensor(0.78814876, shape=(), dtype=float32)\n",
            "epoch: 4 step: 224 val batch loss: tf.Tensor(0.72993505, shape=(), dtype=float32)\n",
            "epoch: 4 step: 225 val batch loss: tf.Tensor(0.66639954, shape=(), dtype=float32)\n",
            "epoch: 4 step: 226 val batch loss: tf.Tensor(0.6972708, shape=(), dtype=float32)\n",
            "epoch: 4 step: 227 val batch loss: tf.Tensor(0.59761846, shape=(), dtype=float32)\n",
            "epoch: 4 step: 228 val batch loss: tf.Tensor(0.50082105, shape=(), dtype=float32)\n",
            "epoch: 4 step: 229 val batch loss: tf.Tensor(1.041878, shape=(), dtype=float32)\n",
            "epoch: 4 step: 230 val batch loss: tf.Tensor(0.7466319, shape=(), dtype=float32)\n",
            "epoch: 4 step: 231 val batch loss: tf.Tensor(0.46340632, shape=(), dtype=float32)\n",
            "epoch: 4 step: 232 val batch loss: tf.Tensor(0.7383906, shape=(), dtype=float32)\n",
            "epoch: 4 step: 233 val batch loss: tf.Tensor(0.5784998, shape=(), dtype=float32)\n",
            "epoch: 4 step: 234 val batch loss: tf.Tensor(0.7008325, shape=(), dtype=float32)\n",
            "epoch: 4 step: 235 val batch loss: tf.Tensor(0.59614015, shape=(), dtype=float32)\n",
            "epoch: 4 step: 236 val batch loss: tf.Tensor(0.7288272, shape=(), dtype=float32)\n",
            "epoch: 4 step: 237 val batch loss: tf.Tensor(0.7235815, shape=(), dtype=float32)\n",
            "epoch: 4 step: 238 val batch loss: tf.Tensor(0.6577461, shape=(), dtype=float32)\n",
            "epoch: 4 step: 239 val batch loss: tf.Tensor(0.6185615, shape=(), dtype=float32)\n",
            "epoch: 4 step: 240 val batch loss: tf.Tensor(0.63801944, shape=(), dtype=float32)\n",
            "epoch: 4 step: 241 val batch loss: tf.Tensor(0.6392206, shape=(), dtype=float32)\n",
            "epoch: 4 step: 242 val batch loss: tf.Tensor(0.6091847, shape=(), dtype=float32)\n",
            "epoch: 4 step: 243 val batch loss: tf.Tensor(0.72624195, shape=(), dtype=float32)\n",
            "epoch: 4 step: 244 val batch loss: tf.Tensor(0.5627906, shape=(), dtype=float32)\n",
            "epoch: 4 step: 245 val batch loss: tf.Tensor(0.43774423, shape=(), dtype=float32)\n",
            "epoch: 4 step: 246 val batch loss: tf.Tensor(0.48218828, shape=(), dtype=float32)\n",
            "epoch: 4 step: 247 val batch loss: tf.Tensor(0.72277755, shape=(), dtype=float32)\n",
            "epoch: 4 step: 248 val batch loss: tf.Tensor(0.8423789, shape=(), dtype=float32)\n",
            "epoch: 4 step: 249 val batch loss: tf.Tensor(0.6897768, shape=(), dtype=float32)\n",
            "Epoch 4 training loss: 0.3637731373310089 val loss: 0.6327879428863525\n",
            "epoch: 5 step: 0 train batch loss: tf.Tensor(0.31027183, shape=(), dtype=float32)\n",
            "epoch: 5 step: 1 train batch loss: tf.Tensor(0.3128392, shape=(), dtype=float32)\n",
            "epoch: 5 step: 2 train batch loss: tf.Tensor(0.35671505, shape=(), dtype=float32)\n",
            "epoch: 5 step: 3 train batch loss: tf.Tensor(0.27122372, shape=(), dtype=float32)\n",
            "epoch: 5 step: 4 train batch loss: tf.Tensor(0.41021943, shape=(), dtype=float32)\n",
            "epoch: 5 step: 5 train batch loss: tf.Tensor(0.27254763, shape=(), dtype=float32)\n",
            "epoch: 5 step: 6 train batch loss: tf.Tensor(0.3435353, shape=(), dtype=float32)\n",
            "epoch: 5 step: 7 train batch loss: tf.Tensor(0.32557684, shape=(), dtype=float32)\n",
            "epoch: 5 step: 8 train batch loss: tf.Tensor(0.31568113, shape=(), dtype=float32)\n",
            "epoch: 5 step: 9 train batch loss: tf.Tensor(0.36390957, shape=(), dtype=float32)\n",
            "epoch: 5 step: 10 train batch loss: tf.Tensor(0.35648522, shape=(), dtype=float32)\n",
            "epoch: 5 step: 11 train batch loss: tf.Tensor(0.36291164, shape=(), dtype=float32)\n",
            "epoch: 5 step: 12 train batch loss: tf.Tensor(0.33283824, shape=(), dtype=float32)\n",
            "epoch: 5 step: 13 train batch loss: tf.Tensor(0.3242086, shape=(), dtype=float32)\n",
            "epoch: 5 step: 14 train batch loss: tf.Tensor(0.32845658, shape=(), dtype=float32)\n",
            "epoch: 5 step: 15 train batch loss: tf.Tensor(0.2892029, shape=(), dtype=float32)\n",
            "epoch: 5 step: 16 train batch loss: tf.Tensor(0.38413316, shape=(), dtype=float32)\n",
            "epoch: 5 step: 17 train batch loss: tf.Tensor(0.3388662, shape=(), dtype=float32)\n",
            "epoch: 5 step: 18 train batch loss: tf.Tensor(0.2873147, shape=(), dtype=float32)\n",
            "epoch: 5 step: 19 train batch loss: tf.Tensor(0.30337802, shape=(), dtype=float32)\n",
            "epoch: 5 step: 20 train batch loss: tf.Tensor(0.36572137, shape=(), dtype=float32)\n",
            "epoch: 5 step: 21 train batch loss: tf.Tensor(0.33385515, shape=(), dtype=float32)\n",
            "epoch: 5 step: 22 train batch loss: tf.Tensor(0.3271587, shape=(), dtype=float32)\n",
            "epoch: 5 step: 23 train batch loss: tf.Tensor(0.2901895, shape=(), dtype=float32)\n",
            "epoch: 5 step: 24 train batch loss: tf.Tensor(0.3555142, shape=(), dtype=float32)\n",
            "epoch: 5 step: 25 train batch loss: tf.Tensor(0.30004492, shape=(), dtype=float32)\n",
            "epoch: 5 step: 26 train batch loss: tf.Tensor(0.33429283, shape=(), dtype=float32)\n",
            "epoch: 5 step: 27 train batch loss: tf.Tensor(0.27317828, shape=(), dtype=float32)\n",
            "epoch: 5 step: 28 train batch loss: tf.Tensor(0.34581882, shape=(), dtype=float32)\n",
            "epoch: 5 step: 29 train batch loss: tf.Tensor(0.29127192, shape=(), dtype=float32)\n",
            "epoch: 5 step: 30 train batch loss: tf.Tensor(0.3434422, shape=(), dtype=float32)\n",
            "epoch: 5 step: 31 train batch loss: tf.Tensor(0.35355744, shape=(), dtype=float32)\n",
            "epoch: 5 step: 32 train batch loss: tf.Tensor(0.3020211, shape=(), dtype=float32)\n",
            "epoch: 5 step: 33 train batch loss: tf.Tensor(0.26336473, shape=(), dtype=float32)\n",
            "epoch: 5 step: 34 train batch loss: tf.Tensor(0.30354533, shape=(), dtype=float32)\n",
            "epoch: 5 step: 35 train batch loss: tf.Tensor(0.3184616, shape=(), dtype=float32)\n",
            "epoch: 5 step: 36 train batch loss: tf.Tensor(0.26480025, shape=(), dtype=float32)\n",
            "epoch: 5 step: 37 train batch loss: tf.Tensor(0.296128, shape=(), dtype=float32)\n",
            "epoch: 5 step: 38 train batch loss: tf.Tensor(0.30483606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 39 train batch loss: tf.Tensor(0.37345305, shape=(), dtype=float32)\n",
            "epoch: 5 step: 40 train batch loss: tf.Tensor(0.5165689, shape=(), dtype=float32)\n",
            "epoch: 5 step: 41 train batch loss: tf.Tensor(0.3563308, shape=(), dtype=float32)\n",
            "epoch: 5 step: 42 train batch loss: tf.Tensor(0.3491207, shape=(), dtype=float32)\n",
            "epoch: 5 step: 43 train batch loss: tf.Tensor(0.3890457, shape=(), dtype=float32)\n",
            "epoch: 5 step: 44 train batch loss: tf.Tensor(0.28744408, shape=(), dtype=float32)\n",
            "epoch: 5 step: 45 train batch loss: tf.Tensor(0.29795608, shape=(), dtype=float32)\n",
            "epoch: 5 step: 46 train batch loss: tf.Tensor(0.30119342, shape=(), dtype=float32)\n",
            "epoch: 5 step: 47 train batch loss: tf.Tensor(0.28046694, shape=(), dtype=float32)\n",
            "epoch: 5 step: 48 train batch loss: tf.Tensor(0.38288742, shape=(), dtype=float32)\n",
            "epoch: 5 step: 49 train batch loss: tf.Tensor(0.3905703, shape=(), dtype=float32)\n",
            "epoch: 5 step: 50 train batch loss: tf.Tensor(0.34899014, shape=(), dtype=float32)\n",
            "epoch: 5 step: 51 train batch loss: tf.Tensor(0.30832654, shape=(), dtype=float32)\n",
            "epoch: 5 step: 52 train batch loss: tf.Tensor(0.30829728, shape=(), dtype=float32)\n",
            "epoch: 5 step: 53 train batch loss: tf.Tensor(0.2747585, shape=(), dtype=float32)\n",
            "epoch: 5 step: 54 train batch loss: tf.Tensor(0.34219146, shape=(), dtype=float32)\n",
            "epoch: 5 step: 55 train batch loss: tf.Tensor(0.3372388, shape=(), dtype=float32)\n",
            "epoch: 5 step: 56 train batch loss: tf.Tensor(0.29611224, shape=(), dtype=float32)\n",
            "epoch: 5 step: 57 train batch loss: tf.Tensor(0.24213329, shape=(), dtype=float32)\n",
            "epoch: 5 step: 58 train batch loss: tf.Tensor(0.3605178, shape=(), dtype=float32)\n",
            "epoch: 5 step: 59 train batch loss: tf.Tensor(0.37011757, shape=(), dtype=float32)\n",
            "epoch: 5 step: 60 train batch loss: tf.Tensor(0.3344092, shape=(), dtype=float32)\n",
            "epoch: 5 step: 61 train batch loss: tf.Tensor(0.35959333, shape=(), dtype=float32)\n",
            "epoch: 5 step: 62 train batch loss: tf.Tensor(0.34587878, shape=(), dtype=float32)\n",
            "epoch: 5 step: 63 train batch loss: tf.Tensor(0.24947095, shape=(), dtype=float32)\n",
            "epoch: 5 step: 64 train batch loss: tf.Tensor(0.35203183, shape=(), dtype=float32)\n",
            "epoch: 5 step: 65 train batch loss: tf.Tensor(0.33424482, shape=(), dtype=float32)\n",
            "epoch: 5 step: 66 train batch loss: tf.Tensor(0.27295527, shape=(), dtype=float32)\n",
            "epoch: 5 step: 67 train batch loss: tf.Tensor(0.24055927, shape=(), dtype=float32)\n",
            "epoch: 5 step: 68 train batch loss: tf.Tensor(0.36504182, shape=(), dtype=float32)\n",
            "epoch: 5 step: 69 train batch loss: tf.Tensor(0.3335291, shape=(), dtype=float32)\n",
            "epoch: 5 step: 70 train batch loss: tf.Tensor(0.31035137, shape=(), dtype=float32)\n",
            "epoch: 5 step: 71 train batch loss: tf.Tensor(0.30535606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 72 train batch loss: tf.Tensor(0.439545, shape=(), dtype=float32)\n",
            "epoch: 5 step: 73 train batch loss: tf.Tensor(0.21402243, shape=(), dtype=float32)\n",
            "epoch: 5 step: 74 train batch loss: tf.Tensor(0.38617986, shape=(), dtype=float32)\n",
            "epoch: 5 step: 75 train batch loss: tf.Tensor(0.29487067, shape=(), dtype=float32)\n",
            "epoch: 5 step: 76 train batch loss: tf.Tensor(0.34029445, shape=(), dtype=float32)\n",
            "epoch: 5 step: 77 train batch loss: tf.Tensor(0.336847, shape=(), dtype=float32)\n",
            "epoch: 5 step: 78 train batch loss: tf.Tensor(0.24349761, shape=(), dtype=float32)\n",
            "epoch: 5 step: 79 train batch loss: tf.Tensor(0.25773785, shape=(), dtype=float32)\n",
            "epoch: 5 step: 80 train batch loss: tf.Tensor(0.29801613, shape=(), dtype=float32)\n",
            "epoch: 5 step: 81 train batch loss: tf.Tensor(0.297215, shape=(), dtype=float32)\n",
            "epoch: 5 step: 82 train batch loss: tf.Tensor(0.33726183, shape=(), dtype=float32)\n",
            "epoch: 5 step: 83 train batch loss: tf.Tensor(0.32219487, shape=(), dtype=float32)\n",
            "epoch: 5 step: 84 train batch loss: tf.Tensor(0.36004204, shape=(), dtype=float32)\n",
            "epoch: 5 step: 85 train batch loss: tf.Tensor(0.314048, shape=(), dtype=float32)\n",
            "epoch: 5 step: 86 train batch loss: tf.Tensor(0.3347755, shape=(), dtype=float32)\n",
            "epoch: 5 step: 87 train batch loss: tf.Tensor(0.33423635, shape=(), dtype=float32)\n",
            "epoch: 5 step: 88 train batch loss: tf.Tensor(0.27328038, shape=(), dtype=float32)\n",
            "epoch: 5 step: 89 train batch loss: tf.Tensor(0.3142199, shape=(), dtype=float32)\n",
            "epoch: 5 step: 90 train batch loss: tf.Tensor(0.28207096, shape=(), dtype=float32)\n",
            "epoch: 5 step: 91 train batch loss: tf.Tensor(0.37192267, shape=(), dtype=float32)\n",
            "epoch: 5 step: 92 train batch loss: tf.Tensor(0.4309606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 93 train batch loss: tf.Tensor(0.2522562, shape=(), dtype=float32)\n",
            "epoch: 5 step: 94 train batch loss: tf.Tensor(0.3063808, shape=(), dtype=float32)\n",
            "epoch: 5 step: 95 train batch loss: tf.Tensor(0.28689334, shape=(), dtype=float32)\n",
            "epoch: 5 step: 96 train batch loss: tf.Tensor(0.31720084, shape=(), dtype=float32)\n",
            "epoch: 5 step: 97 train batch loss: tf.Tensor(0.28060418, shape=(), dtype=float32)\n",
            "epoch: 5 step: 98 train batch loss: tf.Tensor(0.27426046, shape=(), dtype=float32)\n",
            "epoch: 5 step: 99 train batch loss: tf.Tensor(0.32744747, shape=(), dtype=float32)\n",
            "epoch: 5 step: 100 train batch loss: tf.Tensor(0.30594128, shape=(), dtype=float32)\n",
            "epoch: 5 step: 101 train batch loss: tf.Tensor(0.2778229, shape=(), dtype=float32)\n",
            "epoch: 5 step: 102 train batch loss: tf.Tensor(0.37221965, shape=(), dtype=float32)\n",
            "epoch: 5 step: 103 train batch loss: tf.Tensor(0.34119403, shape=(), dtype=float32)\n",
            "epoch: 5 step: 104 train batch loss: tf.Tensor(0.36483237, shape=(), dtype=float32)\n",
            "epoch: 5 step: 105 train batch loss: tf.Tensor(0.31448165, shape=(), dtype=float32)\n",
            "epoch: 5 step: 106 train batch loss: tf.Tensor(0.30801743, shape=(), dtype=float32)\n",
            "epoch: 5 step: 107 train batch loss: tf.Tensor(0.3455612, shape=(), dtype=float32)\n",
            "epoch: 5 step: 108 train batch loss: tf.Tensor(0.26452425, shape=(), dtype=float32)\n",
            "epoch: 5 step: 109 train batch loss: tf.Tensor(0.33179712, shape=(), dtype=float32)\n",
            "epoch: 5 step: 110 train batch loss: tf.Tensor(0.30237257, shape=(), dtype=float32)\n",
            "epoch: 5 step: 111 train batch loss: tf.Tensor(0.29636046, shape=(), dtype=float32)\n",
            "epoch: 5 step: 112 train batch loss: tf.Tensor(0.36856446, shape=(), dtype=float32)\n",
            "epoch: 5 step: 113 train batch loss: tf.Tensor(0.2581196, shape=(), dtype=float32)\n",
            "epoch: 5 step: 114 train batch loss: tf.Tensor(0.2680958, shape=(), dtype=float32)\n",
            "epoch: 5 step: 115 train batch loss: tf.Tensor(0.30655995, shape=(), dtype=float32)\n",
            "epoch: 5 step: 116 train batch loss: tf.Tensor(0.27497754, shape=(), dtype=float32)\n",
            "epoch: 5 step: 117 train batch loss: tf.Tensor(0.31706834, shape=(), dtype=float32)\n",
            "epoch: 5 step: 118 train batch loss: tf.Tensor(0.3772835, shape=(), dtype=float32)\n",
            "epoch: 5 step: 119 train batch loss: tf.Tensor(0.29608408, shape=(), dtype=float32)\n",
            "epoch: 5 step: 120 train batch loss: tf.Tensor(0.24990043, shape=(), dtype=float32)\n",
            "epoch: 5 step: 121 train batch loss: tf.Tensor(0.26399404, shape=(), dtype=float32)\n",
            "epoch: 5 step: 122 train batch loss: tf.Tensor(0.2657588, shape=(), dtype=float32)\n",
            "epoch: 5 step: 123 train batch loss: tf.Tensor(0.32188997, shape=(), dtype=float32)\n",
            "epoch: 5 step: 124 train batch loss: tf.Tensor(0.34081444, shape=(), dtype=float32)\n",
            "epoch: 5 step: 125 train batch loss: tf.Tensor(0.31007347, shape=(), dtype=float32)\n",
            "epoch: 5 step: 126 train batch loss: tf.Tensor(0.28596792, shape=(), dtype=float32)\n",
            "epoch: 5 step: 127 train batch loss: tf.Tensor(0.41769162, shape=(), dtype=float32)\n",
            "epoch: 5 step: 128 train batch loss: tf.Tensor(0.32854065, shape=(), dtype=float32)\n",
            "epoch: 5 step: 129 train batch loss: tf.Tensor(0.27710578, shape=(), dtype=float32)\n",
            "epoch: 5 step: 130 train batch loss: tf.Tensor(0.32933915, shape=(), dtype=float32)\n",
            "epoch: 5 step: 131 train batch loss: tf.Tensor(0.30211514, shape=(), dtype=float32)\n",
            "epoch: 5 step: 132 train batch loss: tf.Tensor(0.25899106, shape=(), dtype=float32)\n",
            "epoch: 5 step: 133 train batch loss: tf.Tensor(0.25124103, shape=(), dtype=float32)\n",
            "epoch: 5 step: 134 train batch loss: tf.Tensor(0.25827393, shape=(), dtype=float32)\n",
            "epoch: 5 step: 135 train batch loss: tf.Tensor(0.2974219, shape=(), dtype=float32)\n",
            "epoch: 5 step: 136 train batch loss: tf.Tensor(0.38462934, shape=(), dtype=float32)\n",
            "epoch: 5 step: 137 train batch loss: tf.Tensor(0.29840878, shape=(), dtype=float32)\n",
            "epoch: 5 step: 138 train batch loss: tf.Tensor(0.3419163, shape=(), dtype=float32)\n",
            "epoch: 5 step: 139 train batch loss: tf.Tensor(0.23719224, shape=(), dtype=float32)\n",
            "epoch: 5 step: 140 train batch loss: tf.Tensor(0.27766794, shape=(), dtype=float32)\n",
            "epoch: 5 step: 141 train batch loss: tf.Tensor(0.374171, shape=(), dtype=float32)\n",
            "epoch: 5 step: 142 train batch loss: tf.Tensor(0.3007725, shape=(), dtype=float32)\n",
            "epoch: 5 step: 143 train batch loss: tf.Tensor(0.36855942, shape=(), dtype=float32)\n",
            "epoch: 5 step: 144 train batch loss: tf.Tensor(0.31218582, shape=(), dtype=float32)\n",
            "epoch: 5 step: 145 train batch loss: tf.Tensor(0.34672183, shape=(), dtype=float32)\n",
            "epoch: 5 step: 146 train batch loss: tf.Tensor(0.2660757, shape=(), dtype=float32)\n",
            "epoch: 5 step: 147 train batch loss: tf.Tensor(0.32429615, shape=(), dtype=float32)\n",
            "epoch: 5 step: 148 train batch loss: tf.Tensor(0.2695014, shape=(), dtype=float32)\n",
            "epoch: 5 step: 149 train batch loss: tf.Tensor(0.3235759, shape=(), dtype=float32)\n",
            "epoch: 5 step: 150 train batch loss: tf.Tensor(0.25063837, shape=(), dtype=float32)\n",
            "epoch: 5 step: 151 train batch loss: tf.Tensor(0.27960488, shape=(), dtype=float32)\n",
            "epoch: 5 step: 152 train batch loss: tf.Tensor(0.26536056, shape=(), dtype=float32)\n",
            "epoch: 5 step: 153 train batch loss: tf.Tensor(0.28360415, shape=(), dtype=float32)\n",
            "epoch: 5 step: 154 train batch loss: tf.Tensor(0.31592372, shape=(), dtype=float32)\n",
            "epoch: 5 step: 155 train batch loss: tf.Tensor(0.29387325, shape=(), dtype=float32)\n",
            "epoch: 5 step: 156 train batch loss: tf.Tensor(0.26400387, shape=(), dtype=float32)\n",
            "epoch: 5 step: 157 train batch loss: tf.Tensor(0.2634308, shape=(), dtype=float32)\n",
            "epoch: 5 step: 158 train batch loss: tf.Tensor(0.34915808, shape=(), dtype=float32)\n",
            "epoch: 5 step: 159 train batch loss: tf.Tensor(0.33411765, shape=(), dtype=float32)\n",
            "epoch: 5 step: 160 train batch loss: tf.Tensor(0.36285684, shape=(), dtype=float32)\n",
            "epoch: 5 step: 161 train batch loss: tf.Tensor(0.305693, shape=(), dtype=float32)\n",
            "epoch: 5 step: 162 train batch loss: tf.Tensor(0.29709542, shape=(), dtype=float32)\n",
            "epoch: 5 step: 163 train batch loss: tf.Tensor(0.3049592, shape=(), dtype=float32)\n",
            "epoch: 5 step: 164 train batch loss: tf.Tensor(0.31546968, shape=(), dtype=float32)\n",
            "epoch: 5 step: 165 train batch loss: tf.Tensor(0.3272348, shape=(), dtype=float32)\n",
            "epoch: 5 step: 166 train batch loss: tf.Tensor(0.30235058, shape=(), dtype=float32)\n",
            "epoch: 5 step: 167 train batch loss: tf.Tensor(0.27868286, shape=(), dtype=float32)\n",
            "epoch: 5 step: 168 train batch loss: tf.Tensor(0.30524647, shape=(), dtype=float32)\n",
            "epoch: 5 step: 169 train batch loss: tf.Tensor(0.27308446, shape=(), dtype=float32)\n",
            "epoch: 5 step: 170 train batch loss: tf.Tensor(0.33214772, shape=(), dtype=float32)\n",
            "epoch: 5 step: 171 train batch loss: tf.Tensor(0.28774595, shape=(), dtype=float32)\n",
            "epoch: 5 step: 172 train batch loss: tf.Tensor(0.26948702, shape=(), dtype=float32)\n",
            "epoch: 5 step: 173 train batch loss: tf.Tensor(0.28698793, shape=(), dtype=float32)\n",
            "epoch: 5 step: 174 train batch loss: tf.Tensor(0.2537085, shape=(), dtype=float32)\n",
            "epoch: 5 step: 175 train batch loss: tf.Tensor(0.36735168, shape=(), dtype=float32)\n",
            "epoch: 5 step: 176 train batch loss: tf.Tensor(0.3046825, shape=(), dtype=float32)\n",
            "epoch: 5 step: 177 train batch loss: tf.Tensor(0.33641115, shape=(), dtype=float32)\n",
            "epoch: 5 step: 178 train batch loss: tf.Tensor(0.26874313, shape=(), dtype=float32)\n",
            "epoch: 5 step: 179 train batch loss: tf.Tensor(0.25498384, shape=(), dtype=float32)\n",
            "epoch: 5 step: 180 train batch loss: tf.Tensor(0.29417345, shape=(), dtype=float32)\n",
            "epoch: 5 step: 181 train batch loss: tf.Tensor(0.2908824, shape=(), dtype=float32)\n",
            "epoch: 5 step: 182 train batch loss: tf.Tensor(0.2536185, shape=(), dtype=float32)\n",
            "epoch: 5 step: 183 train batch loss: tf.Tensor(0.2112641, shape=(), dtype=float32)\n",
            "epoch: 5 step: 184 train batch loss: tf.Tensor(0.27541155, shape=(), dtype=float32)\n",
            "epoch: 5 step: 185 train batch loss: tf.Tensor(0.30609038, shape=(), dtype=float32)\n",
            "epoch: 5 step: 186 train batch loss: tf.Tensor(0.25397116, shape=(), dtype=float32)\n",
            "epoch: 5 step: 187 train batch loss: tf.Tensor(0.2920909, shape=(), dtype=float32)\n",
            "epoch: 5 step: 188 train batch loss: tf.Tensor(0.2751519, shape=(), dtype=float32)\n",
            "epoch: 5 step: 189 train batch loss: tf.Tensor(0.3162679, shape=(), dtype=float32)\n",
            "epoch: 5 step: 190 train batch loss: tf.Tensor(0.32880545, shape=(), dtype=float32)\n",
            "epoch: 5 step: 191 train batch loss: tf.Tensor(0.25707674, shape=(), dtype=float32)\n",
            "epoch: 5 step: 192 train batch loss: tf.Tensor(0.34893665, shape=(), dtype=float32)\n",
            "epoch: 5 step: 193 train batch loss: tf.Tensor(0.31996164, shape=(), dtype=float32)\n",
            "epoch: 5 step: 194 train batch loss: tf.Tensor(0.23638389, shape=(), dtype=float32)\n",
            "epoch: 5 step: 195 train batch loss: tf.Tensor(0.3520156, shape=(), dtype=float32)\n",
            "epoch: 5 step: 196 train batch loss: tf.Tensor(0.30580363, shape=(), dtype=float32)\n",
            "epoch: 5 step: 197 train batch loss: tf.Tensor(0.3499068, shape=(), dtype=float32)\n",
            "epoch: 5 step: 198 train batch loss: tf.Tensor(0.312091, shape=(), dtype=float32)\n",
            "epoch: 5 step: 199 train batch loss: tf.Tensor(0.25081444, shape=(), dtype=float32)\n",
            "epoch: 5 step: 200 train batch loss: tf.Tensor(0.29069597, shape=(), dtype=float32)\n",
            "epoch: 5 step: 201 train batch loss: tf.Tensor(0.26508218, shape=(), dtype=float32)\n",
            "epoch: 5 step: 202 train batch loss: tf.Tensor(0.540446, shape=(), dtype=float32)\n",
            "epoch: 5 step: 203 train batch loss: tf.Tensor(0.27219036, shape=(), dtype=float32)\n",
            "epoch: 5 step: 204 train batch loss: tf.Tensor(0.3210774, shape=(), dtype=float32)\n",
            "epoch: 5 step: 205 train batch loss: tf.Tensor(0.28137383, shape=(), dtype=float32)\n",
            "epoch: 5 step: 206 train batch loss: tf.Tensor(0.333292, shape=(), dtype=float32)\n",
            "epoch: 5 step: 207 train batch loss: tf.Tensor(0.31769976, shape=(), dtype=float32)\n",
            "epoch: 5 step: 208 train batch loss: tf.Tensor(0.32151228, shape=(), dtype=float32)\n",
            "epoch: 5 step: 209 train batch loss: tf.Tensor(0.34148026, shape=(), dtype=float32)\n",
            "epoch: 5 step: 210 train batch loss: tf.Tensor(0.32702702, shape=(), dtype=float32)\n",
            "epoch: 5 step: 211 train batch loss: tf.Tensor(0.2593469, shape=(), dtype=float32)\n",
            "epoch: 5 step: 212 train batch loss: tf.Tensor(0.23137489, shape=(), dtype=float32)\n",
            "epoch: 5 step: 213 train batch loss: tf.Tensor(0.29171568, shape=(), dtype=float32)\n",
            "epoch: 5 step: 214 train batch loss: tf.Tensor(0.32620496, shape=(), dtype=float32)\n",
            "epoch: 5 step: 215 train batch loss: tf.Tensor(0.23905158, shape=(), dtype=float32)\n",
            "epoch: 5 step: 216 train batch loss: tf.Tensor(0.4095057, shape=(), dtype=float32)\n",
            "epoch: 5 step: 217 train batch loss: tf.Tensor(0.2503112, shape=(), dtype=float32)\n",
            "epoch: 5 step: 218 train batch loss: tf.Tensor(0.26875556, shape=(), dtype=float32)\n",
            "epoch: 5 step: 219 train batch loss: tf.Tensor(0.28018522, shape=(), dtype=float32)\n",
            "epoch: 5 step: 220 train batch loss: tf.Tensor(0.28983134, shape=(), dtype=float32)\n",
            "epoch: 5 step: 221 train batch loss: tf.Tensor(0.2665245, shape=(), dtype=float32)\n",
            "epoch: 5 step: 222 train batch loss: tf.Tensor(0.29297334, shape=(), dtype=float32)\n",
            "epoch: 5 step: 223 train batch loss: tf.Tensor(0.27019724, shape=(), dtype=float32)\n",
            "epoch: 5 step: 224 train batch loss: tf.Tensor(0.348423, shape=(), dtype=float32)\n",
            "epoch: 5 step: 225 train batch loss: tf.Tensor(0.31596452, shape=(), dtype=float32)\n",
            "epoch: 5 step: 226 train batch loss: tf.Tensor(0.22077623, shape=(), dtype=float32)\n",
            "epoch: 5 step: 227 train batch loss: tf.Tensor(0.27894703, shape=(), dtype=float32)\n",
            "epoch: 5 step: 228 train batch loss: tf.Tensor(0.3070073, shape=(), dtype=float32)\n",
            "epoch: 5 step: 229 train batch loss: tf.Tensor(0.29942754, shape=(), dtype=float32)\n",
            "epoch: 5 step: 230 train batch loss: tf.Tensor(0.25174996, shape=(), dtype=float32)\n",
            "epoch: 5 step: 231 train batch loss: tf.Tensor(0.24706705, shape=(), dtype=float32)\n",
            "epoch: 5 step: 232 train batch loss: tf.Tensor(0.330634, shape=(), dtype=float32)\n",
            "epoch: 5 step: 233 train batch loss: tf.Tensor(0.39949208, shape=(), dtype=float32)\n",
            "epoch: 5 step: 234 train batch loss: tf.Tensor(0.31060734, shape=(), dtype=float32)\n",
            "epoch: 5 step: 235 train batch loss: tf.Tensor(0.2522958, shape=(), dtype=float32)\n",
            "epoch: 5 step: 236 train batch loss: tf.Tensor(0.32827857, shape=(), dtype=float32)\n",
            "epoch: 5 step: 237 train batch loss: tf.Tensor(0.34762338, shape=(), dtype=float32)\n",
            "epoch: 5 step: 238 train batch loss: tf.Tensor(0.30065003, shape=(), dtype=float32)\n",
            "epoch: 5 step: 239 train batch loss: tf.Tensor(0.30739003, shape=(), dtype=float32)\n",
            "epoch: 5 step: 240 train batch loss: tf.Tensor(0.3295139, shape=(), dtype=float32)\n",
            "epoch: 5 step: 241 train batch loss: tf.Tensor(0.368354, shape=(), dtype=float32)\n",
            "epoch: 5 step: 242 train batch loss: tf.Tensor(0.26461208, shape=(), dtype=float32)\n",
            "epoch: 5 step: 243 train batch loss: tf.Tensor(0.24144581, shape=(), dtype=float32)\n",
            "epoch: 5 step: 244 train batch loss: tf.Tensor(0.3451456, shape=(), dtype=float32)\n",
            "epoch: 5 step: 245 train batch loss: tf.Tensor(0.27060392, shape=(), dtype=float32)\n",
            "epoch: 5 step: 246 train batch loss: tf.Tensor(0.23347764, shape=(), dtype=float32)\n",
            "epoch: 5 step: 247 train batch loss: tf.Tensor(0.305972, shape=(), dtype=float32)\n",
            "epoch: 5 step: 248 train batch loss: tf.Tensor(0.33802158, shape=(), dtype=float32)\n",
            "epoch: 5 step: 249 train batch loss: tf.Tensor(0.25808123, shape=(), dtype=float32)\n",
            "epoch: 5 step: 250 train batch loss: tf.Tensor(0.2845737, shape=(), dtype=float32)\n",
            "epoch: 5 step: 251 train batch loss: tf.Tensor(0.32232207, shape=(), dtype=float32)\n",
            "epoch: 5 step: 252 train batch loss: tf.Tensor(0.3845742, shape=(), dtype=float32)\n",
            "epoch: 5 step: 253 train batch loss: tf.Tensor(0.26979947, shape=(), dtype=float32)\n",
            "epoch: 5 step: 254 train batch loss: tf.Tensor(0.31348974, shape=(), dtype=float32)\n",
            "epoch: 5 step: 255 train batch loss: tf.Tensor(0.36472654, shape=(), dtype=float32)\n",
            "epoch: 5 step: 256 train batch loss: tf.Tensor(0.2665179, shape=(), dtype=float32)\n",
            "epoch: 5 step: 257 train batch loss: tf.Tensor(0.2692737, shape=(), dtype=float32)\n",
            "epoch: 5 step: 258 train batch loss: tf.Tensor(0.25369638, shape=(), dtype=float32)\n",
            "epoch: 5 step: 259 train batch loss: tf.Tensor(0.2688733, shape=(), dtype=float32)\n",
            "epoch: 5 step: 260 train batch loss: tf.Tensor(0.33946654, shape=(), dtype=float32)\n",
            "epoch: 5 step: 261 train batch loss: tf.Tensor(0.2857939, shape=(), dtype=float32)\n",
            "epoch: 5 step: 262 train batch loss: tf.Tensor(0.38316077, shape=(), dtype=float32)\n",
            "epoch: 5 step: 263 train batch loss: tf.Tensor(0.31514338, shape=(), dtype=float32)\n",
            "epoch: 5 step: 264 train batch loss: tf.Tensor(0.30984113, shape=(), dtype=float32)\n",
            "epoch: 5 step: 265 train batch loss: tf.Tensor(0.32847506, shape=(), dtype=float32)\n",
            "epoch: 5 step: 266 train batch loss: tf.Tensor(0.30378267, shape=(), dtype=float32)\n",
            "epoch: 5 step: 267 train batch loss: tf.Tensor(0.266876, shape=(), dtype=float32)\n",
            "epoch: 5 step: 268 train batch loss: tf.Tensor(0.27154607, shape=(), dtype=float32)\n",
            "epoch: 5 step: 269 train batch loss: tf.Tensor(0.285124, shape=(), dtype=float32)\n",
            "epoch: 5 step: 270 train batch loss: tf.Tensor(0.3034237, shape=(), dtype=float32)\n",
            "epoch: 5 step: 271 train batch loss: tf.Tensor(0.3140289, shape=(), dtype=float32)\n",
            "epoch: 5 step: 272 train batch loss: tf.Tensor(0.25776058, shape=(), dtype=float32)\n",
            "epoch: 5 step: 273 train batch loss: tf.Tensor(0.26905158, shape=(), dtype=float32)\n",
            "epoch: 5 step: 274 train batch loss: tf.Tensor(0.47986946, shape=(), dtype=float32)\n",
            "epoch: 5 step: 275 train batch loss: tf.Tensor(0.23580678, shape=(), dtype=float32)\n",
            "epoch: 5 step: 276 train batch loss: tf.Tensor(0.3510031, shape=(), dtype=float32)\n",
            "epoch: 5 step: 277 train batch loss: tf.Tensor(0.32064068, shape=(), dtype=float32)\n",
            "epoch: 5 step: 278 train batch loss: tf.Tensor(0.363141, shape=(), dtype=float32)\n",
            "epoch: 5 step: 279 train batch loss: tf.Tensor(0.2707559, shape=(), dtype=float32)\n",
            "epoch: 5 step: 280 train batch loss: tf.Tensor(0.29312375, shape=(), dtype=float32)\n",
            "epoch: 5 step: 281 train batch loss: tf.Tensor(0.35591412, shape=(), dtype=float32)\n",
            "epoch: 5 step: 282 train batch loss: tf.Tensor(0.27572495, shape=(), dtype=float32)\n",
            "epoch: 5 step: 283 train batch loss: tf.Tensor(0.29365185, shape=(), dtype=float32)\n",
            "epoch: 5 step: 284 train batch loss: tf.Tensor(0.26505503, shape=(), dtype=float32)\n",
            "epoch: 5 step: 285 train batch loss: tf.Tensor(0.30137652, shape=(), dtype=float32)\n",
            "epoch: 5 step: 286 train batch loss: tf.Tensor(0.25841543, shape=(), dtype=float32)\n",
            "epoch: 5 step: 287 train batch loss: tf.Tensor(0.27042407, shape=(), dtype=float32)\n",
            "epoch: 5 step: 288 train batch loss: tf.Tensor(0.291185, shape=(), dtype=float32)\n",
            "epoch: 5 step: 289 train batch loss: tf.Tensor(0.23411337, shape=(), dtype=float32)\n",
            "epoch: 5 step: 290 train batch loss: tf.Tensor(0.33956015, shape=(), dtype=float32)\n",
            "epoch: 5 step: 291 train batch loss: tf.Tensor(0.39932403, shape=(), dtype=float32)\n",
            "epoch: 5 step: 292 train batch loss: tf.Tensor(0.38666272, shape=(), dtype=float32)\n",
            "epoch: 5 step: 293 train batch loss: tf.Tensor(0.36185354, shape=(), dtype=float32)\n",
            "epoch: 5 step: 294 train batch loss: tf.Tensor(0.29920474, shape=(), dtype=float32)\n",
            "epoch: 5 step: 295 train batch loss: tf.Tensor(0.2936395, shape=(), dtype=float32)\n",
            "epoch: 5 step: 296 train batch loss: tf.Tensor(0.27536762, shape=(), dtype=float32)\n",
            "epoch: 5 step: 297 train batch loss: tf.Tensor(0.27900964, shape=(), dtype=float32)\n",
            "epoch: 5 step: 298 train batch loss: tf.Tensor(0.27604413, shape=(), dtype=float32)\n",
            "epoch: 5 step: 299 train batch loss: tf.Tensor(0.26301992, shape=(), dtype=float32)\n",
            "epoch: 5 step: 300 train batch loss: tf.Tensor(0.3044697, shape=(), dtype=float32)\n",
            "epoch: 5 step: 301 train batch loss: tf.Tensor(0.26585266, shape=(), dtype=float32)\n",
            "epoch: 5 step: 302 train batch loss: tf.Tensor(0.2314978, shape=(), dtype=float32)\n",
            "epoch: 5 step: 303 train batch loss: tf.Tensor(0.30901155, shape=(), dtype=float32)\n",
            "epoch: 5 step: 304 train batch loss: tf.Tensor(0.31607634, shape=(), dtype=float32)\n",
            "epoch: 5 step: 305 train batch loss: tf.Tensor(0.31257865, shape=(), dtype=float32)\n",
            "epoch: 5 step: 306 train batch loss: tf.Tensor(0.26743764, shape=(), dtype=float32)\n",
            "epoch: 5 step: 307 train batch loss: tf.Tensor(0.2575444, shape=(), dtype=float32)\n",
            "epoch: 5 step: 308 train batch loss: tf.Tensor(0.27327392, shape=(), dtype=float32)\n",
            "epoch: 5 step: 309 train batch loss: tf.Tensor(0.3058861, shape=(), dtype=float32)\n",
            "epoch: 5 step: 310 train batch loss: tf.Tensor(0.33659083, shape=(), dtype=float32)\n",
            "epoch: 5 step: 311 train batch loss: tf.Tensor(0.33770722, shape=(), dtype=float32)\n",
            "epoch: 5 step: 312 train batch loss: tf.Tensor(0.33299628, shape=(), dtype=float32)\n",
            "epoch: 5 step: 313 train batch loss: tf.Tensor(0.29605716, shape=(), dtype=float32)\n",
            "epoch: 5 step: 314 train batch loss: tf.Tensor(0.29786167, shape=(), dtype=float32)\n",
            "epoch: 5 step: 315 train batch loss: tf.Tensor(0.2636752, shape=(), dtype=float32)\n",
            "epoch: 5 step: 316 train batch loss: tf.Tensor(0.24575381, shape=(), dtype=float32)\n",
            "epoch: 5 step: 317 train batch loss: tf.Tensor(0.28918916, shape=(), dtype=float32)\n",
            "epoch: 5 step: 318 train batch loss: tf.Tensor(0.25735548, shape=(), dtype=float32)\n",
            "epoch: 5 step: 319 train batch loss: tf.Tensor(0.31439367, shape=(), dtype=float32)\n",
            "epoch: 5 step: 320 train batch loss: tf.Tensor(0.2871618, shape=(), dtype=float32)\n",
            "epoch: 5 step: 321 train batch loss: tf.Tensor(0.2878596, shape=(), dtype=float32)\n",
            "epoch: 5 step: 322 train batch loss: tf.Tensor(0.27696615, shape=(), dtype=float32)\n",
            "epoch: 5 step: 323 train batch loss: tf.Tensor(0.29241306, shape=(), dtype=float32)\n",
            "epoch: 5 step: 324 train batch loss: tf.Tensor(0.34904632, shape=(), dtype=float32)\n",
            "epoch: 5 step: 325 train batch loss: tf.Tensor(0.35382298, shape=(), dtype=float32)\n",
            "epoch: 5 step: 326 train batch loss: tf.Tensor(0.28520566, shape=(), dtype=float32)\n",
            "epoch: 5 step: 327 train batch loss: tf.Tensor(0.39994556, shape=(), dtype=float32)\n",
            "epoch: 5 step: 328 train batch loss: tf.Tensor(0.26338214, shape=(), dtype=float32)\n",
            "epoch: 5 step: 329 train batch loss: tf.Tensor(0.31510752, shape=(), dtype=float32)\n",
            "epoch: 5 step: 330 train batch loss: tf.Tensor(0.3287532, shape=(), dtype=float32)\n",
            "epoch: 5 step: 331 train batch loss: tf.Tensor(0.24251935, shape=(), dtype=float32)\n",
            "epoch: 5 step: 332 train batch loss: tf.Tensor(0.30223033, shape=(), dtype=float32)\n",
            "epoch: 5 step: 333 train batch loss: tf.Tensor(0.37044606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 334 train batch loss: tf.Tensor(0.27110487, shape=(), dtype=float32)\n",
            "epoch: 5 step: 335 train batch loss: tf.Tensor(0.29316366, shape=(), dtype=float32)\n",
            "epoch: 5 step: 336 train batch loss: tf.Tensor(0.40478057, shape=(), dtype=float32)\n",
            "epoch: 5 step: 337 train batch loss: tf.Tensor(0.21647196, shape=(), dtype=float32)\n",
            "epoch: 5 step: 338 train batch loss: tf.Tensor(0.2551215, shape=(), dtype=float32)\n",
            "epoch: 5 step: 339 train batch loss: tf.Tensor(0.28648743, shape=(), dtype=float32)\n",
            "epoch: 5 step: 340 train batch loss: tf.Tensor(0.2983878, shape=(), dtype=float32)\n",
            "epoch: 5 step: 341 train batch loss: tf.Tensor(0.26611167, shape=(), dtype=float32)\n",
            "epoch: 5 step: 342 train batch loss: tf.Tensor(0.23305233, shape=(), dtype=float32)\n",
            "epoch: 5 step: 343 train batch loss: tf.Tensor(0.21217373, shape=(), dtype=float32)\n",
            "epoch: 5 step: 344 train batch loss: tf.Tensor(0.30014834, shape=(), dtype=float32)\n",
            "epoch: 5 step: 345 train batch loss: tf.Tensor(0.37435254, shape=(), dtype=float32)\n",
            "epoch: 5 step: 346 train batch loss: tf.Tensor(0.2886287, shape=(), dtype=float32)\n",
            "epoch: 5 step: 347 train batch loss: tf.Tensor(0.30347466, shape=(), dtype=float32)\n",
            "epoch: 5 step: 348 train batch loss: tf.Tensor(0.23957899, shape=(), dtype=float32)\n",
            "epoch: 5 step: 349 train batch loss: tf.Tensor(0.2676477, shape=(), dtype=float32)\n",
            "epoch: 5 step: 350 train batch loss: tf.Tensor(0.3286655, shape=(), dtype=float32)\n",
            "epoch: 5 step: 351 train batch loss: tf.Tensor(0.20571753, shape=(), dtype=float32)\n",
            "epoch: 5 step: 352 train batch loss: tf.Tensor(0.28847876, shape=(), dtype=float32)\n",
            "epoch: 5 step: 353 train batch loss: tf.Tensor(0.36061397, shape=(), dtype=float32)\n",
            "epoch: 5 step: 354 train batch loss: tf.Tensor(0.25571698, shape=(), dtype=float32)\n",
            "epoch: 5 step: 355 train batch loss: tf.Tensor(0.27109143, shape=(), dtype=float32)\n",
            "epoch: 5 step: 356 train batch loss: tf.Tensor(0.28706592, shape=(), dtype=float32)\n",
            "epoch: 5 step: 357 train batch loss: tf.Tensor(0.29795465, shape=(), dtype=float32)\n",
            "epoch: 5 step: 358 train batch loss: tf.Tensor(0.3308862, shape=(), dtype=float32)\n",
            "epoch: 5 step: 359 train batch loss: tf.Tensor(0.29220575, shape=(), dtype=float32)\n",
            "epoch: 5 step: 360 train batch loss: tf.Tensor(0.34964368, shape=(), dtype=float32)\n",
            "epoch: 5 step: 361 train batch loss: tf.Tensor(0.29898492, shape=(), dtype=float32)\n",
            "epoch: 5 step: 362 train batch loss: tf.Tensor(0.2836808, shape=(), dtype=float32)\n",
            "epoch: 5 step: 363 train batch loss: tf.Tensor(0.22113112, shape=(), dtype=float32)\n",
            "epoch: 5 step: 364 train batch loss: tf.Tensor(0.2544942, shape=(), dtype=float32)\n",
            "epoch: 5 step: 365 train batch loss: tf.Tensor(0.26461238, shape=(), dtype=float32)\n",
            "epoch: 5 step: 366 train batch loss: tf.Tensor(0.31275836, shape=(), dtype=float32)\n",
            "epoch: 5 step: 367 train batch loss: tf.Tensor(0.25175327, shape=(), dtype=float32)\n",
            "epoch: 5 step: 368 train batch loss: tf.Tensor(0.27827933, shape=(), dtype=float32)\n",
            "epoch: 5 step: 369 train batch loss: tf.Tensor(0.29940233, shape=(), dtype=float32)\n",
            "epoch: 5 step: 370 train batch loss: tf.Tensor(0.27020952, shape=(), dtype=float32)\n",
            "epoch: 5 step: 371 train batch loss: tf.Tensor(0.23590536, shape=(), dtype=float32)\n",
            "epoch: 5 step: 372 train batch loss: tf.Tensor(0.28580594, shape=(), dtype=float32)\n",
            "epoch: 5 step: 373 train batch loss: tf.Tensor(0.26607522, shape=(), dtype=float32)\n",
            "epoch: 5 step: 374 train batch loss: tf.Tensor(0.26190954, shape=(), dtype=float32)\n",
            "epoch: 5 step: 375 train batch loss: tf.Tensor(0.26917866, shape=(), dtype=float32)\n",
            "epoch: 5 step: 376 train batch loss: tf.Tensor(0.30618683, shape=(), dtype=float32)\n",
            "epoch: 5 step: 377 train batch loss: tf.Tensor(0.32273775, shape=(), dtype=float32)\n",
            "epoch: 5 step: 378 train batch loss: tf.Tensor(0.28300375, shape=(), dtype=float32)\n",
            "epoch: 5 step: 379 train batch loss: tf.Tensor(0.31160346, shape=(), dtype=float32)\n",
            "epoch: 5 step: 380 train batch loss: tf.Tensor(0.2748065, shape=(), dtype=float32)\n",
            "epoch: 5 step: 381 train batch loss: tf.Tensor(0.35964602, shape=(), dtype=float32)\n",
            "epoch: 5 step: 382 train batch loss: tf.Tensor(0.2703627, shape=(), dtype=float32)\n",
            "epoch: 5 step: 383 train batch loss: tf.Tensor(0.3646232, shape=(), dtype=float32)\n",
            "epoch: 5 step: 384 train batch loss: tf.Tensor(0.26653856, shape=(), dtype=float32)\n",
            "epoch: 5 step: 385 train batch loss: tf.Tensor(0.30718154, shape=(), dtype=float32)\n",
            "epoch: 5 step: 386 train batch loss: tf.Tensor(0.32741478, shape=(), dtype=float32)\n",
            "epoch: 5 step: 387 train batch loss: tf.Tensor(0.20010217, shape=(), dtype=float32)\n",
            "epoch: 5 step: 388 train batch loss: tf.Tensor(0.27001062, shape=(), dtype=float32)\n",
            "epoch: 5 step: 389 train batch loss: tf.Tensor(0.3052713, shape=(), dtype=float32)\n",
            "epoch: 5 step: 390 train batch loss: tf.Tensor(0.34365383, shape=(), dtype=float32)\n",
            "epoch: 5 step: 391 train batch loss: tf.Tensor(0.31049907, shape=(), dtype=float32)\n",
            "epoch: 5 step: 392 train batch loss: tf.Tensor(0.37503502, shape=(), dtype=float32)\n",
            "epoch: 5 step: 393 train batch loss: tf.Tensor(0.28332132, shape=(), dtype=float32)\n",
            "epoch: 5 step: 394 train batch loss: tf.Tensor(0.31974354, shape=(), dtype=float32)\n",
            "epoch: 5 step: 395 train batch loss: tf.Tensor(0.334053, shape=(), dtype=float32)\n",
            "epoch: 5 step: 396 train batch loss: tf.Tensor(0.2718723, shape=(), dtype=float32)\n",
            "epoch: 5 step: 397 train batch loss: tf.Tensor(0.31406245, shape=(), dtype=float32)\n",
            "epoch: 5 step: 398 train batch loss: tf.Tensor(0.26373333, shape=(), dtype=float32)\n",
            "epoch: 5 step: 399 train batch loss: tf.Tensor(0.3415169, shape=(), dtype=float32)\n",
            "epoch: 5 step: 400 train batch loss: tf.Tensor(0.27480376, shape=(), dtype=float32)\n",
            "epoch: 5 step: 401 train batch loss: tf.Tensor(0.26172107, shape=(), dtype=float32)\n",
            "epoch: 5 step: 402 train batch loss: tf.Tensor(0.34503824, shape=(), dtype=float32)\n",
            "epoch: 5 step: 403 train batch loss: tf.Tensor(0.24632008, shape=(), dtype=float32)\n",
            "epoch: 5 step: 404 train batch loss: tf.Tensor(0.29332834, shape=(), dtype=float32)\n",
            "epoch: 5 step: 405 train batch loss: tf.Tensor(0.23278661, shape=(), dtype=float32)\n",
            "epoch: 5 step: 406 train batch loss: tf.Tensor(0.21710727, shape=(), dtype=float32)\n",
            "epoch: 5 step: 407 train batch loss: tf.Tensor(0.24603042, shape=(), dtype=float32)\n",
            "epoch: 5 step: 408 train batch loss: tf.Tensor(0.26234114, shape=(), dtype=float32)\n",
            "epoch: 5 step: 409 train batch loss: tf.Tensor(0.31449676, shape=(), dtype=float32)\n",
            "epoch: 5 step: 410 train batch loss: tf.Tensor(0.26046047, shape=(), dtype=float32)\n",
            "epoch: 5 step: 411 train batch loss: tf.Tensor(0.32655984, shape=(), dtype=float32)\n",
            "epoch: 5 step: 412 train batch loss: tf.Tensor(0.275191, shape=(), dtype=float32)\n",
            "epoch: 5 step: 413 train batch loss: tf.Tensor(0.28613743, shape=(), dtype=float32)\n",
            "epoch: 5 step: 414 train batch loss: tf.Tensor(0.33872768, shape=(), dtype=float32)\n",
            "epoch: 5 step: 415 train batch loss: tf.Tensor(0.26638818, shape=(), dtype=float32)\n",
            "epoch: 5 step: 416 train batch loss: tf.Tensor(0.28019828, shape=(), dtype=float32)\n",
            "epoch: 5 step: 417 train batch loss: tf.Tensor(0.29670307, shape=(), dtype=float32)\n",
            "epoch: 5 step: 418 train batch loss: tf.Tensor(0.2449418, shape=(), dtype=float32)\n",
            "epoch: 5 step: 419 train batch loss: tf.Tensor(0.28504404, shape=(), dtype=float32)\n",
            "epoch: 5 step: 420 train batch loss: tf.Tensor(0.31827155, shape=(), dtype=float32)\n",
            "epoch: 5 step: 421 train batch loss: tf.Tensor(0.28455606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 422 train batch loss: tf.Tensor(0.25205046, shape=(), dtype=float32)\n",
            "epoch: 5 step: 423 train batch loss: tf.Tensor(0.25454324, shape=(), dtype=float32)\n",
            "epoch: 5 step: 424 train batch loss: tf.Tensor(0.2954872, shape=(), dtype=float32)\n",
            "epoch: 5 step: 425 train batch loss: tf.Tensor(0.2581026, shape=(), dtype=float32)\n",
            "epoch: 5 step: 426 train batch loss: tf.Tensor(0.2997729, shape=(), dtype=float32)\n",
            "epoch: 5 step: 427 train batch loss: tf.Tensor(0.25902158, shape=(), dtype=float32)\n",
            "epoch: 5 step: 428 train batch loss: tf.Tensor(0.24834748, shape=(), dtype=float32)\n",
            "epoch: 5 step: 429 train batch loss: tf.Tensor(0.24017236, shape=(), dtype=float32)\n",
            "epoch: 5 step: 430 train batch loss: tf.Tensor(0.30708784, shape=(), dtype=float32)\n",
            "epoch: 5 step: 431 train batch loss: tf.Tensor(0.29849124, shape=(), dtype=float32)\n",
            "epoch: 5 step: 432 train batch loss: tf.Tensor(0.3102981, shape=(), dtype=float32)\n",
            "epoch: 5 step: 433 train batch loss: tf.Tensor(0.3178349, shape=(), dtype=float32)\n",
            "epoch: 5 step: 434 train batch loss: tf.Tensor(0.291567, shape=(), dtype=float32)\n",
            "epoch: 5 step: 435 train batch loss: tf.Tensor(0.35618567, shape=(), dtype=float32)\n",
            "epoch: 5 step: 436 train batch loss: tf.Tensor(0.28064597, shape=(), dtype=float32)\n",
            "epoch: 5 step: 437 train batch loss: tf.Tensor(0.42927092, shape=(), dtype=float32)\n",
            "epoch: 5 step: 438 train batch loss: tf.Tensor(0.2624037, shape=(), dtype=float32)\n",
            "epoch: 5 step: 439 train batch loss: tf.Tensor(0.30401763, shape=(), dtype=float32)\n",
            "epoch: 5 step: 440 train batch loss: tf.Tensor(0.2532484, shape=(), dtype=float32)\n",
            "epoch: 5 step: 441 train batch loss: tf.Tensor(0.3700114, shape=(), dtype=float32)\n",
            "epoch: 5 step: 442 train batch loss: tf.Tensor(0.35812414, shape=(), dtype=float32)\n",
            "epoch: 5 step: 443 train batch loss: tf.Tensor(0.21795818, shape=(), dtype=float32)\n",
            "epoch: 5 step: 444 train batch loss: tf.Tensor(0.29779607, shape=(), dtype=float32)\n",
            "epoch: 5 step: 445 train batch loss: tf.Tensor(0.3424186, shape=(), dtype=float32)\n",
            "epoch: 5 step: 446 train batch loss: tf.Tensor(0.2672681, shape=(), dtype=float32)\n",
            "epoch: 5 step: 447 train batch loss: tf.Tensor(0.313521, shape=(), dtype=float32)\n",
            "epoch: 5 step: 448 train batch loss: tf.Tensor(0.21625932, shape=(), dtype=float32)\n",
            "epoch: 5 step: 449 train batch loss: tf.Tensor(0.35324502, shape=(), dtype=float32)\n",
            "epoch: 5 step: 450 train batch loss: tf.Tensor(0.25251648, shape=(), dtype=float32)\n",
            "epoch: 5 step: 451 train batch loss: tf.Tensor(0.34094122, shape=(), dtype=float32)\n",
            "epoch: 5 step: 452 train batch loss: tf.Tensor(0.28099853, shape=(), dtype=float32)\n",
            "epoch: 5 step: 453 train batch loss: tf.Tensor(0.29068762, shape=(), dtype=float32)\n",
            "epoch: 5 step: 454 train batch loss: tf.Tensor(0.3092238, shape=(), dtype=float32)\n",
            "epoch: 5 step: 455 train batch loss: tf.Tensor(0.33879247, shape=(), dtype=float32)\n",
            "epoch: 5 step: 456 train batch loss: tf.Tensor(0.23541006, shape=(), dtype=float32)\n",
            "epoch: 5 step: 457 train batch loss: tf.Tensor(0.32904956, shape=(), dtype=float32)\n",
            "epoch: 5 step: 458 train batch loss: tf.Tensor(0.32095057, shape=(), dtype=float32)\n",
            "epoch: 5 step: 459 train batch loss: tf.Tensor(0.28704682, shape=(), dtype=float32)\n",
            "epoch: 5 step: 460 train batch loss: tf.Tensor(0.24595995, shape=(), dtype=float32)\n",
            "epoch: 5 step: 461 train batch loss: tf.Tensor(0.27060643, shape=(), dtype=float32)\n",
            "epoch: 5 step: 462 train batch loss: tf.Tensor(0.30362207, shape=(), dtype=float32)\n",
            "epoch: 5 step: 463 train batch loss: tf.Tensor(0.27129406, shape=(), dtype=float32)\n",
            "epoch: 5 step: 464 train batch loss: tf.Tensor(0.3084607, shape=(), dtype=float32)\n",
            "epoch: 5 step: 465 train batch loss: tf.Tensor(0.26040053, shape=(), dtype=float32)\n",
            "epoch: 5 step: 466 train batch loss: tf.Tensor(0.2865036, shape=(), dtype=float32)\n",
            "epoch: 5 step: 467 train batch loss: tf.Tensor(0.23644547, shape=(), dtype=float32)\n",
            "epoch: 5 step: 468 train batch loss: tf.Tensor(0.33015734, shape=(), dtype=float32)\n",
            "epoch: 5 step: 469 train batch loss: tf.Tensor(0.3396787, shape=(), dtype=float32)\n",
            "epoch: 5 step: 470 train batch loss: tf.Tensor(0.30755448, shape=(), dtype=float32)\n",
            "epoch: 5 step: 471 train batch loss: tf.Tensor(0.30306238, shape=(), dtype=float32)\n",
            "epoch: 5 step: 472 train batch loss: tf.Tensor(0.30173966, shape=(), dtype=float32)\n",
            "epoch: 5 step: 473 train batch loss: tf.Tensor(0.2879725, shape=(), dtype=float32)\n",
            "epoch: 5 step: 474 train batch loss: tf.Tensor(0.30433953, shape=(), dtype=float32)\n",
            "epoch: 5 step: 475 train batch loss: tf.Tensor(0.27959284, shape=(), dtype=float32)\n",
            "epoch: 5 step: 476 train batch loss: tf.Tensor(0.306352, shape=(), dtype=float32)\n",
            "epoch: 5 step: 477 train batch loss: tf.Tensor(0.25642988, shape=(), dtype=float32)\n",
            "epoch: 5 step: 478 train batch loss: tf.Tensor(0.3001936, shape=(), dtype=float32)\n",
            "epoch: 5 step: 479 train batch loss: tf.Tensor(0.30443686, shape=(), dtype=float32)\n",
            "epoch: 5 step: 480 train batch loss: tf.Tensor(0.37102693, shape=(), dtype=float32)\n",
            "epoch: 5 step: 481 train batch loss: tf.Tensor(0.22797327, shape=(), dtype=float32)\n",
            "epoch: 5 step: 482 train batch loss: tf.Tensor(0.27201003, shape=(), dtype=float32)\n",
            "epoch: 5 step: 483 train batch loss: tf.Tensor(0.2971534, shape=(), dtype=float32)\n",
            "epoch: 5 step: 484 train batch loss: tf.Tensor(0.24446514, shape=(), dtype=float32)\n",
            "epoch: 5 step: 485 train batch loss: tf.Tensor(0.33652654, shape=(), dtype=float32)\n",
            "epoch: 5 step: 486 train batch loss: tf.Tensor(0.24681255, shape=(), dtype=float32)\n",
            "epoch: 5 step: 487 train batch loss: tf.Tensor(0.2450695, shape=(), dtype=float32)\n",
            "epoch: 5 step: 488 train batch loss: tf.Tensor(0.28980878, shape=(), dtype=float32)\n",
            "epoch: 5 step: 489 train batch loss: tf.Tensor(0.23949707, shape=(), dtype=float32)\n",
            "epoch: 5 step: 490 train batch loss: tf.Tensor(0.2672078, shape=(), dtype=float32)\n",
            "epoch: 5 step: 491 train batch loss: tf.Tensor(0.24099994, shape=(), dtype=float32)\n",
            "epoch: 5 step: 492 train batch loss: tf.Tensor(0.32696542, shape=(), dtype=float32)\n",
            "epoch: 5 step: 493 train batch loss: tf.Tensor(0.25852638, shape=(), dtype=float32)\n",
            "epoch: 5 step: 494 train batch loss: tf.Tensor(0.25646576, shape=(), dtype=float32)\n",
            "epoch: 5 step: 495 train batch loss: tf.Tensor(0.20009613, shape=(), dtype=float32)\n",
            "epoch: 5 step: 496 train batch loss: tf.Tensor(0.3468935, shape=(), dtype=float32)\n",
            "epoch: 5 step: 497 train batch loss: tf.Tensor(0.321629, shape=(), dtype=float32)\n",
            "epoch: 5 step: 498 train batch loss: tf.Tensor(0.26607624, shape=(), dtype=float32)\n",
            "epoch: 5 step: 499 train batch loss: tf.Tensor(0.28090268, shape=(), dtype=float32)\n",
            "epoch: 5 step: 500 train batch loss: tf.Tensor(0.32316524, shape=(), dtype=float32)\n",
            "epoch: 5 step: 501 train batch loss: tf.Tensor(0.24182037, shape=(), dtype=float32)\n",
            "epoch: 5 step: 502 train batch loss: tf.Tensor(0.31153315, shape=(), dtype=float32)\n",
            "epoch: 5 step: 503 train batch loss: tf.Tensor(0.28602064, shape=(), dtype=float32)\n",
            "epoch: 5 step: 504 train batch loss: tf.Tensor(0.22592945, shape=(), dtype=float32)\n",
            "epoch: 5 step: 505 train batch loss: tf.Tensor(0.26332587, shape=(), dtype=float32)\n",
            "epoch: 5 step: 506 train batch loss: tf.Tensor(0.2713819, shape=(), dtype=float32)\n",
            "epoch: 5 step: 507 train batch loss: tf.Tensor(0.20116487, shape=(), dtype=float32)\n",
            "epoch: 5 step: 508 train batch loss: tf.Tensor(0.32379872, shape=(), dtype=float32)\n",
            "epoch: 5 step: 509 train batch loss: tf.Tensor(0.28939795, shape=(), dtype=float32)\n",
            "epoch: 5 step: 510 train batch loss: tf.Tensor(0.26259118, shape=(), dtype=float32)\n",
            "epoch: 5 step: 511 train batch loss: tf.Tensor(0.29037338, shape=(), dtype=float32)\n",
            "epoch: 5 step: 512 train batch loss: tf.Tensor(0.294, shape=(), dtype=float32)\n",
            "epoch: 5 step: 513 train batch loss: tf.Tensor(0.36029512, shape=(), dtype=float32)\n",
            "epoch: 5 step: 514 train batch loss: tf.Tensor(0.33624956, shape=(), dtype=float32)\n",
            "epoch: 5 step: 515 train batch loss: tf.Tensor(0.27806515, shape=(), dtype=float32)\n",
            "epoch: 5 step: 516 train batch loss: tf.Tensor(0.22621167, shape=(), dtype=float32)\n",
            "epoch: 5 step: 517 train batch loss: tf.Tensor(0.2091144, shape=(), dtype=float32)\n",
            "epoch: 5 step: 518 train batch loss: tf.Tensor(0.30922574, shape=(), dtype=float32)\n",
            "epoch: 5 step: 519 train batch loss: tf.Tensor(0.24869883, shape=(), dtype=float32)\n",
            "epoch: 5 step: 520 train batch loss: tf.Tensor(0.30176577, shape=(), dtype=float32)\n",
            "epoch: 5 step: 521 train batch loss: tf.Tensor(0.28137285, shape=(), dtype=float32)\n",
            "epoch: 5 step: 522 train batch loss: tf.Tensor(0.31998822, shape=(), dtype=float32)\n",
            "epoch: 5 step: 523 train batch loss: tf.Tensor(0.27900034, shape=(), dtype=float32)\n",
            "epoch: 5 step: 524 train batch loss: tf.Tensor(0.32535812, shape=(), dtype=float32)\n",
            "epoch: 5 step: 525 train batch loss: tf.Tensor(0.27768517, shape=(), dtype=float32)\n",
            "epoch: 5 step: 526 train batch loss: tf.Tensor(0.29192835, shape=(), dtype=float32)\n",
            "epoch: 5 step: 527 train batch loss: tf.Tensor(0.2072428, shape=(), dtype=float32)\n",
            "epoch: 5 step: 528 train batch loss: tf.Tensor(0.2583782, shape=(), dtype=float32)\n",
            "epoch: 5 step: 529 train batch loss: tf.Tensor(0.2997053, shape=(), dtype=float32)\n",
            "epoch: 5 step: 530 train batch loss: tf.Tensor(0.27747416, shape=(), dtype=float32)\n",
            "epoch: 5 step: 531 train batch loss: tf.Tensor(0.33682424, shape=(), dtype=float32)\n",
            "epoch: 5 step: 532 train batch loss: tf.Tensor(0.27308953, shape=(), dtype=float32)\n",
            "epoch: 5 step: 533 train batch loss: tf.Tensor(0.32022828, shape=(), dtype=float32)\n",
            "epoch: 5 step: 534 train batch loss: tf.Tensor(0.25643465, shape=(), dtype=float32)\n",
            "epoch: 5 step: 535 train batch loss: tf.Tensor(0.2183686, shape=(), dtype=float32)\n",
            "epoch: 5 step: 536 train batch loss: tf.Tensor(0.2283731, shape=(), dtype=float32)\n",
            "epoch: 5 step: 537 train batch loss: tf.Tensor(0.2832028, shape=(), dtype=float32)\n",
            "epoch: 5 step: 538 train batch loss: tf.Tensor(0.2518163, shape=(), dtype=float32)\n",
            "epoch: 5 step: 539 train batch loss: tf.Tensor(0.29243213, shape=(), dtype=float32)\n",
            "epoch: 5 step: 540 train batch loss: tf.Tensor(0.30578747, shape=(), dtype=float32)\n",
            "epoch: 5 step: 541 train batch loss: tf.Tensor(0.26098928, shape=(), dtype=float32)\n",
            "epoch: 5 step: 542 train batch loss: tf.Tensor(0.2958744, shape=(), dtype=float32)\n",
            "epoch: 5 step: 543 train batch loss: tf.Tensor(0.36709946, shape=(), dtype=float32)\n",
            "epoch: 5 step: 544 train batch loss: tf.Tensor(0.2734502, shape=(), dtype=float32)\n",
            "epoch: 5 step: 545 train batch loss: tf.Tensor(0.287558, shape=(), dtype=float32)\n",
            "epoch: 5 step: 546 train batch loss: tf.Tensor(0.23600459, shape=(), dtype=float32)\n",
            "epoch: 5 step: 547 train batch loss: tf.Tensor(0.22568616, shape=(), dtype=float32)\n",
            "epoch: 5 step: 548 train batch loss: tf.Tensor(0.2996358, shape=(), dtype=float32)\n",
            "epoch: 5 step: 549 train batch loss: tf.Tensor(0.24787554, shape=(), dtype=float32)\n",
            "epoch: 5 step: 550 train batch loss: tf.Tensor(0.2632365, shape=(), dtype=float32)\n",
            "epoch: 5 step: 551 train batch loss: tf.Tensor(0.26533872, shape=(), dtype=float32)\n",
            "epoch: 5 step: 552 train batch loss: tf.Tensor(0.25962994, shape=(), dtype=float32)\n",
            "epoch: 5 step: 553 train batch loss: tf.Tensor(0.34364772, shape=(), dtype=float32)\n",
            "epoch: 5 step: 554 train batch loss: tf.Tensor(0.36257166, shape=(), dtype=float32)\n",
            "epoch: 5 step: 555 train batch loss: tf.Tensor(0.258527, shape=(), dtype=float32)\n",
            "epoch: 5 step: 556 train batch loss: tf.Tensor(0.30140752, shape=(), dtype=float32)\n",
            "epoch: 5 step: 557 train batch loss: tf.Tensor(0.27541527, shape=(), dtype=float32)\n",
            "epoch: 5 step: 558 train batch loss: tf.Tensor(0.2424725, shape=(), dtype=float32)\n",
            "epoch: 5 step: 559 train batch loss: tf.Tensor(0.30279392, shape=(), dtype=float32)\n",
            "epoch: 5 step: 560 train batch loss: tf.Tensor(0.21970494, shape=(), dtype=float32)\n",
            "epoch: 5 step: 561 train batch loss: tf.Tensor(0.2737215, shape=(), dtype=float32)\n",
            "epoch: 5 step: 562 train batch loss: tf.Tensor(0.2821203, shape=(), dtype=float32)\n",
            "epoch: 5 step: 563 train batch loss: tf.Tensor(0.27931, shape=(), dtype=float32)\n",
            "epoch: 5 step: 564 train batch loss: tf.Tensor(0.2824708, shape=(), dtype=float32)\n",
            "epoch: 5 step: 565 train batch loss: tf.Tensor(0.37374777, shape=(), dtype=float32)\n",
            "epoch: 5 step: 566 train batch loss: tf.Tensor(0.27946693, shape=(), dtype=float32)\n",
            "epoch: 5 step: 567 train batch loss: tf.Tensor(0.24722621, shape=(), dtype=float32)\n",
            "epoch: 5 step: 568 train batch loss: tf.Tensor(0.25715968, shape=(), dtype=float32)\n",
            "epoch: 5 step: 569 train batch loss: tf.Tensor(0.30522758, shape=(), dtype=float32)\n",
            "epoch: 5 step: 570 train batch loss: tf.Tensor(0.2908649, shape=(), dtype=float32)\n",
            "epoch: 5 step: 571 train batch loss: tf.Tensor(0.28117254, shape=(), dtype=float32)\n",
            "epoch: 5 step: 572 train batch loss: tf.Tensor(0.3260446, shape=(), dtype=float32)\n",
            "epoch: 5 step: 573 train batch loss: tf.Tensor(0.2408156, shape=(), dtype=float32)\n",
            "epoch: 5 step: 574 train batch loss: tf.Tensor(0.26966146, shape=(), dtype=float32)\n",
            "epoch: 5 step: 575 train batch loss: tf.Tensor(0.27004826, shape=(), dtype=float32)\n",
            "epoch: 5 step: 576 train batch loss: tf.Tensor(0.31835505, shape=(), dtype=float32)\n",
            "epoch: 5 step: 577 train batch loss: tf.Tensor(0.24351096, shape=(), dtype=float32)\n",
            "epoch: 5 step: 578 train batch loss: tf.Tensor(0.29498377, shape=(), dtype=float32)\n",
            "epoch: 5 step: 579 train batch loss: tf.Tensor(0.35242146, shape=(), dtype=float32)\n",
            "epoch: 5 step: 580 train batch loss: tf.Tensor(0.2624718, shape=(), dtype=float32)\n",
            "epoch: 5 step: 581 train batch loss: tf.Tensor(0.26029402, shape=(), dtype=float32)\n",
            "epoch: 5 step: 582 train batch loss: tf.Tensor(0.28867236, shape=(), dtype=float32)\n",
            "epoch: 5 step: 583 train batch loss: tf.Tensor(0.25852108, shape=(), dtype=float32)\n",
            "epoch: 5 step: 584 train batch loss: tf.Tensor(0.28426015, shape=(), dtype=float32)\n",
            "epoch: 5 step: 585 train batch loss: tf.Tensor(0.27856493, shape=(), dtype=float32)\n",
            "epoch: 5 step: 586 train batch loss: tf.Tensor(0.27042866, shape=(), dtype=float32)\n",
            "epoch: 5 step: 587 train batch loss: tf.Tensor(0.23607402, shape=(), dtype=float32)\n",
            "epoch: 5 step: 588 train batch loss: tf.Tensor(0.25109154, shape=(), dtype=float32)\n",
            "epoch: 5 step: 589 train batch loss: tf.Tensor(0.26429754, shape=(), dtype=float32)\n",
            "epoch: 5 step: 590 train batch loss: tf.Tensor(0.27829984, shape=(), dtype=float32)\n",
            "epoch: 5 step: 591 train batch loss: tf.Tensor(0.26288888, shape=(), dtype=float32)\n",
            "epoch: 5 step: 592 train batch loss: tf.Tensor(0.3605678, shape=(), dtype=float32)\n",
            "epoch: 5 step: 593 train batch loss: tf.Tensor(0.26553488, shape=(), dtype=float32)\n",
            "epoch: 5 step: 594 train batch loss: tf.Tensor(0.22654352, shape=(), dtype=float32)\n",
            "epoch: 5 step: 595 train batch loss: tf.Tensor(0.29906368, shape=(), dtype=float32)\n",
            "epoch: 5 step: 596 train batch loss: tf.Tensor(0.28320247, shape=(), dtype=float32)\n",
            "epoch: 5 step: 597 train batch loss: tf.Tensor(0.24263903, shape=(), dtype=float32)\n",
            "epoch: 5 step: 598 train batch loss: tf.Tensor(0.35214218, shape=(), dtype=float32)\n",
            "epoch: 5 step: 599 train batch loss: tf.Tensor(0.21072273, shape=(), dtype=float32)\n",
            "epoch: 5 step: 600 train batch loss: tf.Tensor(0.33490092, shape=(), dtype=float32)\n",
            "epoch: 5 step: 601 train batch loss: tf.Tensor(0.29463646, shape=(), dtype=float32)\n",
            "epoch: 5 step: 602 train batch loss: tf.Tensor(0.35172647, shape=(), dtype=float32)\n",
            "epoch: 5 step: 603 train batch loss: tf.Tensor(0.31443697, shape=(), dtype=float32)\n",
            "epoch: 5 step: 604 train batch loss: tf.Tensor(0.28972188, shape=(), dtype=float32)\n",
            "epoch: 5 step: 605 train batch loss: tf.Tensor(0.29732904, shape=(), dtype=float32)\n",
            "epoch: 5 step: 606 train batch loss: tf.Tensor(0.25425667, shape=(), dtype=float32)\n",
            "epoch: 5 step: 607 train batch loss: tf.Tensor(0.34346515, shape=(), dtype=float32)\n",
            "epoch: 5 step: 608 train batch loss: tf.Tensor(0.28607923, shape=(), dtype=float32)\n",
            "epoch: 5 step: 609 train batch loss: tf.Tensor(0.2959853, shape=(), dtype=float32)\n",
            "epoch: 5 step: 610 train batch loss: tf.Tensor(0.31773788, shape=(), dtype=float32)\n",
            "epoch: 5 step: 611 train batch loss: tf.Tensor(0.27607906, shape=(), dtype=float32)\n",
            "epoch: 5 step: 612 train batch loss: tf.Tensor(0.32369307, shape=(), dtype=float32)\n",
            "epoch: 5 step: 613 train batch loss: tf.Tensor(0.23262163, shape=(), dtype=float32)\n",
            "epoch: 5 step: 614 train batch loss: tf.Tensor(0.2609449, shape=(), dtype=float32)\n",
            "epoch: 5 step: 615 train batch loss: tf.Tensor(0.29973608, shape=(), dtype=float32)\n",
            "epoch: 5 step: 616 train batch loss: tf.Tensor(0.28681698, shape=(), dtype=float32)\n",
            "epoch: 5 step: 617 train batch loss: tf.Tensor(0.23608525, shape=(), dtype=float32)\n",
            "epoch: 5 step: 618 train batch loss: tf.Tensor(0.21216115, shape=(), dtype=float32)\n",
            "epoch: 5 step: 619 train batch loss: tf.Tensor(0.31396076, shape=(), dtype=float32)\n",
            "epoch: 5 step: 620 train batch loss: tf.Tensor(0.24597104, shape=(), dtype=float32)\n",
            "epoch: 5 step: 621 train batch loss: tf.Tensor(0.23191957, shape=(), dtype=float32)\n",
            "epoch: 5 step: 622 train batch loss: tf.Tensor(0.30200142, shape=(), dtype=float32)\n",
            "epoch: 5 step: 623 train batch loss: tf.Tensor(0.26265985, shape=(), dtype=float32)\n",
            "epoch: 5 step: 624 train batch loss: tf.Tensor(0.28077567, shape=(), dtype=float32)\n",
            "epoch: 5 step: 625 train batch loss: tf.Tensor(0.23149188, shape=(), dtype=float32)\n",
            "epoch: 5 step: 626 train batch loss: tf.Tensor(0.24038737, shape=(), dtype=float32)\n",
            "epoch: 5 step: 627 train batch loss: tf.Tensor(0.32835275, shape=(), dtype=float32)\n",
            "epoch: 5 step: 628 train batch loss: tf.Tensor(0.27766457, shape=(), dtype=float32)\n",
            "epoch: 5 step: 629 train batch loss: tf.Tensor(0.27527732, shape=(), dtype=float32)\n",
            "epoch: 5 step: 630 train batch loss: tf.Tensor(0.27200776, shape=(), dtype=float32)\n",
            "epoch: 5 step: 631 train batch loss: tf.Tensor(0.3222894, shape=(), dtype=float32)\n",
            "epoch: 5 step: 632 train batch loss: tf.Tensor(0.24900246, shape=(), dtype=float32)\n",
            "epoch: 5 step: 633 train batch loss: tf.Tensor(0.2440823, shape=(), dtype=float32)\n",
            "epoch: 5 step: 634 train batch loss: tf.Tensor(0.28025222, shape=(), dtype=float32)\n",
            "epoch: 5 step: 635 train batch loss: tf.Tensor(0.329053, shape=(), dtype=float32)\n",
            "epoch: 5 step: 636 train batch loss: tf.Tensor(0.31978667, shape=(), dtype=float32)\n",
            "epoch: 5 step: 637 train batch loss: tf.Tensor(0.23852399, shape=(), dtype=float32)\n",
            "epoch: 5 step: 638 train batch loss: tf.Tensor(0.34652635, shape=(), dtype=float32)\n",
            "epoch: 5 step: 639 train batch loss: tf.Tensor(0.27559868, shape=(), dtype=float32)\n",
            "epoch: 5 step: 640 train batch loss: tf.Tensor(0.30549392, shape=(), dtype=float32)\n",
            "epoch: 5 step: 641 train batch loss: tf.Tensor(0.39625046, shape=(), dtype=float32)\n",
            "epoch: 5 step: 642 train batch loss: tf.Tensor(0.32711753, shape=(), dtype=float32)\n",
            "epoch: 5 step: 643 train batch loss: tf.Tensor(0.28690258, shape=(), dtype=float32)\n",
            "epoch: 5 step: 644 train batch loss: tf.Tensor(0.27258745, shape=(), dtype=float32)\n",
            "epoch: 5 step: 645 train batch loss: tf.Tensor(0.29106975, shape=(), dtype=float32)\n",
            "epoch: 5 step: 646 train batch loss: tf.Tensor(0.26620972, shape=(), dtype=float32)\n",
            "epoch: 5 step: 647 train batch loss: tf.Tensor(0.21936485, shape=(), dtype=float32)\n",
            "epoch: 5 step: 648 train batch loss: tf.Tensor(0.2352179, shape=(), dtype=float32)\n",
            "epoch: 5 step: 649 train batch loss: tf.Tensor(0.2557811, shape=(), dtype=float32)\n",
            "epoch: 5 step: 650 train batch loss: tf.Tensor(0.30257705, shape=(), dtype=float32)\n",
            "epoch: 5 step: 651 train batch loss: tf.Tensor(0.36867267, shape=(), dtype=float32)\n",
            "epoch: 5 step: 652 train batch loss: tf.Tensor(0.3054785, shape=(), dtype=float32)\n",
            "epoch: 5 step: 653 train batch loss: tf.Tensor(0.25172377, shape=(), dtype=float32)\n",
            "epoch: 5 step: 654 train batch loss: tf.Tensor(0.19458313, shape=(), dtype=float32)\n",
            "epoch: 5 step: 655 train batch loss: tf.Tensor(0.35276034, shape=(), dtype=float32)\n",
            "epoch: 5 step: 656 train batch loss: tf.Tensor(0.33407408, shape=(), dtype=float32)\n",
            "epoch: 5 step: 657 train batch loss: tf.Tensor(0.21446295, shape=(), dtype=float32)\n",
            "epoch: 5 step: 658 train batch loss: tf.Tensor(0.2779071, shape=(), dtype=float32)\n",
            "epoch: 5 step: 659 train batch loss: tf.Tensor(0.26516527, shape=(), dtype=float32)\n",
            "epoch: 5 step: 660 train batch loss: tf.Tensor(0.29397413, shape=(), dtype=float32)\n",
            "epoch: 5 step: 661 train batch loss: tf.Tensor(0.26161033, shape=(), dtype=float32)\n",
            "epoch: 5 step: 662 train batch loss: tf.Tensor(0.2836716, shape=(), dtype=float32)\n",
            "epoch: 5 step: 663 train batch loss: tf.Tensor(0.28291783, shape=(), dtype=float32)\n",
            "epoch: 5 step: 664 train batch loss: tf.Tensor(0.27835187, shape=(), dtype=float32)\n",
            "epoch: 5 step: 665 train batch loss: tf.Tensor(0.31049892, shape=(), dtype=float32)\n",
            "epoch: 5 step: 666 train batch loss: tf.Tensor(0.24614394, shape=(), dtype=float32)\n",
            "epoch: 5 step: 667 train batch loss: tf.Tensor(0.28461444, shape=(), dtype=float32)\n",
            "epoch: 5 step: 668 train batch loss: tf.Tensor(0.29599088, shape=(), dtype=float32)\n",
            "epoch: 5 step: 669 train batch loss: tf.Tensor(0.2389863, shape=(), dtype=float32)\n",
            "epoch: 5 step: 670 train batch loss: tf.Tensor(0.29801315, shape=(), dtype=float32)\n",
            "epoch: 5 step: 671 train batch loss: tf.Tensor(0.24697602, shape=(), dtype=float32)\n",
            "epoch: 5 step: 672 train batch loss: tf.Tensor(0.37710378, shape=(), dtype=float32)\n",
            "epoch: 5 step: 673 train batch loss: tf.Tensor(0.20964494, shape=(), dtype=float32)\n",
            "epoch: 5 step: 674 train batch loss: tf.Tensor(0.26325548, shape=(), dtype=float32)\n",
            "epoch: 5 step: 675 train batch loss: tf.Tensor(0.30868736, shape=(), dtype=float32)\n",
            "epoch: 5 step: 676 train batch loss: tf.Tensor(0.19625677, shape=(), dtype=float32)\n",
            "epoch: 5 step: 677 train batch loss: tf.Tensor(0.35136864, shape=(), dtype=float32)\n",
            "epoch: 5 step: 678 train batch loss: tf.Tensor(0.24958643, shape=(), dtype=float32)\n",
            "epoch: 5 step: 679 train batch loss: tf.Tensor(0.24272053, shape=(), dtype=float32)\n",
            "epoch: 5 step: 680 train batch loss: tf.Tensor(0.2829817, shape=(), dtype=float32)\n",
            "epoch: 5 step: 681 train batch loss: tf.Tensor(0.32886976, shape=(), dtype=float32)\n",
            "epoch: 5 step: 682 train batch loss: tf.Tensor(0.27312863, shape=(), dtype=float32)\n",
            "epoch: 5 step: 683 train batch loss: tf.Tensor(0.2675929, shape=(), dtype=float32)\n",
            "epoch: 5 step: 684 train batch loss: tf.Tensor(0.28694555, shape=(), dtype=float32)\n",
            "epoch: 5 step: 685 train batch loss: tf.Tensor(0.3276785, shape=(), dtype=float32)\n",
            "epoch: 5 step: 686 train batch loss: tf.Tensor(0.27517894, shape=(), dtype=float32)\n",
            "epoch: 5 step: 687 train batch loss: tf.Tensor(0.26995167, shape=(), dtype=float32)\n",
            "epoch: 5 step: 688 train batch loss: tf.Tensor(0.26633435, shape=(), dtype=float32)\n",
            "epoch: 5 step: 689 train batch loss: tf.Tensor(0.3145875, shape=(), dtype=float32)\n",
            "epoch: 5 step: 690 train batch loss: tf.Tensor(0.2559181, shape=(), dtype=float32)\n",
            "epoch: 5 step: 691 train batch loss: tf.Tensor(0.20592597, shape=(), dtype=float32)\n",
            "epoch: 5 step: 692 train batch loss: tf.Tensor(0.24658234, shape=(), dtype=float32)\n",
            "epoch: 5 step: 693 train batch loss: tf.Tensor(0.18495847, shape=(), dtype=float32)\n",
            "epoch: 5 step: 694 train batch loss: tf.Tensor(0.21146247, shape=(), dtype=float32)\n",
            "epoch: 5 step: 695 train batch loss: tf.Tensor(0.27510768, shape=(), dtype=float32)\n",
            "epoch: 5 step: 696 train batch loss: tf.Tensor(0.2504909, shape=(), dtype=float32)\n",
            "epoch: 5 step: 697 train batch loss: tf.Tensor(0.2946782, shape=(), dtype=float32)\n",
            "epoch: 5 step: 698 train batch loss: tf.Tensor(0.27174842, shape=(), dtype=float32)\n",
            "epoch: 5 step: 699 train batch loss: tf.Tensor(0.28559142, shape=(), dtype=float32)\n",
            "epoch: 5 step: 700 train batch loss: tf.Tensor(0.24133189, shape=(), dtype=float32)\n",
            "epoch: 5 step: 701 train batch loss: tf.Tensor(0.29045624, shape=(), dtype=float32)\n",
            "epoch: 5 step: 702 train batch loss: tf.Tensor(0.2867706, shape=(), dtype=float32)\n",
            "epoch: 5 step: 703 train batch loss: tf.Tensor(0.26215842, shape=(), dtype=float32)\n",
            "epoch: 5 step: 704 train batch loss: tf.Tensor(0.30508775, shape=(), dtype=float32)\n",
            "epoch: 5 step: 705 train batch loss: tf.Tensor(0.2852952, shape=(), dtype=float32)\n",
            "epoch: 5 step: 706 train batch loss: tf.Tensor(0.2769326, shape=(), dtype=float32)\n",
            "epoch: 5 step: 707 train batch loss: tf.Tensor(0.19879594, shape=(), dtype=float32)\n",
            "epoch: 5 step: 708 train batch loss: tf.Tensor(0.21468386, shape=(), dtype=float32)\n",
            "epoch: 5 step: 709 train batch loss: tf.Tensor(0.27044195, shape=(), dtype=float32)\n",
            "epoch: 5 step: 710 train batch loss: tf.Tensor(0.28528038, shape=(), dtype=float32)\n",
            "epoch: 5 step: 711 train batch loss: tf.Tensor(0.23652413, shape=(), dtype=float32)\n",
            "epoch: 5 step: 712 train batch loss: tf.Tensor(0.28401297, shape=(), dtype=float32)\n",
            "epoch: 5 step: 713 train batch loss: tf.Tensor(0.2990021, shape=(), dtype=float32)\n",
            "epoch: 5 step: 714 train batch loss: tf.Tensor(0.31673482, shape=(), dtype=float32)\n",
            "epoch: 5 step: 715 train batch loss: tf.Tensor(0.28588015, shape=(), dtype=float32)\n",
            "epoch: 5 step: 716 train batch loss: tf.Tensor(0.2621412, shape=(), dtype=float32)\n",
            "epoch: 5 step: 717 train batch loss: tf.Tensor(0.2636605, shape=(), dtype=float32)\n",
            "epoch: 5 step: 718 train batch loss: tf.Tensor(0.30263835, shape=(), dtype=float32)\n",
            "epoch: 5 step: 719 train batch loss: tf.Tensor(0.39799637, shape=(), dtype=float32)\n",
            "epoch: 5 step: 720 train batch loss: tf.Tensor(0.2783106, shape=(), dtype=float32)\n",
            "epoch: 5 step: 721 train batch loss: tf.Tensor(0.21961105, shape=(), dtype=float32)\n",
            "epoch: 5 step: 722 train batch loss: tf.Tensor(0.3165217, shape=(), dtype=float32)\n",
            "epoch: 5 step: 723 train batch loss: tf.Tensor(0.29315144, shape=(), dtype=float32)\n",
            "epoch: 5 step: 724 train batch loss: tf.Tensor(0.26033646, shape=(), dtype=float32)\n",
            "epoch: 5 step: 725 train batch loss: tf.Tensor(0.23669393, shape=(), dtype=float32)\n",
            "epoch: 5 step: 726 train batch loss: tf.Tensor(0.22646265, shape=(), dtype=float32)\n",
            "epoch: 5 step: 727 train batch loss: tf.Tensor(0.29332146, shape=(), dtype=float32)\n",
            "epoch: 5 step: 728 train batch loss: tf.Tensor(0.30654734, shape=(), dtype=float32)\n",
            "epoch: 5 step: 729 train batch loss: tf.Tensor(0.27663973, shape=(), dtype=float32)\n",
            "epoch: 5 step: 730 train batch loss: tf.Tensor(0.2829669, shape=(), dtype=float32)\n",
            "epoch: 5 step: 731 train batch loss: tf.Tensor(0.22194114, shape=(), dtype=float32)\n",
            "epoch: 5 step: 732 train batch loss: tf.Tensor(0.27320403, shape=(), dtype=float32)\n",
            "epoch: 5 step: 733 train batch loss: tf.Tensor(0.34130162, shape=(), dtype=float32)\n",
            "epoch: 5 step: 734 train batch loss: tf.Tensor(0.31599802, shape=(), dtype=float32)\n",
            "epoch: 5 step: 735 train batch loss: tf.Tensor(0.27116284, shape=(), dtype=float32)\n",
            "epoch: 5 step: 736 train batch loss: tf.Tensor(0.3014251, shape=(), dtype=float32)\n",
            "epoch: 5 step: 737 train batch loss: tf.Tensor(0.27428272, shape=(), dtype=float32)\n",
            "epoch: 5 step: 738 train batch loss: tf.Tensor(0.31868926, shape=(), dtype=float32)\n",
            "epoch: 5 step: 739 train batch loss: tf.Tensor(0.24119708, shape=(), dtype=float32)\n",
            "epoch: 5 step: 740 train batch loss: tf.Tensor(0.32023793, shape=(), dtype=float32)\n",
            "epoch: 5 step: 741 train batch loss: tf.Tensor(0.3202031, shape=(), dtype=float32)\n",
            "epoch: 5 step: 742 train batch loss: tf.Tensor(0.32191822, shape=(), dtype=float32)\n",
            "epoch: 5 step: 743 train batch loss: tf.Tensor(0.23224843, shape=(), dtype=float32)\n",
            "epoch: 5 step: 744 train batch loss: tf.Tensor(0.23396143, shape=(), dtype=float32)\n",
            "epoch: 5 step: 745 train batch loss: tf.Tensor(0.22690551, shape=(), dtype=float32)\n",
            "epoch: 5 step: 746 train batch loss: tf.Tensor(0.32325512, shape=(), dtype=float32)\n",
            "epoch: 5 step: 747 train batch loss: tf.Tensor(0.2743135, shape=(), dtype=float32)\n",
            "epoch: 5 step: 748 train batch loss: tf.Tensor(0.26459825, shape=(), dtype=float32)\n",
            "epoch: 5 step: 749 train batch loss: tf.Tensor(0.31192833, shape=(), dtype=float32)\n",
            "epoch: 5 step: 750 train batch loss: tf.Tensor(0.19917367, shape=(), dtype=float32)\n",
            "epoch: 5 step: 751 train batch loss: tf.Tensor(0.24733083, shape=(), dtype=float32)\n",
            "epoch: 5 step: 752 train batch loss: tf.Tensor(0.32001492, shape=(), dtype=float32)\n",
            "epoch: 5 step: 753 train batch loss: tf.Tensor(0.2571635, shape=(), dtype=float32)\n",
            "epoch: 5 step: 754 train batch loss: tf.Tensor(0.29766518, shape=(), dtype=float32)\n",
            "epoch: 5 step: 755 train batch loss: tf.Tensor(0.23752621, shape=(), dtype=float32)\n",
            "epoch: 5 step: 756 train batch loss: tf.Tensor(0.19832034, shape=(), dtype=float32)\n",
            "epoch: 5 step: 757 train batch loss: tf.Tensor(0.3159968, shape=(), dtype=float32)\n",
            "epoch: 5 step: 758 train batch loss: tf.Tensor(0.29391658, shape=(), dtype=float32)\n",
            "epoch: 5 step: 759 train batch loss: tf.Tensor(0.3019085, shape=(), dtype=float32)\n",
            "epoch: 5 step: 760 train batch loss: tf.Tensor(0.24331243, shape=(), dtype=float32)\n",
            "epoch: 5 step: 761 train batch loss: tf.Tensor(0.24870312, shape=(), dtype=float32)\n",
            "epoch: 5 step: 762 train batch loss: tf.Tensor(0.28394267, shape=(), dtype=float32)\n",
            "epoch: 5 step: 763 train batch loss: tf.Tensor(0.30622354, shape=(), dtype=float32)\n",
            "epoch: 5 step: 764 train batch loss: tf.Tensor(0.33448356, shape=(), dtype=float32)\n",
            "epoch: 5 step: 765 train batch loss: tf.Tensor(0.27625707, shape=(), dtype=float32)\n",
            "epoch: 5 step: 766 train batch loss: tf.Tensor(0.2769542, shape=(), dtype=float32)\n",
            "epoch: 5 step: 767 train batch loss: tf.Tensor(0.2652282, shape=(), dtype=float32)\n",
            "epoch: 5 step: 768 train batch loss: tf.Tensor(0.24724135, shape=(), dtype=float32)\n",
            "epoch: 5 step: 769 train batch loss: tf.Tensor(0.31508487, shape=(), dtype=float32)\n",
            "epoch: 5 step: 770 train batch loss: tf.Tensor(0.18749723, shape=(), dtype=float32)\n",
            "epoch: 5 step: 771 train batch loss: tf.Tensor(0.27147156, shape=(), dtype=float32)\n",
            "epoch: 5 step: 772 train batch loss: tf.Tensor(0.27557606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 773 train batch loss: tf.Tensor(0.22779456, shape=(), dtype=float32)\n",
            "epoch: 5 step: 774 train batch loss: tf.Tensor(0.2492384, shape=(), dtype=float32)\n",
            "epoch: 5 step: 775 train batch loss: tf.Tensor(0.24818137, shape=(), dtype=float32)\n",
            "epoch: 5 step: 776 train batch loss: tf.Tensor(0.32257146, shape=(), dtype=float32)\n",
            "epoch: 5 step: 777 train batch loss: tf.Tensor(0.2892824, shape=(), dtype=float32)\n",
            "epoch: 5 step: 778 train batch loss: tf.Tensor(0.3057558, shape=(), dtype=float32)\n",
            "epoch: 5 step: 779 train batch loss: tf.Tensor(0.20556656, shape=(), dtype=float32)\n",
            "epoch: 5 step: 780 train batch loss: tf.Tensor(0.35414463, shape=(), dtype=float32)\n",
            "epoch: 5 step: 781 train batch loss: tf.Tensor(0.2540303, shape=(), dtype=float32)\n",
            "epoch: 5 step: 782 train batch loss: tf.Tensor(0.29632968, shape=(), dtype=float32)\n",
            "epoch: 5 step: 783 train batch loss: tf.Tensor(0.25075206, shape=(), dtype=float32)\n",
            "epoch: 5 step: 784 train batch loss: tf.Tensor(0.31919807, shape=(), dtype=float32)\n",
            "epoch: 5 step: 785 train batch loss: tf.Tensor(0.2672933, shape=(), dtype=float32)\n",
            "epoch: 5 step: 786 train batch loss: tf.Tensor(0.22539194, shape=(), dtype=float32)\n",
            "epoch: 5 step: 787 train batch loss: tf.Tensor(0.29142496, shape=(), dtype=float32)\n",
            "epoch: 5 step: 788 train batch loss: tf.Tensor(0.24890022, shape=(), dtype=float32)\n",
            "epoch: 5 step: 789 train batch loss: tf.Tensor(0.24827826, shape=(), dtype=float32)\n",
            "epoch: 5 step: 790 train batch loss: tf.Tensor(0.26494578, shape=(), dtype=float32)\n",
            "epoch: 5 step: 791 train batch loss: tf.Tensor(0.26786885, shape=(), dtype=float32)\n",
            "epoch: 5 step: 792 train batch loss: tf.Tensor(0.27497602, shape=(), dtype=float32)\n",
            "epoch: 5 step: 793 train batch loss: tf.Tensor(0.2676753, shape=(), dtype=float32)\n",
            "epoch: 5 step: 794 train batch loss: tf.Tensor(0.23787048, shape=(), dtype=float32)\n",
            "epoch: 5 step: 795 train batch loss: tf.Tensor(0.2095908, shape=(), dtype=float32)\n",
            "epoch: 5 step: 796 train batch loss: tf.Tensor(0.26754197, shape=(), dtype=float32)\n",
            "epoch: 5 step: 797 train batch loss: tf.Tensor(0.29085675, shape=(), dtype=float32)\n",
            "epoch: 5 step: 798 train batch loss: tf.Tensor(0.29213947, shape=(), dtype=float32)\n",
            "epoch: 5 step: 799 train batch loss: tf.Tensor(0.23366222, shape=(), dtype=float32)\n",
            "epoch: 5 step: 800 train batch loss: tf.Tensor(0.28721502, shape=(), dtype=float32)\n",
            "epoch: 5 step: 801 train batch loss: tf.Tensor(0.3161137, shape=(), dtype=float32)\n",
            "epoch: 5 step: 802 train batch loss: tf.Tensor(0.25887212, shape=(), dtype=float32)\n",
            "epoch: 5 step: 803 train batch loss: tf.Tensor(0.38526902, shape=(), dtype=float32)\n",
            "epoch: 5 step: 804 train batch loss: tf.Tensor(0.29882574, shape=(), dtype=float32)\n",
            "epoch: 5 step: 805 train batch loss: tf.Tensor(0.23558907, shape=(), dtype=float32)\n",
            "epoch: 5 step: 806 train batch loss: tf.Tensor(0.24327001, shape=(), dtype=float32)\n",
            "epoch: 5 step: 807 train batch loss: tf.Tensor(0.23954096, shape=(), dtype=float32)\n",
            "epoch: 5 step: 808 train batch loss: tf.Tensor(0.24509856, shape=(), dtype=float32)\n",
            "epoch: 5 step: 809 train batch loss: tf.Tensor(0.26161277, shape=(), dtype=float32)\n",
            "epoch: 5 step: 810 train batch loss: tf.Tensor(0.2907926, shape=(), dtype=float32)\n",
            "epoch: 5 step: 811 train batch loss: tf.Tensor(0.25696144, shape=(), dtype=float32)\n",
            "epoch: 5 step: 812 train batch loss: tf.Tensor(0.25846785, shape=(), dtype=float32)\n",
            "epoch: 5 step: 813 train batch loss: tf.Tensor(0.2567426, shape=(), dtype=float32)\n",
            "epoch: 5 step: 814 train batch loss: tf.Tensor(0.26189277, shape=(), dtype=float32)\n",
            "epoch: 5 step: 815 train batch loss: tf.Tensor(0.30948663, shape=(), dtype=float32)\n",
            "epoch: 5 step: 816 train batch loss: tf.Tensor(0.26923952, shape=(), dtype=float32)\n",
            "epoch: 5 step: 817 train batch loss: tf.Tensor(0.24888225, shape=(), dtype=float32)\n",
            "epoch: 5 step: 818 train batch loss: tf.Tensor(0.2860132, shape=(), dtype=float32)\n",
            "epoch: 5 step: 819 train batch loss: tf.Tensor(0.2501798, shape=(), dtype=float32)\n",
            "epoch: 5 step: 820 train batch loss: tf.Tensor(0.24763095, shape=(), dtype=float32)\n",
            "epoch: 5 step: 821 train batch loss: tf.Tensor(0.22560255, shape=(), dtype=float32)\n",
            "epoch: 5 step: 822 train batch loss: tf.Tensor(0.2474787, shape=(), dtype=float32)\n",
            "epoch: 5 step: 823 train batch loss: tf.Tensor(0.29420358, shape=(), dtype=float32)\n",
            "epoch: 5 step: 824 train batch loss: tf.Tensor(0.23700988, shape=(), dtype=float32)\n",
            "epoch: 5 step: 825 train batch loss: tf.Tensor(0.27411243, shape=(), dtype=float32)\n",
            "epoch: 5 step: 826 train batch loss: tf.Tensor(0.26506186, shape=(), dtype=float32)\n",
            "epoch: 5 step: 827 train batch loss: tf.Tensor(0.276396, shape=(), dtype=float32)\n",
            "epoch: 5 step: 828 train batch loss: tf.Tensor(0.2353617, shape=(), dtype=float32)\n",
            "epoch: 5 step: 829 train batch loss: tf.Tensor(0.3001696, shape=(), dtype=float32)\n",
            "epoch: 5 step: 830 train batch loss: tf.Tensor(0.24358392, shape=(), dtype=float32)\n",
            "epoch: 5 step: 831 train batch loss: tf.Tensor(0.22984567, shape=(), dtype=float32)\n",
            "epoch: 5 step: 832 train batch loss: tf.Tensor(0.26416886, shape=(), dtype=float32)\n",
            "epoch: 5 step: 833 train batch loss: tf.Tensor(0.32786512, shape=(), dtype=float32)\n",
            "epoch: 5 step: 834 train batch loss: tf.Tensor(0.21408062, shape=(), dtype=float32)\n",
            "epoch: 5 step: 835 train batch loss: tf.Tensor(0.23755851, shape=(), dtype=float32)\n",
            "epoch: 5 step: 836 train batch loss: tf.Tensor(0.2778403, shape=(), dtype=float32)\n",
            "epoch: 5 step: 837 train batch loss: tf.Tensor(0.30517948, shape=(), dtype=float32)\n",
            "epoch: 5 step: 838 train batch loss: tf.Tensor(0.2770046, shape=(), dtype=float32)\n",
            "epoch: 5 step: 839 train batch loss: tf.Tensor(0.23575234, shape=(), dtype=float32)\n",
            "epoch: 5 step: 840 train batch loss: tf.Tensor(0.2104815, shape=(), dtype=float32)\n",
            "epoch: 5 step: 841 train batch loss: tf.Tensor(0.25721422, shape=(), dtype=float32)\n",
            "epoch: 5 step: 842 train batch loss: tf.Tensor(0.3061834, shape=(), dtype=float32)\n",
            "epoch: 5 step: 843 train batch loss: tf.Tensor(0.31134057, shape=(), dtype=float32)\n",
            "epoch: 5 step: 844 train batch loss: tf.Tensor(0.25214857, shape=(), dtype=float32)\n",
            "epoch: 5 step: 845 train batch loss: tf.Tensor(0.27024528, shape=(), dtype=float32)\n",
            "epoch: 5 step: 846 train batch loss: tf.Tensor(0.21827805, shape=(), dtype=float32)\n",
            "epoch: 5 step: 847 train batch loss: tf.Tensor(0.3284356, shape=(), dtype=float32)\n",
            "epoch: 5 step: 848 train batch loss: tf.Tensor(0.32396907, shape=(), dtype=float32)\n",
            "epoch: 5 step: 849 train batch loss: tf.Tensor(0.24342713, shape=(), dtype=float32)\n",
            "epoch: 5 step: 850 train batch loss: tf.Tensor(0.25533468, shape=(), dtype=float32)\n",
            "epoch: 5 step: 851 train batch loss: tf.Tensor(0.28712398, shape=(), dtype=float32)\n",
            "epoch: 5 step: 852 train batch loss: tf.Tensor(0.32908946, shape=(), dtype=float32)\n",
            "epoch: 5 step: 853 train batch loss: tf.Tensor(0.21730974, shape=(), dtype=float32)\n",
            "epoch: 5 step: 854 train batch loss: tf.Tensor(0.27900484, shape=(), dtype=float32)\n",
            "epoch: 5 step: 855 train batch loss: tf.Tensor(0.22660375, shape=(), dtype=float32)\n",
            "epoch: 5 step: 856 train batch loss: tf.Tensor(0.31180924, shape=(), dtype=float32)\n",
            "epoch: 5 step: 857 train batch loss: tf.Tensor(0.20571168, shape=(), dtype=float32)\n",
            "epoch: 5 step: 858 train batch loss: tf.Tensor(0.30763295, shape=(), dtype=float32)\n",
            "epoch: 5 step: 859 train batch loss: tf.Tensor(0.21573332, shape=(), dtype=float32)\n",
            "epoch: 5 step: 860 train batch loss: tf.Tensor(0.21175607, shape=(), dtype=float32)\n",
            "epoch: 5 step: 861 train batch loss: tf.Tensor(0.28470302, shape=(), dtype=float32)\n",
            "epoch: 5 step: 862 train batch loss: tf.Tensor(0.34124163, shape=(), dtype=float32)\n",
            "epoch: 5 step: 863 train batch loss: tf.Tensor(0.26829293, shape=(), dtype=float32)\n",
            "epoch: 5 step: 864 train batch loss: tf.Tensor(0.20888826, shape=(), dtype=float32)\n",
            "epoch: 5 step: 865 train batch loss: tf.Tensor(0.22819944, shape=(), dtype=float32)\n",
            "epoch: 5 step: 866 train batch loss: tf.Tensor(0.25994346, shape=(), dtype=float32)\n",
            "epoch: 5 step: 867 train batch loss: tf.Tensor(0.2606755, shape=(), dtype=float32)\n",
            "epoch: 5 step: 868 train batch loss: tf.Tensor(0.2389421, shape=(), dtype=float32)\n",
            "epoch: 5 step: 869 train batch loss: tf.Tensor(0.23469253, shape=(), dtype=float32)\n",
            "epoch: 5 step: 870 train batch loss: tf.Tensor(0.26696736, shape=(), dtype=float32)\n",
            "epoch: 5 step: 871 train batch loss: tf.Tensor(0.24888515, shape=(), dtype=float32)\n",
            "epoch: 5 step: 872 train batch loss: tf.Tensor(0.19009455, shape=(), dtype=float32)\n",
            "epoch: 5 step: 873 train batch loss: tf.Tensor(0.28068587, shape=(), dtype=float32)\n",
            "epoch: 5 step: 874 train batch loss: tf.Tensor(0.32303828, shape=(), dtype=float32)\n",
            "epoch: 5 step: 875 train batch loss: tf.Tensor(0.24521565, shape=(), dtype=float32)\n",
            "epoch: 5 step: 876 train batch loss: tf.Tensor(0.31521577, shape=(), dtype=float32)\n",
            "epoch: 5 step: 877 train batch loss: tf.Tensor(0.24583237, shape=(), dtype=float32)\n",
            "epoch: 5 step: 878 train batch loss: tf.Tensor(0.3091376, shape=(), dtype=float32)\n",
            "epoch: 5 step: 879 train batch loss: tf.Tensor(0.24635044, shape=(), dtype=float32)\n",
            "epoch: 5 step: 880 train batch loss: tf.Tensor(0.27742493, shape=(), dtype=float32)\n",
            "epoch: 5 step: 881 train batch loss: tf.Tensor(0.24071606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 882 train batch loss: tf.Tensor(0.21062365, shape=(), dtype=float32)\n",
            "epoch: 5 step: 883 train batch loss: tf.Tensor(0.27078635, shape=(), dtype=float32)\n",
            "epoch: 5 step: 884 train batch loss: tf.Tensor(0.2566273, shape=(), dtype=float32)\n",
            "epoch: 5 step: 885 train batch loss: tf.Tensor(0.3002116, shape=(), dtype=float32)\n",
            "epoch: 5 step: 886 train batch loss: tf.Tensor(0.25614268, shape=(), dtype=float32)\n",
            "epoch: 5 step: 887 train batch loss: tf.Tensor(0.2739744, shape=(), dtype=float32)\n",
            "epoch: 5 step: 888 train batch loss: tf.Tensor(0.27172497, shape=(), dtype=float32)\n",
            "epoch: 5 step: 889 train batch loss: tf.Tensor(0.18995848, shape=(), dtype=float32)\n",
            "epoch: 5 step: 890 train batch loss: tf.Tensor(0.3384412, shape=(), dtype=float32)\n",
            "epoch: 5 step: 891 train batch loss: tf.Tensor(0.22450054, shape=(), dtype=float32)\n",
            "epoch: 5 step: 892 train batch loss: tf.Tensor(0.22684681, shape=(), dtype=float32)\n",
            "epoch: 5 step: 893 train batch loss: tf.Tensor(0.316504, shape=(), dtype=float32)\n",
            "epoch: 5 step: 894 train batch loss: tf.Tensor(0.2789337, shape=(), dtype=float32)\n",
            "epoch: 5 step: 895 train batch loss: tf.Tensor(0.29170945, shape=(), dtype=float32)\n",
            "epoch: 5 step: 896 train batch loss: tf.Tensor(0.21981224, shape=(), dtype=float32)\n",
            "epoch: 5 step: 897 train batch loss: tf.Tensor(0.30194274, shape=(), dtype=float32)\n",
            "epoch: 5 step: 898 train batch loss: tf.Tensor(0.2917373, shape=(), dtype=float32)\n",
            "epoch: 5 step: 899 train batch loss: tf.Tensor(0.3158803, shape=(), dtype=float32)\n",
            "epoch: 5 step: 900 train batch loss: tf.Tensor(0.23959343, shape=(), dtype=float32)\n",
            "epoch: 5 step: 901 train batch loss: tf.Tensor(0.33026028, shape=(), dtype=float32)\n",
            "epoch: 5 step: 902 train batch loss: tf.Tensor(0.33313388, shape=(), dtype=float32)\n",
            "epoch: 5 step: 903 train batch loss: tf.Tensor(0.19249777, shape=(), dtype=float32)\n",
            "epoch: 5 step: 904 train batch loss: tf.Tensor(0.25659585, shape=(), dtype=float32)\n",
            "epoch: 5 step: 905 train batch loss: tf.Tensor(0.2613277, shape=(), dtype=float32)\n",
            "epoch: 5 step: 906 train batch loss: tf.Tensor(0.2430956, shape=(), dtype=float32)\n",
            "epoch: 5 step: 907 train batch loss: tf.Tensor(0.24275231, shape=(), dtype=float32)\n",
            "epoch: 5 step: 908 train batch loss: tf.Tensor(0.29831165, shape=(), dtype=float32)\n",
            "epoch: 5 step: 909 train batch loss: tf.Tensor(0.2775877, shape=(), dtype=float32)\n",
            "epoch: 5 step: 910 train batch loss: tf.Tensor(0.3438624, shape=(), dtype=float32)\n",
            "epoch: 5 step: 911 train batch loss: tf.Tensor(0.3687355, shape=(), dtype=float32)\n",
            "epoch: 5 step: 912 train batch loss: tf.Tensor(0.2910248, shape=(), dtype=float32)\n",
            "epoch: 5 step: 913 train batch loss: tf.Tensor(0.26961797, shape=(), dtype=float32)\n",
            "epoch: 5 step: 914 train batch loss: tf.Tensor(0.29153234, shape=(), dtype=float32)\n",
            "epoch: 5 step: 915 train batch loss: tf.Tensor(0.2797988, shape=(), dtype=float32)\n",
            "epoch: 5 step: 916 train batch loss: tf.Tensor(0.24463494, shape=(), dtype=float32)\n",
            "epoch: 5 step: 917 train batch loss: tf.Tensor(0.30723315, shape=(), dtype=float32)\n",
            "epoch: 5 step: 918 train batch loss: tf.Tensor(0.20670743, shape=(), dtype=float32)\n",
            "epoch: 5 step: 919 train batch loss: tf.Tensor(0.20845787, shape=(), dtype=float32)\n",
            "epoch: 5 step: 920 train batch loss: tf.Tensor(0.2635618, shape=(), dtype=float32)\n",
            "epoch: 5 step: 921 train batch loss: tf.Tensor(0.26800704, shape=(), dtype=float32)\n",
            "epoch: 5 step: 922 train batch loss: tf.Tensor(0.3024309, shape=(), dtype=float32)\n",
            "epoch: 5 step: 923 train batch loss: tf.Tensor(0.26425028, shape=(), dtype=float32)\n",
            "epoch: 5 step: 924 train batch loss: tf.Tensor(0.32137257, shape=(), dtype=float32)\n",
            "epoch: 5 step: 925 train batch loss: tf.Tensor(0.27429157, shape=(), dtype=float32)\n",
            "epoch: 5 step: 926 train batch loss: tf.Tensor(0.2923389, shape=(), dtype=float32)\n",
            "epoch: 5 step: 927 train batch loss: tf.Tensor(0.23983459, shape=(), dtype=float32)\n",
            "epoch: 5 step: 928 train batch loss: tf.Tensor(0.2594671, shape=(), dtype=float32)\n",
            "epoch: 5 step: 929 train batch loss: tf.Tensor(0.2515263, shape=(), dtype=float32)\n",
            "epoch: 5 step: 930 train batch loss: tf.Tensor(0.31640944, shape=(), dtype=float32)\n",
            "epoch: 5 step: 931 train batch loss: tf.Tensor(0.2058549, shape=(), dtype=float32)\n",
            "epoch: 5 step: 932 train batch loss: tf.Tensor(0.26609516, shape=(), dtype=float32)\n",
            "epoch: 5 step: 933 train batch loss: tf.Tensor(0.21621649, shape=(), dtype=float32)\n",
            "epoch: 5 step: 934 train batch loss: tf.Tensor(0.29982078, shape=(), dtype=float32)\n",
            "epoch: 5 step: 935 train batch loss: tf.Tensor(0.25265923, shape=(), dtype=float32)\n",
            "epoch: 5 step: 936 train batch loss: tf.Tensor(0.24735002, shape=(), dtype=float32)\n",
            "epoch: 5 step: 937 train batch loss: tf.Tensor(0.23191333, shape=(), dtype=float32)\n",
            "epoch: 5 step: 938 train batch loss: tf.Tensor(0.2802146, shape=(), dtype=float32)\n",
            "epoch: 5 step: 939 train batch loss: tf.Tensor(0.23299518, shape=(), dtype=float32)\n",
            "epoch: 5 step: 940 train batch loss: tf.Tensor(0.2611183, shape=(), dtype=float32)\n",
            "epoch: 5 step: 941 train batch loss: tf.Tensor(0.27294618, shape=(), dtype=float32)\n",
            "epoch: 5 step: 942 train batch loss: tf.Tensor(0.29515475, shape=(), dtype=float32)\n",
            "epoch: 5 step: 943 train batch loss: tf.Tensor(0.23935814, shape=(), dtype=float32)\n",
            "epoch: 5 step: 944 train batch loss: tf.Tensor(0.26208323, shape=(), dtype=float32)\n",
            "epoch: 5 step: 945 train batch loss: tf.Tensor(0.31667998, shape=(), dtype=float32)\n",
            "epoch: 5 step: 946 train batch loss: tf.Tensor(0.24847856, shape=(), dtype=float32)\n",
            "epoch: 5 step: 947 train batch loss: tf.Tensor(0.3124448, shape=(), dtype=float32)\n",
            "epoch: 5 step: 948 train batch loss: tf.Tensor(0.24971266, shape=(), dtype=float32)\n",
            "epoch: 5 step: 949 train batch loss: tf.Tensor(0.26043904, shape=(), dtype=float32)\n",
            "epoch: 5 step: 950 train batch loss: tf.Tensor(0.34728304, shape=(), dtype=float32)\n",
            "epoch: 5 step: 951 train batch loss: tf.Tensor(0.27295938, shape=(), dtype=float32)\n",
            "epoch: 5 step: 952 train batch loss: tf.Tensor(0.24804564, shape=(), dtype=float32)\n",
            "epoch: 5 step: 953 train batch loss: tf.Tensor(0.27510828, shape=(), dtype=float32)\n",
            "epoch: 5 step: 954 train batch loss: tf.Tensor(0.19516596, shape=(), dtype=float32)\n",
            "epoch: 5 step: 955 train batch loss: tf.Tensor(0.32260492, shape=(), dtype=float32)\n",
            "epoch: 5 step: 956 train batch loss: tf.Tensor(0.24597606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 957 train batch loss: tf.Tensor(0.2490472, shape=(), dtype=float32)\n",
            "epoch: 5 step: 958 train batch loss: tf.Tensor(0.2723996, shape=(), dtype=float32)\n",
            "epoch: 5 step: 959 train batch loss: tf.Tensor(0.23526207, shape=(), dtype=float32)\n",
            "epoch: 5 step: 960 train batch loss: tf.Tensor(0.29409945, shape=(), dtype=float32)\n",
            "epoch: 5 step: 961 train batch loss: tf.Tensor(0.24494168, shape=(), dtype=float32)\n",
            "epoch: 5 step: 962 train batch loss: tf.Tensor(0.2857334, shape=(), dtype=float32)\n",
            "epoch: 5 step: 963 train batch loss: tf.Tensor(0.25728428, shape=(), dtype=float32)\n",
            "epoch: 5 step: 964 train batch loss: tf.Tensor(0.25839502, shape=(), dtype=float32)\n",
            "epoch: 5 step: 965 train batch loss: tf.Tensor(0.29278025, shape=(), dtype=float32)\n",
            "epoch: 5 step: 966 train batch loss: tf.Tensor(0.29773694, shape=(), dtype=float32)\n",
            "epoch: 5 step: 967 train batch loss: tf.Tensor(0.27597982, shape=(), dtype=float32)\n",
            "epoch: 5 step: 968 train batch loss: tf.Tensor(0.27393413, shape=(), dtype=float32)\n",
            "epoch: 5 step: 969 train batch loss: tf.Tensor(0.2322135, shape=(), dtype=float32)\n",
            "epoch: 5 step: 970 train batch loss: tf.Tensor(0.24392748, shape=(), dtype=float32)\n",
            "epoch: 5 step: 971 train batch loss: tf.Tensor(0.3406207, shape=(), dtype=float32)\n",
            "epoch: 5 step: 972 train batch loss: tf.Tensor(0.21482643, shape=(), dtype=float32)\n",
            "epoch: 5 step: 973 train batch loss: tf.Tensor(0.23393805, shape=(), dtype=float32)\n",
            "epoch: 5 step: 974 train batch loss: tf.Tensor(0.2666847, shape=(), dtype=float32)\n",
            "epoch: 5 step: 975 train batch loss: tf.Tensor(0.28153384, shape=(), dtype=float32)\n",
            "epoch: 5 step: 976 train batch loss: tf.Tensor(0.26439738, shape=(), dtype=float32)\n",
            "epoch: 5 step: 977 train batch loss: tf.Tensor(0.38913897, shape=(), dtype=float32)\n",
            "epoch: 5 step: 978 train batch loss: tf.Tensor(0.30213004, shape=(), dtype=float32)\n",
            "epoch: 5 step: 979 train batch loss: tf.Tensor(0.24627256, shape=(), dtype=float32)\n",
            "epoch: 5 step: 980 train batch loss: tf.Tensor(0.26463127, shape=(), dtype=float32)\n",
            "epoch: 5 step: 981 train batch loss: tf.Tensor(0.1700426, shape=(), dtype=float32)\n",
            "epoch: 5 step: 982 train batch loss: tf.Tensor(0.284195, shape=(), dtype=float32)\n",
            "epoch: 5 step: 983 train batch loss: tf.Tensor(0.2569981, shape=(), dtype=float32)\n",
            "epoch: 5 step: 984 train batch loss: tf.Tensor(0.27191848, shape=(), dtype=float32)\n",
            "epoch: 5 step: 985 train batch loss: tf.Tensor(0.2370034, shape=(), dtype=float32)\n",
            "epoch: 5 step: 986 train batch loss: tf.Tensor(0.3313853, shape=(), dtype=float32)\n",
            "epoch: 5 step: 987 train batch loss: tf.Tensor(0.24587917, shape=(), dtype=float32)\n",
            "epoch: 5 step: 988 train batch loss: tf.Tensor(0.26828045, shape=(), dtype=float32)\n",
            "epoch: 5 step: 989 train batch loss: tf.Tensor(0.2845862, shape=(), dtype=float32)\n",
            "epoch: 5 step: 990 train batch loss: tf.Tensor(0.24207385, shape=(), dtype=float32)\n",
            "epoch: 5 step: 991 train batch loss: tf.Tensor(0.28798813, shape=(), dtype=float32)\n",
            "epoch: 5 step: 992 train batch loss: tf.Tensor(0.32857746, shape=(), dtype=float32)\n",
            "epoch: 5 step: 993 train batch loss: tf.Tensor(0.23867473, shape=(), dtype=float32)\n",
            "epoch: 5 step: 994 train batch loss: tf.Tensor(0.27572837, shape=(), dtype=float32)\n",
            "epoch: 5 step: 995 train batch loss: tf.Tensor(0.25704584, shape=(), dtype=float32)\n",
            "epoch: 5 step: 996 train batch loss: tf.Tensor(0.30410796, shape=(), dtype=float32)\n",
            "epoch: 5 step: 997 train batch loss: tf.Tensor(0.20959905, shape=(), dtype=float32)\n",
            "epoch: 5 step: 998 train batch loss: tf.Tensor(0.27574193, shape=(), dtype=float32)\n",
            "epoch: 5 step: 999 train batch loss: tf.Tensor(0.27055117, shape=(), dtype=float32)\n",
            "epoch: 5 step: 0 val batch loss: tf.Tensor(0.62455356, shape=(), dtype=float32)\n",
            "epoch: 5 step: 1 val batch loss: tf.Tensor(0.76396954, shape=(), dtype=float32)\n",
            "epoch: 5 step: 2 val batch loss: tf.Tensor(0.6504959, shape=(), dtype=float32)\n",
            "epoch: 5 step: 3 val batch loss: tf.Tensor(1.1016843, shape=(), dtype=float32)\n",
            "epoch: 5 step: 4 val batch loss: tf.Tensor(0.6384678, shape=(), dtype=float32)\n",
            "epoch: 5 step: 5 val batch loss: tf.Tensor(0.6254046, shape=(), dtype=float32)\n",
            "epoch: 5 step: 6 val batch loss: tf.Tensor(0.43475604, shape=(), dtype=float32)\n",
            "epoch: 5 step: 7 val batch loss: tf.Tensor(0.5297149, shape=(), dtype=float32)\n",
            "epoch: 5 step: 8 val batch loss: tf.Tensor(0.45414996, shape=(), dtype=float32)\n",
            "epoch: 5 step: 9 val batch loss: tf.Tensor(0.45176506, shape=(), dtype=float32)\n",
            "epoch: 5 step: 10 val batch loss: tf.Tensor(0.61770785, shape=(), dtype=float32)\n",
            "epoch: 5 step: 11 val batch loss: tf.Tensor(0.75263065, shape=(), dtype=float32)\n",
            "epoch: 5 step: 12 val batch loss: tf.Tensor(0.5423356, shape=(), dtype=float32)\n",
            "epoch: 5 step: 13 val batch loss: tf.Tensor(0.44218063, shape=(), dtype=float32)\n",
            "epoch: 5 step: 14 val batch loss: tf.Tensor(0.8011314, shape=(), dtype=float32)\n",
            "epoch: 5 step: 15 val batch loss: tf.Tensor(0.46189117, shape=(), dtype=float32)\n",
            "epoch: 5 step: 16 val batch loss: tf.Tensor(0.7087436, shape=(), dtype=float32)\n",
            "epoch: 5 step: 17 val batch loss: tf.Tensor(0.9463476, shape=(), dtype=float32)\n",
            "epoch: 5 step: 18 val batch loss: tf.Tensor(0.8326408, shape=(), dtype=float32)\n",
            "epoch: 5 step: 19 val batch loss: tf.Tensor(0.4194509, shape=(), dtype=float32)\n",
            "epoch: 5 step: 20 val batch loss: tf.Tensor(0.5089813, shape=(), dtype=float32)\n",
            "epoch: 5 step: 21 val batch loss: tf.Tensor(0.64998543, shape=(), dtype=float32)\n",
            "epoch: 5 step: 22 val batch loss: tf.Tensor(0.69407856, shape=(), dtype=float32)\n",
            "epoch: 5 step: 23 val batch loss: tf.Tensor(0.6644494, shape=(), dtype=float32)\n",
            "epoch: 5 step: 24 val batch loss: tf.Tensor(0.759756, shape=(), dtype=float32)\n",
            "epoch: 5 step: 25 val batch loss: tf.Tensor(0.791794, shape=(), dtype=float32)\n",
            "epoch: 5 step: 26 val batch loss: tf.Tensor(0.5040197, shape=(), dtype=float32)\n",
            "epoch: 5 step: 27 val batch loss: tf.Tensor(0.8704179, shape=(), dtype=float32)\n",
            "epoch: 5 step: 28 val batch loss: tf.Tensor(0.764954, shape=(), dtype=float32)\n",
            "epoch: 5 step: 29 val batch loss: tf.Tensor(0.51062953, shape=(), dtype=float32)\n",
            "epoch: 5 step: 30 val batch loss: tf.Tensor(0.69385934, shape=(), dtype=float32)\n",
            "epoch: 5 step: 31 val batch loss: tf.Tensor(0.5735837, shape=(), dtype=float32)\n",
            "epoch: 5 step: 32 val batch loss: tf.Tensor(0.68631065, shape=(), dtype=float32)\n",
            "epoch: 5 step: 33 val batch loss: tf.Tensor(0.67004883, shape=(), dtype=float32)\n",
            "epoch: 5 step: 34 val batch loss: tf.Tensor(0.7412648, shape=(), dtype=float32)\n",
            "epoch: 5 step: 35 val batch loss: tf.Tensor(0.46532926, shape=(), dtype=float32)\n",
            "epoch: 5 step: 36 val batch loss: tf.Tensor(0.55656254, shape=(), dtype=float32)\n",
            "epoch: 5 step: 37 val batch loss: tf.Tensor(0.539832, shape=(), dtype=float32)\n",
            "epoch: 5 step: 38 val batch loss: tf.Tensor(0.68783146, shape=(), dtype=float32)\n",
            "epoch: 5 step: 39 val batch loss: tf.Tensor(0.4955141, shape=(), dtype=float32)\n",
            "epoch: 5 step: 40 val batch loss: tf.Tensor(0.7048641, shape=(), dtype=float32)\n",
            "epoch: 5 step: 41 val batch loss: tf.Tensor(0.55813026, shape=(), dtype=float32)\n",
            "epoch: 5 step: 42 val batch loss: tf.Tensor(0.7638391, shape=(), dtype=float32)\n",
            "epoch: 5 step: 43 val batch loss: tf.Tensor(0.59060425, shape=(), dtype=float32)\n",
            "epoch: 5 step: 44 val batch loss: tf.Tensor(0.6064022, shape=(), dtype=float32)\n",
            "epoch: 5 step: 45 val batch loss: tf.Tensor(0.715793, shape=(), dtype=float32)\n",
            "epoch: 5 step: 46 val batch loss: tf.Tensor(0.8351699, shape=(), dtype=float32)\n",
            "epoch: 5 step: 47 val batch loss: tf.Tensor(0.5974475, shape=(), dtype=float32)\n",
            "epoch: 5 step: 48 val batch loss: tf.Tensor(0.58436424, shape=(), dtype=float32)\n",
            "epoch: 5 step: 49 val batch loss: tf.Tensor(0.6055186, shape=(), dtype=float32)\n",
            "epoch: 5 step: 50 val batch loss: tf.Tensor(0.63106215, shape=(), dtype=float32)\n",
            "epoch: 5 step: 51 val batch loss: tf.Tensor(0.5833356, shape=(), dtype=float32)\n",
            "epoch: 5 step: 52 val batch loss: tf.Tensor(0.6950559, shape=(), dtype=float32)\n",
            "epoch: 5 step: 53 val batch loss: tf.Tensor(0.6775052, shape=(), dtype=float32)\n",
            "epoch: 5 step: 54 val batch loss: tf.Tensor(0.4366919, shape=(), dtype=float32)\n",
            "epoch: 5 step: 55 val batch loss: tf.Tensor(0.78152716, shape=(), dtype=float32)\n",
            "epoch: 5 step: 56 val batch loss: tf.Tensor(0.5217854, shape=(), dtype=float32)\n",
            "epoch: 5 step: 57 val batch loss: tf.Tensor(0.609249, shape=(), dtype=float32)\n",
            "epoch: 5 step: 58 val batch loss: tf.Tensor(0.6862523, shape=(), dtype=float32)\n",
            "epoch: 5 step: 59 val batch loss: tf.Tensor(0.671303, shape=(), dtype=float32)\n",
            "epoch: 5 step: 60 val batch loss: tf.Tensor(0.5748763, shape=(), dtype=float32)\n",
            "epoch: 5 step: 61 val batch loss: tf.Tensor(0.5791817, shape=(), dtype=float32)\n",
            "epoch: 5 step: 62 val batch loss: tf.Tensor(0.7492846, shape=(), dtype=float32)\n",
            "epoch: 5 step: 63 val batch loss: tf.Tensor(0.564825, shape=(), dtype=float32)\n",
            "epoch: 5 step: 64 val batch loss: tf.Tensor(0.5761785, shape=(), dtype=float32)\n",
            "epoch: 5 step: 65 val batch loss: tf.Tensor(0.55315834, shape=(), dtype=float32)\n",
            "epoch: 5 step: 66 val batch loss: tf.Tensor(0.6674517, shape=(), dtype=float32)\n",
            "epoch: 5 step: 67 val batch loss: tf.Tensor(0.81098396, shape=(), dtype=float32)\n",
            "epoch: 5 step: 68 val batch loss: tf.Tensor(0.66277957, shape=(), dtype=float32)\n",
            "epoch: 5 step: 69 val batch loss: tf.Tensor(0.50154245, shape=(), dtype=float32)\n",
            "epoch: 5 step: 70 val batch loss: tf.Tensor(0.81984997, shape=(), dtype=float32)\n",
            "epoch: 5 step: 71 val batch loss: tf.Tensor(0.64702374, shape=(), dtype=float32)\n",
            "epoch: 5 step: 72 val batch loss: tf.Tensor(0.66373664, shape=(), dtype=float32)\n",
            "epoch: 5 step: 73 val batch loss: tf.Tensor(0.7301546, shape=(), dtype=float32)\n",
            "epoch: 5 step: 74 val batch loss: tf.Tensor(0.6819721, shape=(), dtype=float32)\n",
            "epoch: 5 step: 75 val batch loss: tf.Tensor(0.6002487, shape=(), dtype=float32)\n",
            "epoch: 5 step: 76 val batch loss: tf.Tensor(0.72076523, shape=(), dtype=float32)\n",
            "epoch: 5 step: 77 val batch loss: tf.Tensor(0.7084292, shape=(), dtype=float32)\n",
            "epoch: 5 step: 78 val batch loss: tf.Tensor(0.5856682, shape=(), dtype=float32)\n",
            "epoch: 5 step: 79 val batch loss: tf.Tensor(0.678908, shape=(), dtype=float32)\n",
            "epoch: 5 step: 80 val batch loss: tf.Tensor(0.5142074, shape=(), dtype=float32)\n",
            "epoch: 5 step: 81 val batch loss: tf.Tensor(0.6554109, shape=(), dtype=float32)\n",
            "epoch: 5 step: 82 val batch loss: tf.Tensor(0.6482481, shape=(), dtype=float32)\n",
            "epoch: 5 step: 83 val batch loss: tf.Tensor(0.5998453, shape=(), dtype=float32)\n",
            "epoch: 5 step: 84 val batch loss: tf.Tensor(0.51591194, shape=(), dtype=float32)\n",
            "epoch: 5 step: 85 val batch loss: tf.Tensor(0.6106186, shape=(), dtype=float32)\n",
            "epoch: 5 step: 86 val batch loss: tf.Tensor(0.6645456, shape=(), dtype=float32)\n",
            "epoch: 5 step: 87 val batch loss: tf.Tensor(0.64533806, shape=(), dtype=float32)\n",
            "epoch: 5 step: 88 val batch loss: tf.Tensor(0.6258606, shape=(), dtype=float32)\n",
            "epoch: 5 step: 89 val batch loss: tf.Tensor(0.5006632, shape=(), dtype=float32)\n",
            "epoch: 5 step: 90 val batch loss: tf.Tensor(0.9025482, shape=(), dtype=float32)\n",
            "epoch: 5 step: 91 val batch loss: tf.Tensor(0.6997277, shape=(), dtype=float32)\n",
            "epoch: 5 step: 92 val batch loss: tf.Tensor(0.6528312, shape=(), dtype=float32)\n",
            "epoch: 5 step: 93 val batch loss: tf.Tensor(0.6054262, shape=(), dtype=float32)\n",
            "epoch: 5 step: 94 val batch loss: tf.Tensor(0.44350544, shape=(), dtype=float32)\n",
            "epoch: 5 step: 95 val batch loss: tf.Tensor(0.53471774, shape=(), dtype=float32)\n",
            "epoch: 5 step: 96 val batch loss: tf.Tensor(0.44811973, shape=(), dtype=float32)\n",
            "epoch: 5 step: 97 val batch loss: tf.Tensor(0.80562747, shape=(), dtype=float32)\n",
            "epoch: 5 step: 98 val batch loss: tf.Tensor(0.7344445, shape=(), dtype=float32)\n",
            "epoch: 5 step: 99 val batch loss: tf.Tensor(0.66186154, shape=(), dtype=float32)\n",
            "epoch: 5 step: 100 val batch loss: tf.Tensor(0.8764408, shape=(), dtype=float32)\n",
            "epoch: 5 step: 101 val batch loss: tf.Tensor(0.6571451, shape=(), dtype=float32)\n",
            "epoch: 5 step: 102 val batch loss: tf.Tensor(0.7390219, shape=(), dtype=float32)\n",
            "epoch: 5 step: 103 val batch loss: tf.Tensor(0.6015657, shape=(), dtype=float32)\n",
            "epoch: 5 step: 104 val batch loss: tf.Tensor(0.64399457, shape=(), dtype=float32)\n",
            "epoch: 5 step: 105 val batch loss: tf.Tensor(0.45232993, shape=(), dtype=float32)\n",
            "epoch: 5 step: 106 val batch loss: tf.Tensor(0.6887312, shape=(), dtype=float32)\n",
            "epoch: 5 step: 107 val batch loss: tf.Tensor(0.58244336, shape=(), dtype=float32)\n",
            "epoch: 5 step: 108 val batch loss: tf.Tensor(0.61941135, shape=(), dtype=float32)\n",
            "epoch: 5 step: 109 val batch loss: tf.Tensor(0.65027344, shape=(), dtype=float32)\n",
            "epoch: 5 step: 110 val batch loss: tf.Tensor(0.5416115, shape=(), dtype=float32)\n",
            "epoch: 5 step: 111 val batch loss: tf.Tensor(0.5547339, shape=(), dtype=float32)\n",
            "epoch: 5 step: 112 val batch loss: tf.Tensor(0.6766597, shape=(), dtype=float32)\n",
            "epoch: 5 step: 113 val batch loss: tf.Tensor(0.6451697, shape=(), dtype=float32)\n",
            "epoch: 5 step: 114 val batch loss: tf.Tensor(0.60610163, shape=(), dtype=float32)\n",
            "epoch: 5 step: 115 val batch loss: tf.Tensor(0.5412125, shape=(), dtype=float32)\n",
            "epoch: 5 step: 116 val batch loss: tf.Tensor(0.6821898, shape=(), dtype=float32)\n",
            "epoch: 5 step: 117 val batch loss: tf.Tensor(0.5235488, shape=(), dtype=float32)\n",
            "epoch: 5 step: 118 val batch loss: tf.Tensor(0.5806321, shape=(), dtype=float32)\n",
            "epoch: 5 step: 119 val batch loss: tf.Tensor(0.67764187, shape=(), dtype=float32)\n",
            "epoch: 5 step: 120 val batch loss: tf.Tensor(0.5296339, shape=(), dtype=float32)\n",
            "epoch: 5 step: 121 val batch loss: tf.Tensor(0.8976141, shape=(), dtype=float32)\n",
            "epoch: 5 step: 122 val batch loss: tf.Tensor(0.6164161, shape=(), dtype=float32)\n",
            "epoch: 5 step: 123 val batch loss: tf.Tensor(0.6383561, shape=(), dtype=float32)\n",
            "epoch: 5 step: 124 val batch loss: tf.Tensor(0.6871215, shape=(), dtype=float32)\n",
            "epoch: 5 step: 125 val batch loss: tf.Tensor(0.6222734, shape=(), dtype=float32)\n",
            "epoch: 5 step: 126 val batch loss: tf.Tensor(0.7303448, shape=(), dtype=float32)\n",
            "epoch: 5 step: 127 val batch loss: tf.Tensor(0.7190865, shape=(), dtype=float32)\n",
            "epoch: 5 step: 128 val batch loss: tf.Tensor(0.63616, shape=(), dtype=float32)\n",
            "epoch: 5 step: 129 val batch loss: tf.Tensor(0.6168648, shape=(), dtype=float32)\n",
            "epoch: 5 step: 130 val batch loss: tf.Tensor(0.54223907, shape=(), dtype=float32)\n",
            "epoch: 5 step: 131 val batch loss: tf.Tensor(0.6706868, shape=(), dtype=float32)\n",
            "epoch: 5 step: 132 val batch loss: tf.Tensor(0.54152536, shape=(), dtype=float32)\n",
            "epoch: 5 step: 133 val batch loss: tf.Tensor(0.5326599, shape=(), dtype=float32)\n",
            "epoch: 5 step: 134 val batch loss: tf.Tensor(0.65345675, shape=(), dtype=float32)\n",
            "epoch: 5 step: 135 val batch loss: tf.Tensor(0.5220686, shape=(), dtype=float32)\n",
            "epoch: 5 step: 136 val batch loss: tf.Tensor(0.5591201, shape=(), dtype=float32)\n",
            "epoch: 5 step: 137 val batch loss: tf.Tensor(0.5733436, shape=(), dtype=float32)\n",
            "epoch: 5 step: 138 val batch loss: tf.Tensor(0.52676344, shape=(), dtype=float32)\n",
            "epoch: 5 step: 139 val batch loss: tf.Tensor(0.6509197, shape=(), dtype=float32)\n",
            "epoch: 5 step: 140 val batch loss: tf.Tensor(0.5485033, shape=(), dtype=float32)\n",
            "epoch: 5 step: 141 val batch loss: tf.Tensor(0.54476523, shape=(), dtype=float32)\n",
            "epoch: 5 step: 142 val batch loss: tf.Tensor(0.7530712, shape=(), dtype=float32)\n",
            "epoch: 5 step: 143 val batch loss: tf.Tensor(0.650649, shape=(), dtype=float32)\n",
            "epoch: 5 step: 144 val batch loss: tf.Tensor(0.7054996, shape=(), dtype=float32)\n",
            "epoch: 5 step: 145 val batch loss: tf.Tensor(0.65674615, shape=(), dtype=float32)\n",
            "epoch: 5 step: 146 val batch loss: tf.Tensor(0.59281516, shape=(), dtype=float32)\n",
            "epoch: 5 step: 147 val batch loss: tf.Tensor(0.69585484, shape=(), dtype=float32)\n",
            "epoch: 5 step: 148 val batch loss: tf.Tensor(0.60791475, shape=(), dtype=float32)\n",
            "epoch: 5 step: 149 val batch loss: tf.Tensor(0.68181294, shape=(), dtype=float32)\n",
            "epoch: 5 step: 150 val batch loss: tf.Tensor(0.5720099, shape=(), dtype=float32)\n",
            "epoch: 5 step: 151 val batch loss: tf.Tensor(0.7423841, shape=(), dtype=float32)\n",
            "epoch: 5 step: 152 val batch loss: tf.Tensor(0.59454834, shape=(), dtype=float32)\n",
            "epoch: 5 step: 153 val batch loss: tf.Tensor(0.5084721, shape=(), dtype=float32)\n",
            "epoch: 5 step: 154 val batch loss: tf.Tensor(0.558486, shape=(), dtype=float32)\n",
            "epoch: 5 step: 155 val batch loss: tf.Tensor(0.70924276, shape=(), dtype=float32)\n",
            "epoch: 5 step: 156 val batch loss: tf.Tensor(0.54925406, shape=(), dtype=float32)\n",
            "epoch: 5 step: 157 val batch loss: tf.Tensor(0.5562049, shape=(), dtype=float32)\n",
            "epoch: 5 step: 158 val batch loss: tf.Tensor(0.5152391, shape=(), dtype=float32)\n",
            "epoch: 5 step: 159 val batch loss: tf.Tensor(0.4863884, shape=(), dtype=float32)\n",
            "epoch: 5 step: 160 val batch loss: tf.Tensor(0.6421729, shape=(), dtype=float32)\n",
            "epoch: 5 step: 161 val batch loss: tf.Tensor(0.47011274, shape=(), dtype=float32)\n",
            "epoch: 5 step: 162 val batch loss: tf.Tensor(0.753257, shape=(), dtype=float32)\n",
            "epoch: 5 step: 163 val batch loss: tf.Tensor(0.6613172, shape=(), dtype=float32)\n",
            "epoch: 5 step: 164 val batch loss: tf.Tensor(0.81888324, shape=(), dtype=float32)\n",
            "epoch: 5 step: 165 val batch loss: tf.Tensor(0.71557057, shape=(), dtype=float32)\n",
            "epoch: 5 step: 166 val batch loss: tf.Tensor(0.62356246, shape=(), dtype=float32)\n",
            "epoch: 5 step: 167 val batch loss: tf.Tensor(0.5695492, shape=(), dtype=float32)\n",
            "epoch: 5 step: 168 val batch loss: tf.Tensor(0.5437777, shape=(), dtype=float32)\n",
            "epoch: 5 step: 169 val batch loss: tf.Tensor(0.45055157, shape=(), dtype=float32)\n",
            "epoch: 5 step: 170 val batch loss: tf.Tensor(0.64098775, shape=(), dtype=float32)\n",
            "epoch: 5 step: 171 val batch loss: tf.Tensor(0.6211708, shape=(), dtype=float32)\n",
            "epoch: 5 step: 172 val batch loss: tf.Tensor(0.51204574, shape=(), dtype=float32)\n",
            "epoch: 5 step: 173 val batch loss: tf.Tensor(0.55553746, shape=(), dtype=float32)\n",
            "epoch: 5 step: 174 val batch loss: tf.Tensor(0.75957906, shape=(), dtype=float32)\n",
            "epoch: 5 step: 175 val batch loss: tf.Tensor(0.7437746, shape=(), dtype=float32)\n",
            "epoch: 5 step: 176 val batch loss: tf.Tensor(0.5420675, shape=(), dtype=float32)\n",
            "epoch: 5 step: 177 val batch loss: tf.Tensor(0.56715477, shape=(), dtype=float32)\n",
            "epoch: 5 step: 178 val batch loss: tf.Tensor(0.4619051, shape=(), dtype=float32)\n",
            "epoch: 5 step: 179 val batch loss: tf.Tensor(0.6072625, shape=(), dtype=float32)\n",
            "epoch: 5 step: 180 val batch loss: tf.Tensor(0.7114562, shape=(), dtype=float32)\n",
            "epoch: 5 step: 181 val batch loss: tf.Tensor(0.49446097, shape=(), dtype=float32)\n",
            "epoch: 5 step: 182 val batch loss: tf.Tensor(0.61985624, shape=(), dtype=float32)\n",
            "epoch: 5 step: 183 val batch loss: tf.Tensor(0.55706227, shape=(), dtype=float32)\n",
            "epoch: 5 step: 184 val batch loss: tf.Tensor(0.5197901, shape=(), dtype=float32)\n",
            "epoch: 5 step: 185 val batch loss: tf.Tensor(0.66736674, shape=(), dtype=float32)\n",
            "epoch: 5 step: 186 val batch loss: tf.Tensor(0.7632017, shape=(), dtype=float32)\n",
            "epoch: 5 step: 187 val batch loss: tf.Tensor(0.7061168, shape=(), dtype=float32)\n",
            "epoch: 5 step: 188 val batch loss: tf.Tensor(0.54469526, shape=(), dtype=float32)\n",
            "epoch: 5 step: 189 val batch loss: tf.Tensor(0.54454947, shape=(), dtype=float32)\n",
            "epoch: 5 step: 190 val batch loss: tf.Tensor(0.57161605, shape=(), dtype=float32)\n",
            "epoch: 5 step: 191 val batch loss: tf.Tensor(0.6561927, shape=(), dtype=float32)\n",
            "epoch: 5 step: 192 val batch loss: tf.Tensor(0.6658166, shape=(), dtype=float32)\n",
            "epoch: 5 step: 193 val batch loss: tf.Tensor(0.520496, shape=(), dtype=float32)\n",
            "epoch: 5 step: 194 val batch loss: tf.Tensor(0.5647256, shape=(), dtype=float32)\n",
            "epoch: 5 step: 195 val batch loss: tf.Tensor(0.62400496, shape=(), dtype=float32)\n",
            "epoch: 5 step: 196 val batch loss: tf.Tensor(0.62026644, shape=(), dtype=float32)\n",
            "epoch: 5 step: 197 val batch loss: tf.Tensor(0.61275667, shape=(), dtype=float32)\n",
            "epoch: 5 step: 198 val batch loss: tf.Tensor(0.5277201, shape=(), dtype=float32)\n",
            "epoch: 5 step: 199 val batch loss: tf.Tensor(0.45449638, shape=(), dtype=float32)\n",
            "epoch: 5 step: 200 val batch loss: tf.Tensor(0.5374566, shape=(), dtype=float32)\n",
            "epoch: 5 step: 201 val batch loss: tf.Tensor(0.7279666, shape=(), dtype=float32)\n",
            "epoch: 5 step: 202 val batch loss: tf.Tensor(0.547747, shape=(), dtype=float32)\n",
            "epoch: 5 step: 203 val batch loss: tf.Tensor(0.9064562, shape=(), dtype=float32)\n",
            "epoch: 5 step: 204 val batch loss: tf.Tensor(0.50119627, shape=(), dtype=float32)\n",
            "epoch: 5 step: 205 val batch loss: tf.Tensor(0.5929402, shape=(), dtype=float32)\n",
            "epoch: 5 step: 206 val batch loss: tf.Tensor(0.46385306, shape=(), dtype=float32)\n",
            "epoch: 5 step: 207 val batch loss: tf.Tensor(0.549353, shape=(), dtype=float32)\n",
            "epoch: 5 step: 208 val batch loss: tf.Tensor(0.47928548, shape=(), dtype=float32)\n",
            "epoch: 5 step: 209 val batch loss: tf.Tensor(0.45069146, shape=(), dtype=float32)\n",
            "epoch: 5 step: 210 val batch loss: tf.Tensor(0.5482333, shape=(), dtype=float32)\n",
            "epoch: 5 step: 211 val batch loss: tf.Tensor(0.5149354, shape=(), dtype=float32)\n",
            "epoch: 5 step: 212 val batch loss: tf.Tensor(0.6427729, shape=(), dtype=float32)\n",
            "epoch: 5 step: 213 val batch loss: tf.Tensor(0.73986685, shape=(), dtype=float32)\n",
            "epoch: 5 step: 214 val batch loss: tf.Tensor(0.64247763, shape=(), dtype=float32)\n",
            "epoch: 5 step: 215 val batch loss: tf.Tensor(0.55725473, shape=(), dtype=float32)\n",
            "epoch: 5 step: 216 val batch loss: tf.Tensor(0.5179341, shape=(), dtype=float32)\n",
            "epoch: 5 step: 217 val batch loss: tf.Tensor(0.56086755, shape=(), dtype=float32)\n",
            "epoch: 5 step: 218 val batch loss: tf.Tensor(0.6294826, shape=(), dtype=float32)\n",
            "epoch: 5 step: 219 val batch loss: tf.Tensor(0.40481573, shape=(), dtype=float32)\n",
            "epoch: 5 step: 220 val batch loss: tf.Tensor(0.41831598, shape=(), dtype=float32)\n",
            "epoch: 5 step: 221 val batch loss: tf.Tensor(0.69193923, shape=(), dtype=float32)\n",
            "epoch: 5 step: 222 val batch loss: tf.Tensor(0.7554968, shape=(), dtype=float32)\n",
            "epoch: 5 step: 223 val batch loss: tf.Tensor(0.78349036, shape=(), dtype=float32)\n",
            "epoch: 5 step: 224 val batch loss: tf.Tensor(0.7358501, shape=(), dtype=float32)\n",
            "epoch: 5 step: 225 val batch loss: tf.Tensor(0.660892, shape=(), dtype=float32)\n",
            "epoch: 5 step: 226 val batch loss: tf.Tensor(0.7055939, shape=(), dtype=float32)\n",
            "epoch: 5 step: 227 val batch loss: tf.Tensor(0.5966273, shape=(), dtype=float32)\n",
            "epoch: 5 step: 228 val batch loss: tf.Tensor(0.49297857, shape=(), dtype=float32)\n",
            "epoch: 5 step: 229 val batch loss: tf.Tensor(1.0454977, shape=(), dtype=float32)\n",
            "epoch: 5 step: 230 val batch loss: tf.Tensor(0.74330586, shape=(), dtype=float32)\n",
            "epoch: 5 step: 231 val batch loss: tf.Tensor(0.4600916, shape=(), dtype=float32)\n",
            "epoch: 5 step: 232 val batch loss: tf.Tensor(0.73392576, shape=(), dtype=float32)\n",
            "epoch: 5 step: 233 val batch loss: tf.Tensor(0.56260455, shape=(), dtype=float32)\n",
            "epoch: 5 step: 234 val batch loss: tf.Tensor(0.71012866, shape=(), dtype=float32)\n",
            "epoch: 5 step: 235 val batch loss: tf.Tensor(0.58477664, shape=(), dtype=float32)\n",
            "epoch: 5 step: 236 val batch loss: tf.Tensor(0.7301403, shape=(), dtype=float32)\n",
            "epoch: 5 step: 237 val batch loss: tf.Tensor(0.7019197, shape=(), dtype=float32)\n",
            "epoch: 5 step: 238 val batch loss: tf.Tensor(0.6492106, shape=(), dtype=float32)\n",
            "epoch: 5 step: 239 val batch loss: tf.Tensor(0.596612, shape=(), dtype=float32)\n",
            "epoch: 5 step: 240 val batch loss: tf.Tensor(0.63746536, shape=(), dtype=float32)\n",
            "epoch: 5 step: 241 val batch loss: tf.Tensor(0.6449943, shape=(), dtype=float32)\n",
            "epoch: 5 step: 242 val batch loss: tf.Tensor(0.6058508, shape=(), dtype=float32)\n",
            "epoch: 5 step: 243 val batch loss: tf.Tensor(0.71563447, shape=(), dtype=float32)\n",
            "epoch: 5 step: 244 val batch loss: tf.Tensor(0.5694162, shape=(), dtype=float32)\n",
            "epoch: 5 step: 245 val batch loss: tf.Tensor(0.43213263, shape=(), dtype=float32)\n",
            "epoch: 5 step: 246 val batch loss: tf.Tensor(0.461301, shape=(), dtype=float32)\n",
            "epoch: 5 step: 247 val batch loss: tf.Tensor(0.7125011, shape=(), dtype=float32)\n",
            "epoch: 5 step: 248 val batch loss: tf.Tensor(0.85037184, shape=(), dtype=float32)\n",
            "epoch: 5 step: 249 val batch loss: tf.Tensor(0.7076534, shape=(), dtype=float32)\n",
            "Epoch 5 training loss: 0.2878071665763855 val loss: 0.6267526745796204\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 389
        },
        "id": "MhMqedbwQV9K",
        "outputId": "db2c20da-5f13-46b7-bc4f-7d28a65f661f"
      },
      "source": [
        "print(\"Training loss per epoch:\", train_loss_results)\n",
        "print(\"Validation loss per epoch:\", val_loss_results)\n",
        "\n",
        "fig, axes = plt.subplots(1, 2, figsize=(12, 5))\n",
        "\n",
        "# set x axis intervals to 1\n",
        "axes[0].xaxis.set_ticks(np.arange(0, len(train_loss_results), 1))\n",
        "axes[1].xaxis.set_ticks(np.arange(0, len(train_loss_results), 1))\n",
        "\n",
        "axes[0].set_xlabel(\"Epochs\", fontsize=14)\n",
        "axes[0].set_ylabel(\"Loss\", fontsize=14)\n",
        "axes[0].set_title('Training Loss vs epochs')\n",
        "axes[0].plot(train_loss_results)\n",
        "\n",
        "axes[1].set_xlabel(\"Epochs\", fontsize=14)\n",
        "axes[1].set_ylabel(\"Loss\", fontsize=14)\n",
        "axes[1].set_title('Val Loss vs epochs')\n",
        "axes[1].plot(val_loss_results)\n",
        "plt.show()"
      ],
      "execution_count": 88,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training loss per epoch: [1.0224138, 0.7304562, 0.57923293, 0.4592395, 0.36377314, 0.28780717]\n",
            "Validation loss per epoch: [0.8419979, 0.73068273, 0.6761334, 0.646281, 0.63278794, 0.6267527]\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QgPYP3AVQV9K"
      },
      "source": [
        "## 7. Use the model to translate\n",
        "Now it's time to put your model into practice! You should run your translation for five randomly sampled English sentences from the dataset. For each sentence, the process is as follows:\n",
        "* Preprocess and embed the English sentence according to the model requirements.\n",
        "* Pass the embedded sentence through the encoder to get the encoder hidden and cell states.\n",
        "* Starting with the special  `\"<start>\"` token, use this token and the final encoder hidden and cell states to get the one-step prediction from the decoder, as well as the decoder’s updated hidden and cell states.\n",
        "* Create a loop to get the next step prediction and updated hidden and cell states from the decoder, using the most recent hidden and cell states. Terminate the loop when the `\"<end>\"` token is emitted, or when the sentence has reached a maximum length.\n",
        "* Decode the output token sequence into German text and print the English text and the model's German translation."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5lM6V3XiQV9L",
        "outputId": "ca6ad5c7-e1aa-40f4-e897-d79a98870024"
      },
      "source": [
        "total_sent = len(data_examples)\n",
        "random_samples = random.sample(range(0, total_sent), 5)\n",
        "\n",
        "english_sentences =[]\n",
        "embedded_eng_sentences = []\n",
        "\n",
        "for sentence_index in random_samples:\n",
        "    eng, deu, rest = data_examples[sentence_index].split('\\t')\n",
        "    eng = preprocess_sentence(eng)\n",
        "    english_sentences.append(eng)\n",
        "\n",
        "for sent in english_sentences:\n",
        "    print(\"ENGLISH:\", sent)\n",
        "    eng_embedded_sent = embedding_layer(tf.strings.split(sent))\n",
        "    paddings = tf.constant([[13, 0], [0, 0]])\n",
        "    x = tf.pad(eng_embedded_sent, paddings, \"CONSTANT\")\n",
        "    padded_eng_sent = x[-13:,:]\n",
        "    padded_eng_sent = tf.expand_dims(padded_eng_sent, axis=0)\n",
        "    discard_output_seq, eng_final_hidden_state, eng_final_cell_state = encoder(padded_eng_sent)\n",
        "    deu_sentence = \"\"\n",
        "    decoder_hidden_state, decoder_cell_state = eng_final_hidden_state, eng_final_cell_state\n",
        "    token = start_token = input_sentence = '<start>'\n",
        "    counter = 0\n",
        "    while token != '<end>':\n",
        "        deu_index = tokenizer.texts_to_sequences([input_sentence])\n",
        "        output_seq, decoder_hidden_state, decoder_cell_state = decoder(tf.constant(deu_index), decoder_hidden_state, decoder_cell_state)\n",
        "        seq = []\n",
        "        seq.append([np.argmax(output_seq[:,-1])])\n",
        "        token = tokenizer.sequences_to_texts(seq)[0]\n",
        "        if token != '<end>':\n",
        "            deu_sentence += \" {0}\".format(token)\n",
        "            input_sentence = \"{0}{1}\".format(start_token, deu_sentence)\n",
        "        counter += 1\n",
        "        if counter > 12:\n",
        "            break\n",
        "    print(\"GERMAN:\", deu_sentence,\"\\n\")\n"
      ],
      "execution_count": 89,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "ENGLISH: he sang and sang .\n",
            "GERMAN:  er hat aus . \n",
            "\n",
            "ENGLISH: i was sick .\n",
            "GERMAN:  ich bin tun . \n",
            "\n",
            "ENGLISH: is this art ?\n",
            "GERMAN:  ist das ? \n",
            "\n",
            "ENGLISH: tom talks funny .\n",
            "GERMAN:  tom gut . \n",
            "\n",
            "ENGLISH: pick any card .\n",
            "GERMAN:  mach die regeln ! \n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pamIAB_AQV9L"
      },
      "source": [
        ""
      ],
      "execution_count": 89,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0LsnMBvIQV9L"
      },
      "source": [
        ""
      ],
      "execution_count": 89,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lh89zp97QV9L"
      },
      "source": [
        ""
      ],
      "execution_count": 89,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MBA9-Yd_QV9M"
      },
      "source": [
        ""
      ],
      "execution_count": 89,
      "outputs": []
    }
  ]
}