cnn.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The autoreload extension is already loaded. To reload it, use:\n",
      "  %reload_ext autoreload\n"
     ]
    }
   ],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "import numpy as np\n",
    "\n",
    "from datasets import DataCifar10\n",
    "from models import ThreeLayerConvNet\n",
    "from models.layers import Conv, ConvNaive, MaxPool, MaxPoolNaive\n",
    "from optimers import OptimerAdam, OptimerSGD\n",
    "from utils import check_rel_error, check_gradient, check_accuracy, show_training_info"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training data shape:  (49000, 3, 32, 32)\n",
      "Training labels shape:  (49000,)\n",
      "Validating data shape:  (100, 3, 32, 32)\n",
      "Validating labels shape:  (100,)\n",
      "Testing data shape:  (1000, 3, 32, 32)\n",
      "Testing labels shape:  (1000,)\n"
     ]
    }
   ],
   "source": [
    "dataloader = DataCifar10('./datasets/cifar-10-batches-py', \n",
    "                         num_train=49000, num_test=1000, num_val=100)\n",
    "\n",
    "print('Training data shape: ', dataloader.x_train.shape)\n",
    "print('Training labels shape: ', dataloader.y_train.shape)\n",
    "print('Validating data shape: ', dataloader.x_val.shape)\n",
    "print('Validating labels shape: ', dataloader.y_val.shape)\n",
    "print('Testing data shape: ', dataloader.x_test.shape)\n",
    "print('Testing labels shape: ', dataloader.y_test.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sanity Check"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  Image processing via convolutions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# get a image\n",
    "images, labels = dataloader.get_batch(1)\n",
    "\n",
    "# identical filter\n",
    "conv = Conv(3, 3, 3, 3, S=1, P=1)\n",
    "b = np.zeros(3)\n",
    "W = np.zeros([3, 3, 3, 3])\n",
    "W[0, 0, :, :] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]\n",
    "W[1, 1, :, :] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]\n",
    "W[2, 2, :, :] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]\n",
    "\n",
    "images_i, _ = conv.forward(images, {'W': W, 'b': b})\n",
    "\n",
    "# grey filter\n",
    "conv = Conv(3, 1, 3, 3, S=1, P=1)\n",
    "b = np.zeros(1)\n",
    "W = np.zeros([1, 3, 3, 3])\n",
    "W[0, 0, :, :] = [[0, 0, 0], [0, 0.3, 0], [0, 0, 0]]\n",
    "W[0, 1, :, :] = [[0, 0, 0], [0, 0.6, 0], [0, 0, 0]]\n",
    "W[0, 2, :, :] = [[0, 0, 0], [0, 0.1, 0], [0, 0, 0]]\n",
    "\n",
    "images_g, _ = conv.forward(images, {'W': W, 'b': b})\n",
    "\n",
    "# other filter\n",
    "conv = Conv(3, 3, 3, 3, S=1, P=1)\n",
    "b = np.zeros(3)\n",
    "W = np.zeros([3, 3, 3, 3])\n",
    "W[0, 0, :, :] = [[0, 0.2, 0], [0.2, 0.2, 0.2], [0, 0.2, 0]]\n",
    "W[1, 1, :, :] = [[0, 0.2, 0], [0.2, 0.2, 0.2], [0, 0.2, 0]]\n",
    "W[2, 2, :, :] = [[0, 0.2, 0], [0.2, 0.2, 0.2], [0, 0.2, 0]]\n",
    "\n",
    "images_u, _ = conv.forward(images, {'W': W, 'b': b})\n",
    "\n",
    "# show image\n",
    "dataloader.show_by_data(images_i[0], labels[0])\n",
    "dataloader.show_by_data(images_g[0], labels[0])\n",
    "dataloader.show_by_data(images_u[0], labels[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Compare img2col/col2img with naive implementation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Relative error of conv layer: 5.8435187450990335e-08\n",
      "Relative error of max pooling layer: 0.0\n"
     ]
    }
   ],
   "source": [
    "# get a image\n",
    "images, labels = dataloader.get_batch(1)\n",
    "\n",
    "# for conv layer\n",
    "conv = Conv(3, 3, 3, 3, S=1, P=1)\n",
    "conv_naive = ConvNaive(3, 3, 3, 3, S=1, P=1)\n",
    "param = conv.get_init_param()\n",
    "\n",
    "result, _ = conv.forward(images, param)\n",
    "result_naive, _ = conv_naive.forward(images, param)\n",
    "\n",
    "print('Relative error of conv layer:', check_rel_error(result , result_naive))\n",
    "\n",
    "# for max pooling layer\n",
    "pool = MaxPool(2, 2, S=2)\n",
    "pool_naive = MaxPoolNaive(2, 2, S=2)\n",
    "\n",
    "result, _ = pool.forward(images, {})\n",
    "result_naive, _ = pool_naive.forward(images, {})\n",
    "\n",
    "print('Relative error of max pooling layer:', check_rel_error(result , result_naive))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Gradient check"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Gradient check result:\n",
      "Layer | Key | Numerical gradient | Calculated gradient | Relative error\n",
      "<class 'models.layers.Conv'> W 0.030910036796782944 0.030910036822071135 8.181223411503938e-10\n",
      "<class 'models.layers.Conv'> b 0.0011852527181943628 0.0011852527333532858 1.2789612388472066e-08\n",
      "<class 'models.layers.Linear'> W 0.00016849097672633204 0.00016849099097202472 8.454869425909958e-08\n",
      "<class 'models.layers.Linear'> b -0.0012017592476709638 -0.001201759233535293 1.1762481400564651e-08\n",
      "<class 'models.layers.Linear'> W 0.03185399519711751 0.03185399522069988 7.403267650603517e-10\n",
      "<class 'models.layers.Linear'> b -0.012401299831488187 -0.012401299835136975 2.942262438707505e-10\n"
     ]
    }
   ],
   "source": [
    "def get_init_model(hyperparams):\n",
    "    return ThreeLayerConvNet((3, 32, 32), 10, hyperparams=hyperparams)\n",
    "\n",
    "# init model\n",
    "model = get_init_model({'init_scale': 1e-2, 'num_filter': 3, 'num_hidden': 5})\n",
    "\n",
    "# gradient check\n",
    "print('\\nGradient check result:')\n",
    "check_gradient(model, dataloader.x_train[0: 10, :], dataloader.y_train[0: 10])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Overfit small dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 / 100 loss is 2.386580730421604\n",
      "10 / 100 loss is 1.6291386191460113\n",
      "20 / 100 loss is 0.6072363212522237\n",
      "30 / 100 loss is 0.09134781434109701\n",
      "40 / 100 loss is 0.030913320706545126\n",
      "50 / 100 loss is 0.007406126816653422\n",
      "60 / 100 loss is 0.0027921466034255098\n",
      "70 / 100 loss is 0.0024747101097354032\n",
      "80 / 100 loss is 0.0016293615051948237\n",
      "90 / 100 loss is 0.0013996051855285935\n"
     ]
    }
   ],
   "source": [
    "# load small dataset\n",
    "smallloader = DataCifar10('./datasets/cifar-10-batches-py', \n",
    "                          num_train=100, num_test=100, num_val=100)\n",
    "\n",
    "# init model\n",
    "model = get_init_model({'reg': 0., \n",
    "                        'filter_size': 7,\n",
    "                        'num_filter': 32, \n",
    "                        'num_hidden': 100, \n",
    "                        'init_scale': 1e-2})\n",
    "\n",
    "optimer = OptimerAdam({'learn_rate': 1e-3, \n",
    "                       'num_iters': 100, \n",
    "                       'batch_size': 50}, print_every=10)\n",
    "\n",
    "# train model\n",
    "optimer.train(model, smallloader)\n",
    "\n",
    "# save model\n",
    "model.save('./saves/ThreeLayerConvNet/model_small.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# plot loss curve\n",
    "show_training_info(optimer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The accuracy on testing dataset is  0.16\n"
     ]
    }
   ],
   "source": [
    "# load model\n",
    "model = ThreeLayerConvNet.load('./saves/ThreeLayerConvNet/model_small.pkl')\n",
    "\n",
    "# check accuracy\n",
    "scores = model.predict(smallloader.x_test)\n",
    "accuracy = check_accuracy(scores, smallloader.y_test)\n",
    "    \n",
    "print('The accuracy on testing dataset is ', accuracy)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train 3-layer convnet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 / 10000 loss is 2.3232238405906163\n",
      "100 / 10000 loss is 2.170192833534839\n",
      "200 / 10000 loss is 1.9385105533940545\n",
      "300 / 10000 loss is 1.7486496421394757\n",
      "400 / 10000 loss is 1.6488785776656472\n",
      "500 / 10000 loss is 1.7984916521006307\n",
      "600 / 10000 loss is 1.5403184437090798\n",
      "700 / 10000 loss is 1.7109441691119245\n",
      "800 / 10000 loss is 1.2686219238978438\n",
      "900 / 10000 loss is 1.3187309394840745\n",
      "1000 / 10000 loss is 1.4525718986914051\n",
      "1100 / 10000 loss is 1.4146489262832183\n",
      "1200 / 10000 loss is 1.4539469248075625\n",
      "1300 / 10000 loss is 1.2641733618831448\n",
      "1400 / 10000 loss is 1.397223603896143\n",
      "1500 / 10000 loss is 1.268911952285924\n",
      "1600 / 10000 loss is 1.3716131834907146\n",
      "1700 / 10000 loss is 1.248670387844385\n",
      "1800 / 10000 loss is 1.4614190063958266\n",
      "1900 / 10000 loss is 1.1177854883926073\n",
      "2000 / 10000 loss is 1.385941425493822\n",
      "2100 / 10000 loss is 1.3280310187578162\n",
      "2200 / 10000 loss is 1.270276856258146\n",
      "2300 / 10000 loss is 1.3617826091604406\n",
      "2400 / 10000 loss is 1.2192145974404156\n",
      "2500 / 10000 loss is 1.428426314576365\n",
      "2600 / 10000 loss is 1.0173565267007447\n",
      "2700 / 10000 loss is 1.2752092840047464\n",
      "2800 / 10000 loss is 1.1574421612856027\n",
      "2900 / 10000 loss is 1.4020529035819609\n",
      "3000 / 10000 loss is 1.0233220880956984\n",
      "3100 / 10000 loss is 1.2099366019636237\n",
      "3200 / 10000 loss is 1.2585606940348204\n",
      "3300 / 10000 loss is 1.3272597744951327\n",
      "3400 / 10000 loss is 1.3385020703340416\n",
      "3500 / 10000 loss is 1.1009389538230272\n",
      "3600 / 10000 loss is 1.0388832797463994\n",
      "3700 / 10000 loss is 1.0972588744264966\n",
      "3800 / 10000 loss is 1.1997413613417334\n",
      "3900 / 10000 loss is 0.9461905127431638\n",
      "4000 / 10000 loss is 1.075957172204839\n",
      "4100 / 10000 loss is 1.2219089550329576\n",
      "4200 / 10000 loss is 1.0842582845526905\n",
      "4300 / 10000 loss is 1.2116269804935011\n",
      "4400 / 10000 loss is 0.9143120163761485\n",
      "4500 / 10000 loss is 1.085234367627073\n",
      "4600 / 10000 loss is 0.9530204054254487\n",
      "4700 / 10000 loss is 1.0814711840792572\n",
      "4800 / 10000 loss is 0.9631827790487303\n",
      "4900 / 10000 loss is 1.0045639222879448\n",
      "5000 / 10000 loss is 0.9775715286581238\n",
      "5100 / 10000 loss is 0.9337689812361349\n",
      "5200 / 10000 loss is 1.095669901085911\n",
      "5300 / 10000 loss is 1.042799871457335\n",
      "5400 / 10000 loss is 1.0358530219226731\n",
      "5500 / 10000 loss is 0.8536487588665789\n",
      "5600 / 10000 loss is 0.9834955448155732\n",
      "5700 / 10000 loss is 1.1430346545096597\n",
      "5800 / 10000 loss is 1.0330883457572249\n",
      "5900 / 10000 loss is 0.9511492659997428\n",
      "6000 / 10000 loss is 1.1638526769492679\n",
      "6100 / 10000 loss is 0.98776352647831\n",
      "6200 / 10000 loss is 0.8694765586610925\n",
      "6300 / 10000 loss is 0.8531148722408739\n",
      "6400 / 10000 loss is 0.6627334178870583\n",
      "6500 / 10000 loss is 0.7523162899517383\n",
      "6600 / 10000 loss is 0.9984316105759171\n",
      "6700 / 10000 loss is 1.0004484216653013\n",
      "6800 / 10000 loss is 0.9289660347683916\n",
      "6900 / 10000 loss is 0.9837474123422545\n",
      "7000 / 10000 loss is 0.8646790769080799\n",
      "7100 / 10000 loss is 0.9176647692217416\n",
      "7200 / 10000 loss is 0.9442091381522822\n",
      "7300 / 10000 loss is 1.0645131149956257\n",
      "7400 / 10000 loss is 0.8199347434806015\n",
      "7500 / 10000 loss is 0.965440312527952\n",
      "7600 / 10000 loss is 0.70564275682124\n",
      "7700 / 10000 loss is 0.9362680123938143\n",
      "7800 / 10000 loss is 0.8574051833315731\n",
      "7900 / 10000 loss is 0.7247306585325259\n",
      "8000 / 10000 loss is 0.8737448529634201\n",
      "8100 / 10000 loss is 0.7990862194128355\n",
      "8200 / 10000 loss is 0.8489454203984906\n",
      "8300 / 10000 loss is 0.9398196747773908\n",
      "8400 / 10000 loss is 0.9060989562494537\n",
      "8500 / 10000 loss is 0.7579559309445872\n",
      "8600 / 10000 loss is 0.7931839377746789\n",
      "8700 / 10000 loss is 0.6812163769519107\n",
      "8800 / 10000 loss is 0.8739644307983748\n",
      "8900 / 10000 loss is 0.9999336198547188\n",
      "9000 / 10000 loss is 1.0025109103609424\n",
      "9100 / 10000 loss is 0.9636960393289791\n",
      "9200 / 10000 loss is 0.7954080540114866\n",
      "9300 / 10000 loss is 0.9162885742178003\n",
      "9400 / 10000 loss is 0.7791641090928806\n",
      "9500 / 10000 loss is 0.6707933868743778\n",
      "9600 / 10000 loss is 0.676158155103925\n",
      "9700 / 10000 loss is 0.6821096141335452\n",
      "9800 / 10000 loss is 0.6765953817115531\n",
      "9900 / 10000 loss is 0.9919169310542643\n"
     ]
    }
   ],
   "source": [
    "# init model\n",
    "model = get_init_model({'reg': 0.01, 'init_scale': 1e-3,\n",
    "                        'filter_size': 7, 'num_filter': 32, 'num_hidden': 500})\n",
    "\n",
    "optimer = OptimerAdam({'learn_rate': 1e-4, 'learn_rate_decay':0.8,\n",
    "                       'num_iters': 10000, 'batch_size': 64}, print_every=100)\n",
    "\n",
    "# train model\n",
    "optimer.train(model, dataloader)\n",
    "\n",
    "# save model\n",
    "model.save('./saves/ThreeLayerConvNet/model.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# plot loss curve\n",
    "show_training_info(optimer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The accuracy on testing dataset is  0.628\n"
     ]
    }
   ],
   "source": [
    "# load model\n",
    "model = ThreeLayerConvNet.load('./saves/ThreeLayerConvNet/model.pkl')\n",
    "\n",
    "# check accuracy\n",
    "scores = model.predict(dataloader.x_test)\n",
    "accuracy = check_accuracy(scores, dataloader.y_test)\n",
    "    \n",
    "print('The accuracy on testing dataset is ', accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}