CNN/CNN_CIFAR_10.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "import keras\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from PIL import Image\n",
    "import os\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>frog</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>truck</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>truck</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>deer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>automobile</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         label\n",
       "id            \n",
       "1         frog\n",
       "2        truck\n",
       "3        truck\n",
       "4         deer\n",
       "5   automobile"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels = pd.read_csv('cifar10_Labels.csv', index_col=0)\n",
    "labels.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "automobile\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.PngImagePlugin.PngImageFile image mode=RGB size=32x32 at 0x17CE569E518>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# View an image\n",
    "img_idx = 5\n",
    "print(labels.label[img_idx])\n",
    "Image.open('cifar10/'+str(img_idx)+'.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 174,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Splitting data into Train and Test data\n",
    "from sklearn.model_selection import train_test_split\n",
    "y_train, y_test = train_test_split(labels.label, test_size=0.3, random_state=42)\n",
    "train_idx, test_idx = y_train.index, y_test.index # Stroing indexes for later use"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Reading images for training \n",
    "temp = []\n",
    "# cnt = 1\n",
    "for img_idx in y_train.index:\n",
    "    img_path = os.path.join('cifar10/', str(img_idx) + '.png')\n",
    "    img = np.array(Image.open(img_path)).astype('float32')        \n",
    "    temp.append(img)    \n",
    "#     print(cnt)\n",
    "#     cnt+=1\n",
    "X_train = np.stack(temp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Reading images for testing \n",
    "temp = []\n",
    "# cnt = 1\n",
    "for img_idx in y_test.index:\n",
    "    img_path = os.path.join('cifar10/', str(img_idx) + '.png')\n",
    "    img = np.array(Image.open(img_path)).astype('float32')        \n",
    "    temp.append(img)\n",
    "#     print(cnt)\n",
    "#     cnt+=1\n",
    "X_test = np.stack(temp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Normalizing image data\n",
    "X_train = X_train/255.\n",
    "X_test = X_test/255."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# One-hot encoding 10 output classes\n",
    "encode_X = LabelEncoder()\n",
    "encode_X_fit = encode_X.fit_transform(y_train)\n",
    "y_train = keras.utils.np_utils.to_categorical(encode_X_fit)\n",
    "\n",
    "encode_y = LabelEncoder()\n",
    "encode_y_fit = encode_y.fit_transform(y_test)\n",
    "y_test = keras.utils.np_utils.to_categorical(encode_y_fit)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 320,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Defining CNN network\n",
    "\n",
    "num_classes = 10\n",
    "model = keras.models.Sequential([    \n",
    "    # Adding first convolutional layer\n",
    "    keras.layers.Conv2D(filters=32, kernel_size=(3, 3), strides=1, padding='same', activation='relu', \n",
    "                        kernel_regularizer=keras.regularizers.l2(0.001), input_shape=(32, 32, 3), name='Conv_1'), \n",
    "    # Normalizing the parameters from last layer to speed up the performance (optional)\n",
    "    keras.layers.BatchNormalization(name='BN_1'),\n",
    "    # Adding first pooling layer\n",
    "    keras.layers.MaxPool2D(pool_size=(2, 2), name='MaxPool_1'),\n",
    "    # Adding second convolutional layer\n",
    "    keras.layers.Conv2D(filters=64, kernel_size=(3, 3), strides=1, padding='same', activation='relu', \n",
    "                        kernel_regularizer=keras.regularizers.l2(0.001), name='Conv_2'),    \n",
    "    keras.layers.BatchNormalization(name='BN_2'),\n",
    "    # Adding second pooling layer\n",
    "    keras.layers.MaxPool2D(pool_size=(2, 2), name='MaxPool_2'),\n",
    "    # Flattens the input\n",
    "    keras.layers.Flatten(name='Flat'),\n",
    "    # Fully-Connected layer\n",
    "    keras.layers.Dense(num_classes, activation='softmax', name='pred_layer')    \n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 323,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "Conv_1 (Conv2D)              (None, 32, 32, 32)        896       \n",
      "_________________________________________________________________\n",
      "BN_1 (BatchNormalization)    (None, 32, 32, 32)        128       \n",
      "_________________________________________________________________\n",
      "MaxPool_1 (MaxPooling2D)     (None, 16, 16, 32)        0         \n",
      "_________________________________________________________________\n",
      "Conv_2 (Conv2D)              (None, 16, 16, 64)        18496     \n",
      "_________________________________________________________________\n",
      "BN_2 (BatchNormalization)    (None, 16, 16, 64)        256       \n",
      "_________________________________________________________________\n",
      "MaxPool_2 (MaxPooling2D)     (None, 8, 8, 64)          0         \n",
      "_________________________________________________________________\n",
      "Flat (Flatten)               (None, 4096)              0         \n",
      "_________________________________________________________________\n",
      "pred_layer (Dense)           (None, 10)                40970     \n",
      "=================================================================\n",
      "Total params: 60,746\n",
      "Trainable params: 60,554\n",
      "Non-trainable params: 192\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 324,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss='categorical_crossentropy', \n",
    "              optimizer=keras.optimizers.Adam(), \n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 28000 samples, validate on 7000 samples\n",
      "Epoch 1/5\n",
      "28000/28000 [==============================] - 379s 14ms/step - loss: 1.7326 - acc: 0.4736 - val_loss: 1.7605 - val_acc: 0.4581\n",
      "\n",
      "Epoch 00001: val_acc improved from -inf to 0.45814, saving model to CIFAR10_checkpoint.hdf5\n",
      "Epoch 2/5\n",
      "28000/28000 [==============================] - 351s 13ms/step - loss: 1.2379 - acc: 0.6019 - val_loss: 1.7944 - val_acc: 0.4793\n",
      "\n",
      "Epoch 00002: val_acc improved from 0.45814 to 0.47929, saving model to CIFAR10_checkpoint.hdf5\n",
      "Epoch 3/5\n",
      "28000/28000 [==============================] - 353s 13ms/step - loss: 1.0563 - acc: 0.6564 - val_loss: 1.2411 - val_acc: 0.6099\n",
      "\n",
      "Epoch 00003: val_acc improved from 0.47929 to 0.60986, saving model to CIFAR10_checkpoint.hdf5\n",
      "Epoch 4/5\n",
      "28000/28000 [==============================] - 379s 14ms/step - loss: 0.9514 - acc: 0.6893 - val_loss: 1.3454 - val_acc: 0.5811\n",
      "\n",
      "Epoch 00004: val_acc did not improve\n",
      "Epoch 5/5\n",
      "28000/28000 [==============================] - 392s 14ms/step - loss: 0.8666 - acc: 0.7163 - val_loss: 1.2015 - val_acc: 0.6327\n",
      "\n",
      "Epoch 00005: val_acc improved from 0.60986 to 0.63271, saving model to CIFAR10_checkpoint.hdf5\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x17c9b07fe48>"
      ]
     },
     "execution_count": 102,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cpfile = r'CIFAR10_checkpoint.hdf5' # Weights to be stored in HDF5 format\n",
    "cb_checkpoint = keras.callbacks.ModelCheckpoint(cpfile, monitor='val_acc', verbose=1, save_best_only=True, mode='max')\n",
    "epochs = 5\n",
    "model.fit(X_train, y_train, epochs=epochs, validation_split=0.2, callbacks=[cb_checkpoint])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 234,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>predicted</th>\n",
       "      <th>actual</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>truck</td>\n",
       "      <td>horse</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>ship</td>\n",
       "      <td>ship</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>ship</td>\n",
       "      <td>airplane</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>frog</td>\n",
       "      <td>frog</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>automobile</td>\n",
       "      <td>automobile</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>frog</td>\n",
       "      <td>frog</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>ship</td>\n",
       "      <td>ship</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>airplane</td>\n",
       "      <td>airplane</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>frog</td>\n",
       "      <td>frog</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>ship</td>\n",
       "      <td>dog</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    predicted      actual\n",
       "0       truck       horse\n",
       "1        ship        ship\n",
       "2        ship    airplane\n",
       "3        frog        frog\n",
       "4  automobile  automobile\n",
       "5        frog        frog\n",
       "6        ship        ship\n",
       "7    airplane    airplane\n",
       "8        frog        frog\n",
       "9        ship         dog"
      ]
     },
     "execution_count": 234,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# << DeprecationWarning: The truth value of an empty array is ambiguous >> can arise due to a NumPy version higher than 1.13.3.\n",
    "# The issue will be updated in upcoming version.\n",
    "pred = encode_y.inverse_transform(model.predict_classes(X_test[:10])) \n",
    "act = encode_y.inverse_transform([np.argmax(x) for x in y_test[:10]])\n",
    "\n",
    "res = pd.DataFrame([pred, act]).T\n",
    "res.columns = ['predicted', 'actual']\n",
    "res"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CNN Conv and MaxPool Layers' filters visualization\n",
    "- Provide an absolute path for both `temp_folder` and `input_folder` \n",
    "- Make sure to put test images inside `img` folder\n",
    "- Make sure to create an empty `tmp` folder\n",
    "- The visualization will start on `http://localhost:5000`\n",
    "- Proceed by clicking each layer sequentially, halt until the images for a clicked layer are generated/saved in `tmp` folder.\n",
    "- Interrupt the kernel to stop the execution."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "from quiver_engine import server # pip install quiver_engine | for CNN layer's filters visualization.\n",
    "server.launch(model, temp_folder='/Desktop/tmp', \n",
    "              input_folder='/Desktop/img') "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CNN Dense layer visualization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 285,
   "metadata": {},
   "outputs": [],
   "source": [
    "from vis.visualization import visualize_saliency, visualize_cam, overlay\n",
    "from vis.utils import utils"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 244,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog',\n",
       "       'horse', 'ship', 'truck'], dtype=object)"
      ]
     },
     "execution_count": 244,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Indexes of categories for our model\n",
    "classes = encode_y.inverse_transform(np.arange(10))\n",
    "classes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "From `res` dataframe, pick an index where `predicted` class is same as `actual` class.\n",
    "Say index `6` for `ship`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 282,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Fetching the ship image\n",
    "ship_img = utils.load_img('cifar10/'+str(test_idx[6])+'.png') # can use Image.open() also.\n",
    "plt.imshow(ship_img)\n",
    "plt.title('Ship image')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 278,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Predicting the probability for each of the class\n",
    "ship_prob = model.predict(X_test[6:7]).ravel().copy()\n",
    "pd.Series(ship_prob, index=classes).plot.barh()\n",
    "plt.title('Predicted probability of actual ship image by Conv model')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To visualize activation over final dense layer outputs, we need to switch the softmax activation out for linear since gradient of output node will depend on all the other node activations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 327,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Utility to search for layer index by name. \n",
    "layer_idx = utils.find_layer_idx(model, 'pred_layer')\n",
    "\n",
    "# Swap softmax with linear\n",
    "model.layers[layer_idx].activation = keras.activations.linear\n",
    "model = utils.apply_modifications(model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1. Saliency map\n",
    "Saliency maps clarifies which part does our model focuses on to get prediction."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 284,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x432 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(12,6))\n",
    "for i in range(len(classes)):\n",
    "    plt.subplot(2, 5, i + 1)\n",
    "    grads = visualize_saliency(model, layer_idx, filter_indices=i, seed_input=ship_img, backprop_modifier='guided')\n",
    "    plt.xticks([])\n",
    "    plt.yticks([])\n",
    "    plt.xlabel(classes[i])\n",
    "    plt.title('p=' + str(np.round(ship_prob[i], 4)))\n",
    "    plt.imshow(grads, cmap='jet')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As shown in probability graph only first two and last two nodes are active, leaving center 6 almost dead. The second last node meant for `ship` has the highest probability and as seen in above plot only emphasize on region of ship. Other 3 active nodes includes noise and thus reduced probability."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2. Class activation maps or Grad-cam\n",
    "These contain more detail since they use Conv or Pooling features that contain more spatial detail which is lost in Dense layers. The only additional detail compared to saliency is the `penultimate_layer_idx`. This specifies the pre-layer whose gradients should be used."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 328,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x432 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(12,6))\n",
    "for i in range(len(classes)):\n",
    "    plt.subplot(2, 5, i + 1)\n",
    "    cam_grads = visualize_cam(model, layer_idx, filter_indices=i, seed_input=ship_img, backprop_modifier='guided',\n",
    "                             penultimate_layer_idx=utils.find_layer_idx(model, 'BN_2'))# batch_normalization_14\n",
    "    plt.xticks([])\n",
    "    plt.yticks([])\n",
    "    plt.xlabel(classes[i])\n",
    "    plt.title('p=' + str(np.round(ship_prob[i], 4)))\n",
    "    plt.imshow(overlay(cam_grads, ship_img, alpha=0.3))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The ninth subplot constitute of a contour plot overlayed over original image which captures ship structure better than other three countour plots which are heavily affected by noise.  \n",
    "Center six nodes are almost dead so no countour plot."
   ]
  }
 ],
 "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}