How_to_train_resnet18_with_uL2Q.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-03T08:23:22.758160Z",
     "start_time": "2020-04-03T08:23:21.900626Z"
    }
   },
   "outputs": [],
   "source": [
    "from pylab import *\n",
    "%matplotlib inline\n",
    "caffe_root = './'  # this file should be run from {caffe_root} (otherwise change this line)\n",
    "import sys\n",
    "sys.path.insert(0, caffe_root + 'python')\n",
    "import os\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"3\"\n",
    "import caffe\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-03T08:24:07.784228Z",
     "start_time": "2020-04-03T08:24:07.744694Z"
    }
   },
   "outputs": [],
   "source": [
    "# define solver\n",
    "solver_config_path=caffe_root+'models/ResNet-18/ULQ/resnet18-ulq2_solver-pycaffe.prototxt'\n",
    "from caffe.proto import caffe_pb2\n",
    "s = caffe_pb2.SolverParameter()\n",
    "s.random_seed = 0xCAFFE\n",
    "# load model prototxt\n",
    "s.net = \"models/ResNet-18/ULQ/resnet18_ulq2.prototxt\"\n",
    "# s.net=\"../caffe/models/ResNet-18/train.prototxt\"\n",
    "# s.test_net.append(test_net_path)\n",
    "s.test_interval = 1000  # Test after every 1000 training iterations.\n",
    "s.test_iter.append(100) # Test on 100 batches each time we test.\n",
    "s.max_iter = 55000   # no. of times to update the net (training iterations)\n",
    "# EDIT HERE to try different solvers\n",
    "# solver types include \"SGD\", \"Adam\", and \"Nesterov\" among others.\n",
    "s.type = \"SGD\"\n",
    "# s.type=\"Adam\"\n",
    "# Set the initial learning rate for SGD.\n",
    "s.base_lr = 0.001  # EDIT HERE to try different learning rates\n",
    "# Set momentum to accelerate learning by\n",
    "# taking weighted average of current and previous updates.\n",
    "s.momentum = 0.9\n",
    "# Set weight decay to regularize and prevent overfitting\n",
    "s.weight_decay = 0\n",
    "# Set `lr_policy` to define how the learning rate changes during training.\n",
    "# This is the same policy as our default LeNet.\n",
    "s.lr_policy = 'multistep'\n",
    "s.stepvalue.append(30000)\n",
    "s.stepvalue.append(45000)\n",
    "s.stepvalue.append(50000)\n",
    "s.gamma = 0.1\n",
    "# s.power=1\n",
    "s.test_initialization=False\n",
    "# EDIT HERE to try the fixed rate (and compare with adaptive solvers)\n",
    "# `fixed` is the simplest policy that keeps the learning rate constant.\n",
    "# s.lr_policy = 'fixed'\n",
    "# Display the current training loss and accuracy every 1000 iterations.\n",
    "s.display = 100\n",
    "# Snapshots are files used to store networks we've trained.\n",
    "# We'll snapshot every 5K iterations -- twice during training.\n",
    "s.snapshot = 5000\n",
    "s.snapshot_prefix = \"models/ResNet-18/ULQ/model/resnet-18_ulq2\"\n",
    "# Train on the GPU\n",
    "s.solver_mode = caffe_pb2.SolverParameter.GPU\n",
    "# Write the solver to a temporary file and return its filename.\n",
    "with open(solver_config_path, 'w') as f:\n",
    "    f.write(str(s))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-03T08:24:26.949697Z",
     "start_time": "2020-04-03T08:24:15.582162Z"
    }
   },
   "outputs": [],
   "source": [
    "# If you have the pretraining weights\n",
    "weights=\"../caffe/models/ResNet-18/ULQ/model/resnet-18_ulq2.caffemodel\"\n",
    "### load the solver and create train and test nets\n",
    "# caffe.set_device(0)#set your own gpu id\n",
    "caffe.set_mode_gpu()\n",
    "solver = None  # ignore this workaround for lmdb data (can't instantiate two solvers on the same data)\n",
    "solver = caffe.get_solver(solver_config_path)\n",
    "# solver=caffe.get_solver(\"models/ResNet-18/ULQ/resnet18_ulq2_solver.prototxt\")\n",
    "solver.net.copy_from(weights)\n",
    "solver.test_nets[0].copy_from(weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-03T09:28:27.886462Z",
     "start_time": "2020-04-03T09:28:27.837658Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training iterations= 20000 , Test iterations= 1000\n"
     ]
    }
   ],
   "source": [
    "training_batch_size=64\n",
    "iter_size=4\n",
    "training_samples=1280000\n",
    "train_iterations=int(training_samples/training_batch_size)\n",
    "# train_iterations=10\n",
    "test_batch_size=50\n",
    "test_samples=50000\n",
    "test_iterations=int(test_samples/test_batch_size)\n",
    "# test_iterations=10\n",
    "print \"Training iterations=\", train_iterations, \", Test iterations=\",test_iterations\n",
    "def evaluate(solver, test_iterations):\n",
    "    correct1=0\n",
    "    correct5=0\n",
    "    loss=0\n",
    "    for test_it in range(test_iterations):\n",
    "        print \"\\rTest %.2f%%\"%((test_it+1.0)*100/test_iterations),\n",
    "        solver.test_nets[0].forward()\n",
    "#         correct += sum(solver.test_nets[0].blobs['score'].data.argmax(1)== solver.test_nets[0].blobs['label'].data)\n",
    "        correct1+=solver.test_nets[0].blobs['acc/top-1'].data\n",
    "        correct5+=solver.test_nets[0].blobs['acc/top-5'].data\n",
    "        loss+=solver.test_nets[0].blobs['loss'].data\n",
    "    correct1 = correct1 / test_iterations\n",
    "    correct5 = correct5 / test_iterations\n",
    "    loss=loss/test_epoch\n",
    "    print(\" loss=%.4f, test Top 1 accuracy=%.4f, Top 5 accuracy=%.4f\"%(loss,correct1,correct5))\n",
    "    return loss,correct1,correct5\n",
    "def fit(solver,epoches,train_iterations):\n",
    "    # the main solver loop\n",
    "    history={\"top1\":[],\"top5\":[],\"loss\":[]}\n",
    "    for ep in range(epoches):\n",
    "        for train_it in range(train_iterations):\n",
    "            print \"\\rEpoch %d: Training %.2f%%\"%(ep,(train_it+1.0)*100/train_iterations),\n",
    "            solver.step(1)\n",
    "#         solver.step(int(train_iterations))  # run one epoch\n",
    "        solver.test_nets[0].forward()\n",
    "        history[\"top1\"].append(float(solver.test_nets[0].blobs['acc/top-1'].data))\n",
    "        history[\"top5\"].append(float(solver.test_nets[0].blobs['acc/top-5'].data))\n",
    "        history[\"loss\"].append(float(solver.test_nets[0].blobs['loss'].data))\n",
    "        print(\" loss=%.4f, test Top 1 accuracy=%.4f, Top 5 accuracy=%.4f\"%(history[\"loss\"][ep],history[\"top1\"][ep],history[\"top5\"][ep]))\n",
    "    return history"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-03T08:59:25.510337Z",
     "start_time": "2020-04-03T08:57:39.854551Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0: Training 100.00%  loss=1.6505, test Top 1 accuracy=0.5800, Top 5 accuracy=0.7600\n",
      "Epoch 1: Training 100.00%  loss=1.2609, test Top 1 accuracy=0.6000, Top 5 accuracy=0.9200\n",
      "Epoch 2: Training 100.00%  loss=1.4065, test Top 1 accuracy=0.7000, Top 5 accuracy=0.8400\n",
      "Epoch 3: Training 100.00%  loss=1.2101, test Top 1 accuracy=0.7200, Top 5 accuracy=0.9400\n",
      "Epoch 4: Training 100.00%  loss=1.7304, test Top 1 accuracy=0.6600, Top 5 accuracy=0.8600\n",
      "Epoch 5: Training 100.00%  loss=1.5374, test Top 1 accuracy=0.6000, Top 5 accuracy=0.8800\n",
      "Epoch 6: Training 100.00%  loss=0.8420, test Top 1 accuracy=0.7400, Top 5 accuracy=0.9400\n",
      "Epoch 7: Training 100.00%  loss=1.4797, test Top 1 accuracy=0.6800, Top 5 accuracy=0.8600\n",
      "Epoch 8: Training 100.00%  loss=1.2261, test Top 1 accuracy=0.6400, Top 5 accuracy=0.9000\n",
      "Epoch 9: Training 100.00%  loss=1.3313, test Top 1 accuracy=0.6600, Top 5 accuracy=0.8600\n"
     ]
    }
   ],
   "source": [
    "his=fit(solver,epoches=10,train_iterations=train_iterations)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-03T08:59:27.590713Z",
     "start_time": "2020-04-03T08:59:25.514511Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test 100.00%  loss=14.3659, test Top 1 accuracy=0.6800, Top 5 accuracy=0.8700\n"
     ]
    }
   ],
   "source": [
    "loss,top1,top5=evaluate(solver,test_iterations)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-03T08:59:27.921941Z",
     "start_time": "2020-04-03T08:59:27.597071Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'Test Accuracy: 0.87')"
      ]
     },
     "execution_count": 85,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#plot figure\n",
    "epoches=10\n",
    "_, ax1 = subplots()\n",
    "ax2 = ax1.twinx()\n",
    "ax1.plot(arange(epoches), his['loss'])\n",
    "ax2.plot(arange(epoches), his['top5'], 'r')\n",
    "ax1.set_xlabel('iteration')\n",
    "ax1.set_ylabel('train loss')\n",
    "ax2.set_ylabel('test accuracy')\n",
    "ax2.set_title('Test Accuracy: {:.2f}'.format(top5))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  },
  "latex_envs": {
   "LaTeX_envs_menu_present": true,
   "autoclose": false,
   "autocomplete": true,
   "bibliofile": "biblio.bib",
   "cite_by": "apalike",
   "current_citInitial": 1,
   "eqLabelWithNumbers": true,
   "eqNumInitial": 1,
   "hotkeys": {
    "equation": "Ctrl-E",
    "itemize": "Ctrl-I"
   },
   "labels_anchors": false,
   "latex_user_defs": false,
   "report_style_numbering": false,
   "user_envs_cfg": false
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}