Kerasautoencoders.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Kerasautoencoders.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyOFyfAbKlW3bp/x2Jgv09D1",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/nouranali/SureStart_VAILTraining/blob/master/Kerasautoencoders.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iNnOszXEJNQs"
      },
      "source": [
        "## In this notebook I'll be learning about autoencoders and how to use them with upsampling"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CIJmIvMIJmjw"
      },
      "source": [
        "### I'm learning from this [tutorial](https://blog.keras.io/building-autoencoders-in-keras.html) which has both demonstration and code example "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-3Fd1UKgJK9Q"
      },
      "source": [
        "# simple example\n",
        "from keras.datasets import mnist\n",
        "import numpy as np\n",
        "\n",
        "# This is the size of our encoded representations\n",
        "encoding_dim = 32  # 32 floats -> compression of factor 24.5, assuming the input is 784 floats\n",
        "\n",
        "# This is our input image\n",
        "input_img = keras.Input(shape=(784,))\n",
        "# \"encoded\" is the encoded representation of the input\n",
        "encoded = layers.Dense(encoding_dim, activation='relu')(input_img)\n",
        "# \"decoded\" is the lossy reconstruction of the input\n",
        "decoded = layers.Dense(784, activation='sigmoid')(encoded)\n",
        "# This model maps an input to its reconstruction\n",
        "autoencoder = keras.Model(input_img, decoded)"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "s44QZjUWLJCt"
      },
      "source": [
        "encoder = keras.Model(input_img, encoded)"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ywC_-mwALOQs"
      },
      "source": [
        "# This is our encoded (32-dimensional) input\n",
        "encoded_input = keras.Input(shape=(encoding_dim,))\n",
        "# Retrieve the last layer of the autoencoder model\n",
        "decoder_layer = autoencoder.layers[-1]\n",
        "# Create the decoder model\n",
        "decoder = keras.Model(encoded_input, decoder_layer(encoded_input))"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V--RyUEiLQxj"
      },
      "source": [
        "autoencoder.compile(optimizer='adam', loss='binary_crossentropy')"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4lwDARmyLWWp"
      },
      "source": [
        "MNIST with autoencoders, we'll discard the labels as we'll decode and encode the images"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Sh2Y2KrNLVrD",
        "outputId": "75577508-c05d-4f0f-efc5-380713eaa936"
      },
      "source": [
        "from keras.datasets import mnist\n",
        "import numpy as np\n",
        "(x_train, _), (x_test, _) = mnist.load_data()"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz\n",
            "11493376/11490434 [==============================] - 0s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "SGJcrQqrLiCc",
        "outputId": "37bd67b0-5ce1-40b2-af06-b4d69a2ad4d2"
      },
      "source": [
        "# normalization of data and reshaping\n",
        "x_train = x_train.astype('float32') / 255.\n",
        "x_test = x_test.astype('float32') / 255.\n",
        "x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))\n",
        "x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))\n",
        "print(x_train.shape)\n",
        "print(x_test.shape)"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(60000, 784)\n",
            "(10000, 784)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "FuEea9XXLmdT",
        "outputId": "7ca00f5e-47b8-4c0b-df40-7923d91d4923"
      },
      "source": [
        "# train the autoencoder\n",
        "autoencoder.fit(x_train, x_train,\n",
        "                epochs=50,\n",
        "                batch_size=256,\n",
        "                shuffle=True,\n",
        "                validation_data=(x_test, x_test))"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "235/235 [==============================] - 3s 10ms/step - loss: 0.3830 - val_loss: 0.1875\n",
            "Epoch 2/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.1790 - val_loss: 0.1532\n",
            "Epoch 3/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.1491 - val_loss: 0.1336\n",
            "Epoch 4/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.1317 - val_loss: 0.1212\n",
            "Epoch 5/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.1205 - val_loss: 0.1130\n",
            "Epoch 6/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.1126 - val_loss: 0.1072\n",
            "Epoch 7/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.1071 - val_loss: 0.1030\n",
            "Epoch 8/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.1033 - val_loss: 0.0998\n",
            "Epoch 9/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.1003 - val_loss: 0.0974\n",
            "Epoch 10/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0979 - val_loss: 0.0958\n",
            "Epoch 11/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0967 - val_loss: 0.0946\n",
            "Epoch 12/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0957 - val_loss: 0.0939\n",
            "Epoch 13/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0948 - val_loss: 0.0935\n",
            "Epoch 14/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0949 - val_loss: 0.0932\n",
            "Epoch 15/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0944 - val_loss: 0.0930\n",
            "Epoch 16/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0941 - val_loss: 0.0928\n",
            "Epoch 17/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0939 - val_loss: 0.0927\n",
            "Epoch 18/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0937 - val_loss: 0.0925\n",
            "Epoch 19/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0937 - val_loss: 0.0924\n",
            "Epoch 20/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0936 - val_loss: 0.0923\n",
            "Epoch 21/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0934 - val_loss: 0.0922\n",
            "Epoch 22/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0935 - val_loss: 0.0922\n",
            "Epoch 23/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0933 - val_loss: 0.0921\n",
            "Epoch 24/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0933 - val_loss: 0.0920\n",
            "Epoch 25/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0931 - val_loss: 0.0920\n",
            "Epoch 26/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0931 - val_loss: 0.0920\n",
            "Epoch 27/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0932 - val_loss: 0.0920\n",
            "Epoch 28/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0929 - val_loss: 0.0919\n",
            "Epoch 29/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0930 - val_loss: 0.0919\n",
            "Epoch 30/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0931 - val_loss: 0.0918\n",
            "Epoch 31/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0932 - val_loss: 0.0918\n",
            "Epoch 32/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0928 - val_loss: 0.0918\n",
            "Epoch 33/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0929 - val_loss: 0.0918\n",
            "Epoch 34/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0930 - val_loss: 0.0918\n",
            "Epoch 35/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0931 - val_loss: 0.0917\n",
            "Epoch 36/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0928 - val_loss: 0.0917\n",
            "Epoch 37/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0929 - val_loss: 0.0917\n",
            "Epoch 38/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0927 - val_loss: 0.0918\n",
            "Epoch 39/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0927 - val_loss: 0.0916\n",
            "Epoch 40/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0928 - val_loss: 0.0916\n",
            "Epoch 41/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0928 - val_loss: 0.0917\n",
            "Epoch 42/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0927 - val_loss: 0.0916\n",
            "Epoch 43/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0928 - val_loss: 0.0915\n",
            "Epoch 44/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0929 - val_loss: 0.0916\n",
            "Epoch 45/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0926 - val_loss: 0.0916\n",
            "Epoch 46/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0925 - val_loss: 0.0916\n",
            "Epoch 47/50\n",
            "235/235 [==============================] - 2s 9ms/step - loss: 0.0928 - val_loss: 0.0915\n",
            "Epoch 48/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0924 - val_loss: 0.0916\n",
            "Epoch 49/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0926 - val_loss: 0.0915\n",
            "Epoch 50/50\n",
            "235/235 [==============================] - 2s 8ms/step - loss: 0.0928 - val_loss: 0.0915\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f7a86eff210>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OeOTdafXLreO"
      },
      "source": [
        "# encode/decode images\n",
        "encoded_imgs = encoder.predict(x_test)\n",
        "decoded_imgs = decoder.predict(encoded_imgs)"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 248
        },
        "id": "IacE_uFiMpVO",
        "outputId": "911e3ce2-c7cd-4c39-a390-b72b7258f06d"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "n = 10  # How many digits we will display\n",
        "plt.figure(figsize=(20, 4))\n",
        "for i in range(n):\n",
        "    # Display original\n",
        "    ax = plt.subplot(2, n, i + 1)\n",
        "    plt.imshow(x_test[i].reshape(28, 28))\n",
        "    plt.gray()\n",
        "    ax.get_xaxis().set_visible(False)\n",
        "    ax.get_yaxis().set_visible(False)\n",
        "\n",
        "    # Display reconstruction\n",
        "    ax = plt.subplot(2, n, i + 1 + n)\n",
        "    plt.imshow(decoded_imgs[i].reshape(28, 28))\n",
        "    plt.gray()\n",
        "    ax.get_xaxis().set_visible(False)\n",
        "    ax.get_yaxis().set_visible(False)\n",
        "plt.show()"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1440x288 with 20 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OobwOiOSMyOn"
      },
      "source": [
        "Wow, the neural network learnt to deconsruct then construct the images"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "37rG3hzWM9yJ"
      },
      "source": [
        "we'll add a contstraint on hiddedn layer size"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t9DyTBpPMu08"
      },
      "source": [
        "from keras import regularizers\n",
        "\n",
        "encoding_dim = 32\n",
        "\n",
        "input_img = keras.Input(shape=(784,))\n",
        "# Add a Dense layer with a L1 activity regularizer\n",
        "encoded = layers.Dense(encoding_dim, activation='relu',\n",
        "                activity_regularizer=regularizers.l1(10e-5))(input_img)\n",
        "decoded = layers.Dense(784, activation='sigmoid')(encoded)\n",
        "\n",
        "autoencoder = keras.Model(input_img, decoded)"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DQ5XyYTuNlTt"
      },
      "source": [
        "## Deep auto encoder with more layers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b9AjN18mNZfy"
      },
      "source": [
        "input_img = keras.Input(shape=(784,))\n",
        "encoded = layers.Dense(128, activation='relu')(input_img)\n",
        "encoded = layers.Dense(64, activation='relu')(encoded)\n",
        "encoded = layers.Dense(32, activation='relu')(encoded)\n",
        "\n",
        "decoded = layers.Dense(64, activation='relu')(encoded)\n",
        "decoded = layers.Dense(128, activation='relu')(decoded)\n",
        "decoded = layers.Dense(784, activation='sigmoid')(decoded)"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DjhVG-b8NtRs",
        "outputId": "17999eee-a28b-49d7-9a34-24a1269ecb1e"
      },
      "source": [
        "autoencoder = keras.Model(input_img, decoded)\n",
        "autoencoder.compile(optimizer='adam', loss='binary_crossentropy')\n",
        "\n",
        "autoencoder.fit(x_train, x_train,\n",
        "                epochs=100,\n",
        "                batch_size=256,\n",
        "                shuffle=True,\n",
        "                validation_data=(x_test, x_test))"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.3354 - val_loss: 0.1639\n",
            "Epoch 2/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.1584 - val_loss: 0.1422\n",
            "Epoch 3/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.1396 - val_loss: 0.1288\n",
            "Epoch 4/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.1282 - val_loss: 0.1206\n",
            "Epoch 5/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.1205 - val_loss: 0.1153\n",
            "Epoch 6/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.1152 - val_loss: 0.1105\n",
            "Epoch 7/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.1110 - val_loss: 0.1071\n",
            "Epoch 8/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.1075 - val_loss: 0.1047\n",
            "Epoch 9/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.1055 - val_loss: 0.1032\n",
            "Epoch 10/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.1039 - val_loss: 0.1013\n",
            "Epoch 11/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.1024 - val_loss: 0.1002\n",
            "Epoch 12/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.1010 - val_loss: 0.0988\n",
            "Epoch 13/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0998 - val_loss: 0.0980\n",
            "Epoch 14/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0986 - val_loss: 0.0969\n",
            "Epoch 15/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0976 - val_loss: 0.0959\n",
            "Epoch 16/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0965 - val_loss: 0.0950\n",
            "Epoch 17/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0956 - val_loss: 0.0945\n",
            "Epoch 18/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0951 - val_loss: 0.0936\n",
            "Epoch 19/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0942 - val_loss: 0.0931\n",
            "Epoch 20/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0938 - val_loss: 0.0923\n",
            "Epoch 21/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0931 - val_loss: 0.0919\n",
            "Epoch 22/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0925 - val_loss: 0.0913\n",
            "Epoch 23/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0919 - val_loss: 0.0909\n",
            "Epoch 24/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0914 - val_loss: 0.0903\n",
            "Epoch 25/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0906 - val_loss: 0.0897\n",
            "Epoch 26/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0902 - val_loss: 0.0894\n",
            "Epoch 27/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0897 - val_loss: 0.0889\n",
            "Epoch 28/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0896 - val_loss: 0.0885\n",
            "Epoch 29/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0892 - val_loss: 0.0883\n",
            "Epoch 30/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0886 - val_loss: 0.0880\n",
            "Epoch 31/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0884 - val_loss: 0.0877\n",
            "Epoch 32/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0884 - val_loss: 0.0874\n",
            "Epoch 33/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0879 - val_loss: 0.0872\n",
            "Epoch 34/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0876 - val_loss: 0.0868\n",
            "Epoch 35/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0876 - val_loss: 0.0867\n",
            "Epoch 36/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0870 - val_loss: 0.0865\n",
            "Epoch 37/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0869 - val_loss: 0.0862\n",
            "Epoch 38/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0868 - val_loss: 0.0859\n",
            "Epoch 39/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0864 - val_loss: 0.0860\n",
            "Epoch 40/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0864 - val_loss: 0.0854\n",
            "Epoch 41/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0858 - val_loss: 0.0849\n",
            "Epoch 42/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0855 - val_loss: 0.0857\n",
            "Epoch 43/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0855 - val_loss: 0.0848\n",
            "Epoch 44/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0851 - val_loss: 0.0845\n",
            "Epoch 45/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0851 - val_loss: 0.0848\n",
            "Epoch 46/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0849 - val_loss: 0.0843\n",
            "Epoch 47/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0848 - val_loss: 0.0842\n",
            "Epoch 48/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0845 - val_loss: 0.0840\n",
            "Epoch 49/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0845 - val_loss: 0.0838\n",
            "Epoch 50/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0843 - val_loss: 0.0839\n",
            "Epoch 51/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0842 - val_loss: 0.0837\n",
            "Epoch 52/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0842 - val_loss: 0.0835\n",
            "Epoch 53/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0840 - val_loss: 0.0835\n",
            "Epoch 54/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0841 - val_loss: 0.0833\n",
            "Epoch 55/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0837 - val_loss: 0.0832\n",
            "Epoch 56/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0837 - val_loss: 0.0829\n",
            "Epoch 57/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0834 - val_loss: 0.0832\n",
            "Epoch 58/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0834 - val_loss: 0.0830\n",
            "Epoch 59/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0834 - val_loss: 0.0830\n",
            "Epoch 60/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0830 - val_loss: 0.0829\n",
            "Epoch 61/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0830 - val_loss: 0.0826\n",
            "Epoch 62/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0829 - val_loss: 0.0825\n",
            "Epoch 63/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0830 - val_loss: 0.0826\n",
            "Epoch 64/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0827 - val_loss: 0.0824\n",
            "Epoch 65/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0827 - val_loss: 0.0823\n",
            "Epoch 66/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0827 - val_loss: 0.0823\n",
            "Epoch 67/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0826 - val_loss: 0.0820\n",
            "Epoch 68/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0824 - val_loss: 0.0822\n",
            "Epoch 69/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0823 - val_loss: 0.0822\n",
            "Epoch 70/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0824 - val_loss: 0.0821\n",
            "Epoch 71/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0823 - val_loss: 0.0822\n",
            "Epoch 72/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0822 - val_loss: 0.0822\n",
            "Epoch 73/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0825 - val_loss: 0.0820\n",
            "Epoch 74/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0822 - val_loss: 0.0819\n",
            "Epoch 75/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0820 - val_loss: 0.0819\n",
            "Epoch 76/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0822 - val_loss: 0.0816\n",
            "Epoch 77/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0820 - val_loss: 0.0818\n",
            "Epoch 78/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0818 - val_loss: 0.0818\n",
            "Epoch 79/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0819 - val_loss: 0.0819\n",
            "Epoch 80/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0821 - val_loss: 0.0816\n",
            "Epoch 81/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0819 - val_loss: 0.0815\n",
            "Epoch 82/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0817 - val_loss: 0.0817\n",
            "Epoch 83/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0816 - val_loss: 0.0816\n",
            "Epoch 84/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0815 - val_loss: 0.0815\n",
            "Epoch 85/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0816 - val_loss: 0.0817\n",
            "Epoch 86/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0818 - val_loss: 0.0816\n",
            "Epoch 87/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0816 - val_loss: 0.0816\n",
            "Epoch 88/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0817 - val_loss: 0.0813\n",
            "Epoch 89/100\n",
            "235/235 [==============================] - 3s 13ms/step - loss: 0.0817 - val_loss: 0.0814\n",
            "Epoch 90/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0817 - val_loss: 0.0813\n",
            "Epoch 91/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0814 - val_loss: 0.0814\n",
            "Epoch 92/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0813 - val_loss: 0.0815\n",
            "Epoch 93/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0816 - val_loss: 0.0814\n",
            "Epoch 94/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0816 - val_loss: 0.0811\n",
            "Epoch 95/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0812 - val_loss: 0.0812\n",
            "Epoch 96/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0813 - val_loss: 0.0813\n",
            "Epoch 97/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0815 - val_loss: 0.0814\n",
            "Epoch 98/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0814 - val_loss: 0.0814\n",
            "Epoch 99/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0812 - val_loss: 0.0812\n",
            "Epoch 100/100\n",
            "235/235 [==============================] - 3s 12ms/step - loss: 0.0812 - val_loss: 0.0811\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f7a8232a290>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zMeArQG7N2Lu"
      },
      "source": [
        "## convolutional autoencoders"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uM7xNmwgNu96"
      },
      "source": [
        "input_img = keras.Input(shape=(28, 28, 1))\n",
        "\n",
        "x = layers.Conv2D(16, (3, 3), activation='relu', padding='same')(input_img)\n",
        "x = layers.MaxPooling2D((2, 2), padding='same')(x)\n",
        "x = layers.Conv2D(8, (3, 3), activation='relu', padding='same')(x)\n",
        "x = layers.MaxPooling2D((2, 2), padding='same')(x)\n",
        "x = layers.Conv2D(8, (3, 3), activation='relu', padding='same')(x)\n",
        "encoded = layers.MaxPooling2D((2, 2), padding='same')(x)\n",
        "\n",
        "# at this point the representation is (4, 4, 8) i.e. 128-dimensional\n",
        "\n",
        "x = layers.Conv2D(8, (3, 3), activation='relu', padding='same')(encoded)\n",
        "x = layers.UpSampling2D((2, 2))(x)\n",
        "x = layers.Conv2D(8, (3, 3), activation='relu', padding='same')(x)\n",
        "x = layers.UpSampling2D((2, 2))(x)\n",
        "x = layers.Conv2D(16, (3, 3), activation='relu')(x)\n",
        "x = layers.UpSampling2D((2, 2))(x)\n",
        "decoded = layers.Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)\n",
        "\n",
        "autoencoder = keras.Model(input_img, decoded)\n",
        "autoencoder.compile(optimizer='adam', loss='binary_crossentropy')"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OnompxjaN-2P"
      },
      "source": [
        "(x_train, _), (x_test, _) = mnist.load_data()\n",
        "\n",
        "x_train = x_train.astype('float32') / 255.\n",
        "x_test = x_test.astype('float32') / 255.\n",
        "x_train = np.reshape(x_train, (len(x_train), 28, 28, 1))\n",
        "x_test = np.reshape(x_test, (len(x_test), 28, 28, 1))"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yiayB7Z6OFxK"
      },
      "source": [
        "%load_ext tensorboard\n",
        "%tensorboard --logdir=\"/tmp/autoencoder\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "UFzeq64GOG4j",
        "outputId": "8885f7af-4ce1-4144-9c10-4341bb0d2cc2"
      },
      "source": [
        "from keras.callbacks import TensorBoard\n",
        "\n",
        "autoencoder.fit(x_train, x_train,\n",
        "                epochs=50,\n",
        "                batch_size=128,\n",
        "                shuffle=True,\n",
        "                validation_data=(x_test, x_test),\n",
        "                callbacks=[TensorBoard(log_dir='/tmp/autoencoder')])"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "469/469 [==============================] - 82s 173ms/step - loss: 0.3269 - val_loss: 0.1457\n",
            "Epoch 2/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.1402 - val_loss: 0.1254\n",
            "Epoch 3/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.1245 - val_loss: 0.1179\n",
            "Epoch 4/50\n",
            "469/469 [==============================] - 80s 171ms/step - loss: 0.1179 - val_loss: 0.1134\n",
            "Epoch 5/50\n",
            "469/469 [==============================] - 80s 172ms/step - loss: 0.1137 - val_loss: 0.1104\n",
            "Epoch 6/50\n",
            "469/469 [==============================] - 80s 172ms/step - loss: 0.1111 - val_loss: 0.1082\n",
            "Epoch 7/50\n",
            "469/469 [==============================] - 80s 172ms/step - loss: 0.1091 - val_loss: 0.1068\n",
            "Epoch 8/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.1075 - val_loss: 0.1058\n",
            "Epoch 9/50\n",
            "469/469 [==============================] - 80s 172ms/step - loss: 0.1066 - val_loss: 0.1041\n",
            "Epoch 10/50\n",
            "469/469 [==============================] - 80s 171ms/step - loss: 0.1051 - val_loss: 0.1030\n",
            "Epoch 11/50\n",
            "469/469 [==============================] - 80s 171ms/step - loss: 0.1044 - val_loss: 0.1028\n",
            "Epoch 12/50\n",
            "469/469 [==============================] - 80s 171ms/step - loss: 0.1036 - val_loss: 0.1015\n",
            "Epoch 13/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.1028 - val_loss: 0.1014\n",
            "Epoch 14/50\n",
            "469/469 [==============================] - 80s 171ms/step - loss: 0.1025 - val_loss: 0.1009\n",
            "Epoch 15/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.1016 - val_loss: 0.1000\n",
            "Epoch 16/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.1010 - val_loss: 0.0999\n",
            "Epoch 17/50\n",
            "469/469 [==============================] - 80s 172ms/step - loss: 0.1009 - val_loss: 0.0992\n",
            "Epoch 18/50\n",
            "469/469 [==============================] - 80s 171ms/step - loss: 0.1003 - val_loss: 0.0987\n",
            "Epoch 19/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.0997 - val_loss: 0.0982\n",
            "Epoch 20/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.0993 - val_loss: 0.0978\n",
            "Epoch 21/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.0989 - val_loss: 0.0982\n",
            "Epoch 22/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0988 - val_loss: 0.0975\n",
            "Epoch 23/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0983 - val_loss: 0.0969\n",
            "Epoch 24/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.0982 - val_loss: 0.0968\n",
            "Epoch 25/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0976 - val_loss: 0.0962\n",
            "Epoch 26/50\n",
            "469/469 [==============================] - 81s 174ms/step - loss: 0.0975 - val_loss: 0.0960\n",
            "Epoch 27/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0972 - val_loss: 0.0957\n",
            "Epoch 28/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0970 - val_loss: 0.0955\n",
            "Epoch 29/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0966 - val_loss: 0.0953\n",
            "Epoch 30/50\n",
            "469/469 [==============================] - 81s 174ms/step - loss: 0.0966 - val_loss: 0.0953\n",
            "Epoch 31/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0963 - val_loss: 0.0949\n",
            "Epoch 32/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0960 - val_loss: 0.0949\n",
            "Epoch 33/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0959 - val_loss: 0.0945\n",
            "Epoch 34/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0956 - val_loss: 0.0943\n",
            "Epoch 35/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0954 - val_loss: 0.0944\n",
            "Epoch 36/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0954 - val_loss: 0.0942\n",
            "Epoch 37/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0951 - val_loss: 0.0939\n",
            "Epoch 38/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0952 - val_loss: 0.0942\n",
            "Epoch 39/50\n",
            "469/469 [==============================] - 81s 172ms/step - loss: 0.0951 - val_loss: 0.0936\n",
            "Epoch 40/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0948 - val_loss: 0.0934\n",
            "Epoch 41/50\n",
            "469/469 [==============================] - 81s 174ms/step - loss: 0.0946 - val_loss: 0.0938\n",
            "Epoch 42/50\n",
            "469/469 [==============================] - 82s 174ms/step - loss: 0.0946 - val_loss: 0.0930\n",
            "Epoch 43/50\n",
            "469/469 [==============================] - 81s 174ms/step - loss: 0.0941 - val_loss: 0.0931\n",
            "Epoch 44/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0942 - val_loss: 0.0930\n",
            "Epoch 45/50\n",
            "469/469 [==============================] - 81s 174ms/step - loss: 0.0943 - val_loss: 0.0932\n",
            "Epoch 46/50\n",
            "469/469 [==============================] - 82s 174ms/step - loss: 0.0942 - val_loss: 0.0927\n",
            "Epoch 47/50\n",
            "469/469 [==============================] - 81s 174ms/step - loss: 0.0939 - val_loss: 0.0926\n",
            "Epoch 48/50\n",
            "469/469 [==============================] - 82s 174ms/step - loss: 0.0937 - val_loss: 0.0926\n",
            "Epoch 49/50\n",
            "469/469 [==============================] - 82s 174ms/step - loss: 0.0935 - val_loss: 0.0929\n",
            "Epoch 50/50\n",
            "469/469 [==============================] - 81s 173ms/step - loss: 0.0935 - val_loss: 0.0925\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f7a7fa5dad0>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 25
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 248
        },
        "id": "fCirTxYLOq3r",
        "outputId": "2daa9390-526a-4c73-a1a6-558f4f354f52"
      },
      "source": [
        "decoded_imgs = autoencoder.predict(x_test)\n",
        "\n",
        "n = 10\n",
        "plt.figure(figsize=(20, 4))\n",
        "for i in range(1, n + 1):\n",
        "    # Display original\n",
        "    ax = plt.subplot(2, n, i)\n",
        "    plt.imshow(x_test[i].reshape(28, 28))\n",
        "    plt.gray()\n",
        "    ax.get_xaxis().set_visible(False)\n",
        "    ax.get_yaxis().set_visible(False)\n",
        "\n",
        "    # Display reconstruction\n",
        "    ax = plt.subplot(2, n, i + n)\n",
        "    plt.imshow(decoded_imgs[i].reshape(28, 28))\n",
        "    plt.gray()\n",
        "    ax.get_xaxis().set_visible(False)\n",
        "    ax.get_yaxis().set_visible(False)\n",
        "plt.show()"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1440x288 with 20 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 469
        },
        "id": "67VYAKLjOwUh",
        "outputId": "bc222e0f-a7cd-4f97-968e-5c2109c9acbd"
      },
      "source": [
        "encoder = keras.Model(input_img, encoded)\n",
        "encoded_imgs = encoder.predict(x_test)\n",
        "\n",
        "n = 10\n",
        "plt.figure(figsize=(20, 8))\n",
        "for i in range(1, n + 1):\n",
        "    ax = plt.subplot(1, n, i)\n",
        "    plt.imshow(encoded_imgs[i].reshape((4, 4 * 8)).T)\n",
        "    plt.gray()\n",
        "    ax.get_xaxis().set_visible(False)\n",
        "    ax.get_yaxis().set_visible(False)\n",
        "plt.show()"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1440x576 with 10 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qnl6W31KPcmo"
      },
      "source": [
        "## Image denoising"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bIAnorljPcOs"
      },
      "source": [
        "(x_train, _), (x_test, _) = mnist.load_data()\n",
        "\n",
        "x_train = x_train.astype('float32') / 255.\n",
        "x_test = x_test.astype('float32') / 255.\n",
        "x_train = np.reshape(x_train, (len(x_train), 28, 28, 1))\n",
        "x_test = np.reshape(x_test, (len(x_test), 28, 28, 1))\n",
        "## add noise to immages\n",
        "noise_factor = 0.5\n",
        "x_train_noisy = x_train + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_train.shape) \n",
        "x_test_noisy = x_test + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_test.shape) \n",
        "\n",
        "x_train_noisy = np.clip(x_train_noisy, 0., 1.)\n",
        "x_test_noisy = np.clip(x_test_noisy, 0., 1.)\n"
      ],
      "execution_count": 28,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 129
        },
        "id": "kJ13H-6hPk5n",
        "outputId": "7b6e6ed0-3eb3-4cd1-8438-d26fb4339891"
      },
      "source": [
        "n = 10\n",
        "plt.figure(figsize=(20, 2))\n",
        "for i in range(1, n + 1):\n",
        "    ax = plt.subplot(1, n, i)\n",
        "    plt.imshow(x_test_noisy[i].reshape(28, 28))\n",
        "    plt.gray()\n",
        "    ax.get_xaxis().set_visible(False)\n",
        "    ax.get_yaxis().set_visible(False)\n",
        "plt.show()"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1440x144 with 10 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FvOrNkp3PmyX"
      },
      "source": [
        "# build conv autoencoder to denoise\n",
        "input_img = keras.Input(shape=(28, 28, 1))\n",
        "\n",
        "x = layers.Conv2D(32, (3, 3), activation='relu', padding='same')(input_img)\n",
        "x = layers.MaxPooling2D((2, 2), padding='same')(x)\n",
        "x = layers.Conv2D(32, (3, 3), activation='relu', padding='same')(x)\n",
        "encoded = layers.MaxPooling2D((2, 2), padding='same')(x)\n",
        "\n",
        "# At this point the representation is (7, 7, 32)\n",
        "\n",
        "x = layers.Conv2D(32, (3, 3), activation='relu', padding='same')(encoded)\n",
        "x = layers.UpSampling2D((2, 2))(x)\n",
        "x = layers.Conv2D(32, (3, 3), activation='relu', padding='same')(x)\n",
        "x = layers.UpSampling2D((2, 2))(x)\n",
        "decoded = layers.Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)\n",
        "\n",
        "autoencoder = keras.Model(input_img, decoded)\n",
        "autoencoder.compile(optimizer='adam', loss='binary_crossentropy')"
      ],
      "execution_count": 30,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4UxnX-hYPsY-",
        "outputId": "99170f0a-cb09-4b78-b0d8-53eeae192920"
      },
      "source": [
        "autoencoder.fit(x_train_noisy, x_train,\n",
        "                epochs=100,\n",
        "                batch_size=128,\n",
        "                shuffle=True,\n",
        "                validation_data=(x_test_noisy, x_test),\n",
        "                callbacks=[TensorBoard(log_dir='/tmp/tb', histogram_freq=0, write_graph=False)])"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/100\n",
            "469/469 [==============================] - 147s 312ms/step - loss: 0.2628 - val_loss: 0.1158\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 147s 313ms/step - loss: 0.1147 - val_loss: 0.1086\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 147s 314ms/step - loss: 0.1085 - val_loss: 0.1054\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 147s 313ms/step - loss: 0.1053 - val_loss: 0.1029\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 147s 314ms/step - loss: 0.1035 - val_loss: 0.1016\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 146s 312ms/step - loss: 0.1021 - val_loss: 0.1009\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 147s 313ms/step - loss: 0.1009 - val_loss: 0.0995\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 146s 311ms/step - loss: 0.1001 - val_loss: 0.0988\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 145s 310ms/step - loss: 0.0992 - val_loss: 0.0979\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 146s 311ms/step - loss: 0.0984 - val_loss: 0.0977\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 146s 311ms/step - loss: 0.0981 - val_loss: 0.0972\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 146s 311ms/step - loss: 0.0976 - val_loss: 0.0967\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 146s 311ms/step - loss: 0.0972 - val_loss: 0.0966\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 146s 311ms/step - loss: 0.0969 - val_loss: 0.0961\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 145s 309ms/step - loss: 0.0965 - val_loss: 0.0959\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 145s 309ms/step - loss: 0.0964 - val_loss: 0.0958\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 145s 309ms/step - loss: 0.0963 - val_loss: 0.0955\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 145s 309ms/step - loss: 0.0959 - val_loss: 0.0952\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 145s 309ms/step - loss: 0.0956 - val_loss: 0.0954\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 145s 310ms/step - loss: 0.0956 - val_loss: 0.0951\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 146s 311ms/step - loss: 0.0954 - val_loss: 0.0949\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 145s 310ms/step - loss: 0.0953 - val_loss: 0.0949\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 145s 309ms/step - loss: 0.0954 - val_loss: 0.0947\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 145s 309ms/step - loss: 0.0950 - val_loss: 0.0947\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 145s 308ms/step - loss: 0.0950 - val_loss: 0.0945\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 145s 308ms/step - loss: 0.0950 - val_loss: 0.0946\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 144s 308ms/step - loss: 0.0950 - val_loss: 0.0944\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 144s 307ms/step - loss: 0.0947 - val_loss: 0.0944\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 144s 307ms/step - loss: 0.0947 - val_loss: 0.0943\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 144s 307ms/step - loss: 0.0946 - val_loss: 0.0942\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 144s 308ms/step - loss: 0.0947 - val_loss: 0.0945\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 144s 307ms/step - loss: 0.0945 - val_loss: 0.0941\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 143s 305ms/step - loss: 0.0943 - val_loss: 0.0942\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 143s 305ms/step - loss: 0.0945 - val_loss: 0.0942\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 143s 305ms/step - loss: 0.0942 - val_loss: 0.0942\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 144s 306ms/step - loss: 0.0942 - val_loss: 0.0942\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 144s 307ms/step - loss: 0.0944 - val_loss: 0.0940\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 144s 307ms/step - loss: 0.0943 - val_loss: 0.0939\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 144s 306ms/step - loss: 0.0942 - val_loss: 0.0940\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 144s 307ms/step - loss: 0.0941 - val_loss: 0.0940\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 143s 305ms/step - loss: 0.0942 - val_loss: 0.0939\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0942 - val_loss: 0.0937\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0939 - val_loss: 0.0937\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0940 - val_loss: 0.0938\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 143s 304ms/step - loss: 0.0939 - val_loss: 0.0940\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0939 - val_loss: 0.0938\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0938 - val_loss: 0.0936\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0936 - val_loss: 0.0936\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0939 - val_loss: 0.0939\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0938 - val_loss: 0.0936\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0939 - val_loss: 0.0938\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0937 - val_loss: 0.0939\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0938 - val_loss: 0.0937\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0938 - val_loss: 0.0936\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 141s 302ms/step - loss: 0.0937 - val_loss: 0.0938\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0939 - val_loss: 0.0935\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0936 - val_loss: 0.0935\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0935 - val_loss: 0.0935\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0937 - val_loss: 0.0934\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0936 - val_loss: 0.0940\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0936 - val_loss: 0.0936\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0936 - val_loss: 0.0935\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0935 - val_loss: 0.0935\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0934 - val_loss: 0.0933\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0934 - val_loss: 0.0935\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0935 - val_loss: 0.0934\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 141s 302ms/step - loss: 0.0934 - val_loss: 0.0936\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0934 - val_loss: 0.0934\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0933 - val_loss: 0.0934\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0935 - val_loss: 0.0934\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0934 - val_loss: 0.0934\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0934 - val_loss: 0.0936\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0934 - val_loss: 0.0933\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0935 - val_loss: 0.0935\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0932 - val_loss: 0.0933\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0934 - val_loss: 0.0933\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0934 - val_loss: 0.0934\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0934 - val_loss: 0.0936\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0933 - val_loss: 0.0935\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0933 - val_loss: 0.0933\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0932 - val_loss: 0.0933\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0932 - val_loss: 0.0933\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0930 - val_loss: 0.0933\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0932 - val_loss: 0.0932\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 141s 302ms/step - loss: 0.0934 - val_loss: 0.0934\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0933 - val_loss: 0.0933\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0933 - val_loss: 0.0934\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0934 - val_loss: 0.0933\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0932 - val_loss: 0.0935\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0933 - val_loss: 0.0934\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0932 - val_loss: 0.0935\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0932 - val_loss: 0.0934\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0931 - val_loss: 0.0932\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0933 - val_loss: 0.0933\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0932 - val_loss: 0.0932\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0930 - val_loss: 0.0932\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 141s 302ms/step - loss: 0.0932 - val_loss: 0.0935\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 142s 303ms/step - loss: 0.0930 - val_loss: 0.0935\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 142s 302ms/step - loss: 0.0931 - val_loss: 0.0932\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 141s 301ms/step - loss: 0.0932 - val_loss: 0.0932\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f7a828250d0>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 31
        }
      ]
    }
  ]
}