cnn_landmarks_hog.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "import cv2\n",
    "import dlib\n",
    "import itertools\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import keras\n",
    "from keras.callbacks import ModelCheckpoint\n",
    "from keras.layers import Activation, BatchNormalization, Conv2D, Dense, Dropout, Flatten, Input, MaxPool2D, Reshape\n",
    "from keras.models import Model, Sequential\n",
    "from keras.utils import np_utils\n",
    "from skimage.feature import hog\n",
    "from sklearn.metrics import confusion_matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_csv(\"fer2013.csv\")\n",
    "emotion_labels = {0: \"Angry\", 1: \"Disgust\", 2:\"Fear\", 3:\"Happy\", 4:\"Sad\", 5:\"Surprise\", 6:\"Neutral\"}\n",
    "labels = [\"Angry\", \"Disgust\", \"Fear\", \"Happy\", \"Sad\", \"Surprise\", \"Neutral\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": true
   },
   "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>emotion</th>\n",
       "      <th>pixels</th>\n",
       "      <th>Usage</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>70 80 82 72 58 58 60 63 54 58 60 48 89 115 121...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>151 150 147 155 148 133 111 140 170 174 182 15...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>231 212 156 164 174 138 161 173 182 200 106 38...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>24 32 36 30 32 23 19 20 30 41 21 22 32 34 21 1...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>6</td>\n",
       "      <td>4 0 0 0 0 0 0 0 0 0 0 0 3 15 23 28 48 50 58 84...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   emotion                                             pixels     Usage\n",
       "0        0  70 80 82 72 58 58 60 63 54 58 60 48 89 115 121...  Training\n",
       "1        0  151 150 147 155 148 133 111 140 170 174 182 15...  Training\n",
       "2        2  231 212 156 164 174 138 161 173 182 200 106 38...  Training\n",
       "3        4  24 32 36 30 32 23 19 20 30 41 21 22 32 34 21 1...  Training\n",
       "4        6  4 0 0 0 0 0 0 0 0 0 0 0 3 15 23 28 48 50 58 84...  Training"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Training       28709\n",
       "PrivateTest     3589\n",
       "PublicTest      3589\n",
       "Name: Usage, dtype: int64"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[\"Usage\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def process_data(main_df, use):\n",
    "    subset_df = main_df[main_df[\"Usage\"] == use]\n",
    "    subset_df.drop(columns=['Usage'], inplace=True)\n",
    "    subset_df['pixels'] = subset_df['pixels'].apply(lambda pixel_str: np.fromstring(pixel_str, sep=' '))\n",
    "    x = np.vstack(subset_df['pixels'].values)\n",
    "    x = x.reshape(-1, 48, 48, 1)\n",
    "    y = subset_df['emotion'].values\n",
    "    y = np_utils.to_categorical(y)\n",
    "    \n",
    "    return x, y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# takes the 'x' output of process_data as input\n",
    "def extract_hog_features(images_array):\n",
    "    hog_features_res = []\n",
    "    for img in images_array:\n",
    "        hog_feature = hog(img, orientations=8, pixels_per_cell=(16,16), cells_per_block=(1,1))\n",
    "        hog_features_res.append(hog_feature)\n",
    "    # print(len(hog_features_res), len(hog_features_res[0]))\n",
    "    return np.asarray(hog_features_res)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_landmark_features(images_array):\n",
    "    face_landmarks_res = []\n",
    "    # load detector and predictors from dlib\n",
    "    predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')\n",
    "    \n",
    "    # since our images are already cropped to contain only the face, we don't need to use a detector \n",
    "    rect = [dlib.rectangle(left=1, top=1, right=47, bottom=47)]\n",
    "        \n",
    "    # detect face using openCV i.e cv2\n",
    "    for img in images_array:\n",
    "        cv2.imwrite('temp.jpg', img)\n",
    "        saved_img = cv2.imread('temp.jpg', 0)\n",
    "        face_landmarks = predictor( saved_img, rect[0])\n",
    "        face_landmarks_pixels = np.array([[point.x, point.y] for point in face_landmarks.parts()])\n",
    "        # print(face_landmarks)\n",
    "        face_landmarks_res.append(face_landmarks_pixels)\n",
    "    face_landmarks_res = [res.flatten() for res in face_landmarks_res]\n",
    "    return np.asarray(face_landmarks_res)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/ritvikareddy 1/Documents/GitHub/ml-project/venv/lib/python3.6/site-packages/pandas/core/frame.py:3697: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  errors=errors)\n",
      "/Users/ritvikareddy 1/Documents/GitHub/ml-project/venv/lib/python3.6/site-packages/ipykernel_launcher.py:4: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  after removing the cwd from sys.path.\n"
     ]
    }
   ],
   "source": [
    "# process the training, validation and test data\n",
    "x_train, y_train = process_data(df, 'Training')\n",
    "x_validation, y_validation = process_data(df, 'PublicTest')\n",
    "x_test, y_test = process_data(df, \"PrivateTest\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/ritvikareddy 1/Documents/GitHub/ml-project/venv/lib/python3.6/site-packages/skimage/feature/_hog.py:150: skimage_deprecation: Default value of `block_norm`==`L1` is deprecated and will be changed to `L2-Hys` in v0.15. To supress this message specify explicitly the normalization method.\n",
      "  skimage_deprecation)\n"
     ]
    }
   ],
   "source": [
    "# get HoG features\n",
    "x_train_hog = extract_hog_features(x_train)\n",
    "x_validation_hog = extract_hog_features(x_validation)\n",
    "x_test_hog = extract_hog_features(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get face landmarks\n",
    "x_train_landmarks = extract_landmark_features(x_train)\n",
    "x_validation_landmarks = extract_landmark_features(x_validation)\n",
    "x_test_landmarks = extract_landmark_features(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train_features = np.concatenate((x_train_landmarks, x_train_hog), axis=1)\n",
    "x_validation_features = np.concatenate((x_validation_landmarks, x_validation_hog), axis=1)\n",
    "x_test_features = np.concatenate((x_test_landmarks, x_test_hog), axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((28709, 48, 48, 1), (28709, 72), (28709, 136), (28709, 208))"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_train.shape, x_train_hog.shape, x_train_landmarks.shape, x_train_features.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 48, 48, 1)    0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_6 (Conv2D)               (None, 46, 46, 64)   640         input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_7 (BatchNor (None, 46, 46, 64)   256         conv2d_6[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_7 (Conv2D)               (None, 44, 44, 64)   36928       batch_normalization_7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_8 (BatchNor (None, 44, 44, 64)   256         conv2d_7[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2D)  (None, 22, 22, 64)   0           batch_normalization_8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "dropout_4 (Dropout)             (None, 22, 22, 64)   0           max_pooling2d_3[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_8 (Conv2D)               (None, 20, 20, 32)   18464       dropout_4[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_9 (BatchNor (None, 20, 20, 32)   128         conv2d_8[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_9 (Conv2D)               (None, 18, 18, 32)   9248        batch_normalization_9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_10 (BatchNo (None, 18, 18, 32)   128         conv2d_9[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_10 (Conv2D)              (None, 16, 16, 32)   9248        batch_normalization_10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_11 (BatchNo (None, 16, 16, 32)   128         conv2d_10[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2D)  (None, 8, 8, 32)     0           batch_normalization_11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_5 (Dropout)             (None, 8, 8, 32)     0           max_pooling2d_4[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "input_2 (InputLayer)            (None, 208)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "flatten_2 (Flatten)             (None, 2048)         0           dropout_5[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_4 (Dense)                 (None, 1024)         214016      input_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_3 (Dense)                 (None, 128)          262272      flatten_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_13 (BatchNo (None, 1024)         4096        dense_4[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_12 (BatchNo (None, 128)          512         dense_3[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_5 (Dense)                 (None, 128)          131200      batch_normalization_13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_6 (Dropout)             (None, 128)          0           batch_normalization_12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_14 (BatchNo (None, 128)          512         dense_5[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, 256)          0           dropout_6[0][0]                  \n",
      "                                                                 batch_normalization_14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dense_6 (Dense)                 (None, 7)            1799        concatenate_1[0][0]              \n",
      "==================================================================================================\n",
      "Total params: 689,831\n",
      "Trainable params: 686,823\n",
      "Non-trainable params: 3,008\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# cnn for just the raw pixels\n",
    "images_nn = Input(shape=(48,48,1))\n",
    "images_conv1 = Conv2D(64, 3, kernel_initializer='he_normal', activation='relu')(images_nn)\n",
    "images_batchnorm1 = BatchNormalization()(images_conv1)\n",
    "\n",
    "images_conv2 = Conv2D(64, 3, activation='relu')(images_batchnorm1)\n",
    "images_batchnorm2 = BatchNormalization()(images_conv2)\n",
    "images_pool1 = MaxPool2D(pool_size=(2,2), strides=2)(images_batchnorm2)\n",
    "images_dropout1 = Dropout(0.6)(images_pool1)\n",
    "\n",
    "images_conv3 = Conv2D(32, 3, activation='relu')(images_dropout1)\n",
    "images_batchnorm3 = BatchNormalization()(images_conv3)\n",
    "\n",
    "images_conv4 = Conv2D(32, 3, activation='relu')(images_batchnorm3)\n",
    "images_batchnorm4 = BatchNormalization()(images_conv4)\n",
    "\n",
    "images_conv5 = Conv2D(32, 3, activation='relu')(images_batchnorm4)\n",
    "images_batchnorm5 = BatchNormalization()(images_conv5)\n",
    "images_pool2 = MaxPool2D(pool_size=(2,2), strides=2)(images_batchnorm5)\n",
    "images_dropout2 = Dropout(0.6)(images_pool2)\n",
    "\n",
    "images_flat1 = Flatten()(images_dropout2)\n",
    "images_dense1 = Dense(128, activation='relu')(images_flat1)\n",
    "images_batchnorm6 = BatchNormalization()(images_dense1)\n",
    "images_dropout3 = Dropout(0.6)(images_batchnorm6)\n",
    "\n",
    "# cnn for hog + landmark features\n",
    "# input shape will be 72 (from hog) + 68*2 (from landamrks)\n",
    "extra_features_nn = Input(shape=(208,))\n",
    "extra_features_dense1 = Dense(1024, activation='relu')(extra_features_nn)\n",
    "extra_features_batchnorm1 = BatchNormalization()(extra_features_dense1)\n",
    "\n",
    "extra_features_dense2 = Dense(128, activation='relu') (extra_features_batchnorm1)\n",
    "extra_features_batchnorm2 = BatchNormalization()(extra_features_dense2)\n",
    "\n",
    "# merge final outputs from both networks\n",
    "merged_net = keras.layers.merge.concatenate([images_dropout3, extra_features_batchnorm2])\n",
    "\n",
    "output_layer = Dense(7, activation='softmax')(merged_net)\n",
    "\n",
    "model = Model(inputs=[images_nn, extra_features_nn], outputs=output_layer)\n",
    "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 28709 samples, validate on 3589 samples\n",
      "Epoch 1/12\n",
      " - 626s - loss: 1.7676 - acc: 0.3920 - val_loss: 2.5206 - val_acc: 0.3084\n",
      "\n",
      "Epoch 00001: val_loss improved from inf to 2.52059, saving model to cnn_feats_3\n",
      "Epoch 2/12\n",
      " - 718s - loss: 1.4323 - acc: 0.4608 - val_loss: 2.4292 - val_acc: 0.2154\n",
      "\n",
      "Epoch 00002: val_loss improved from 2.52059 to 2.42921, saving model to cnn_feats_3\n",
      "Epoch 3/12\n",
      " - 643s - loss: 1.3595 - acc: 0.4859 - val_loss: 2.0938 - val_acc: 0.2583\n",
      "\n",
      "Epoch 00003: val_loss improved from 2.42921 to 2.09381, saving model to cnn_feats_3\n",
      "Epoch 4/12\n",
      " - 763s - loss: 1.3296 - acc: 0.4932 - val_loss: 2.1358 - val_acc: 0.3302\n",
      "\n",
      "Epoch 00004: val_loss did not improve from 2.09381\n",
      "Epoch 5/12\n",
      " - 770s - loss: 1.3110 - acc: 0.5013 - val_loss: 1.7190 - val_acc: 0.3279\n",
      "\n",
      "Epoch 00005: val_loss improved from 2.09381 to 1.71895, saving model to cnn_feats_3\n",
      "Epoch 6/12\n",
      " - 804s - loss: 1.2995 - acc: 0.5066 - val_loss: 3.3469 - val_acc: 0.1315\n",
      "\n",
      "Epoch 00006: val_loss did not improve from 1.71895\n",
      "Epoch 7/12\n",
      " - 936s - loss: 1.2893 - acc: 0.5105 - val_loss: 1.6727 - val_acc: 0.3865\n",
      "\n",
      "Epoch 00007: val_loss improved from 1.71895 to 1.67266, saving model to cnn_feats_3\n",
      "Epoch 8/12\n",
      " - 1004s - loss: 1.2716 - acc: 0.5154 - val_loss: 1.7469 - val_acc: 0.3569\n",
      "\n",
      "Epoch 00008: val_loss did not improve from 1.67266\n",
      "Epoch 9/12\n",
      " - 886s - loss: 1.2557 - acc: 0.5227 - val_loss: 2.0873 - val_acc: 0.3199\n",
      "\n",
      "Epoch 00009: val_loss did not improve from 1.67266\n",
      "Epoch 10/12\n",
      " - 987s - loss: 1.2459 - acc: 0.5284 - val_loss: 2.1568 - val_acc: 0.2686\n",
      "\n",
      "Epoch 00010: val_loss did not improve from 1.67266\n",
      "Epoch 11/12\n",
      " - 951s - loss: 1.2283 - acc: 0.5313 - val_loss: 1.4960 - val_acc: 0.4436\n",
      "\n",
      "Epoch 00011: val_loss improved from 1.67266 to 1.49604, saving model to cnn_feats_3\n",
      "Epoch 12/12\n",
      " - 739s - loss: 1.2240 - acc: 0.5338 - val_loss: 1.4935 - val_acc: 0.4514\n",
      "\n",
      "Epoch 00012: val_loss improved from 1.49604 to 1.49349, saving model to cnn_feats_3\n"
     ]
    }
   ],
   "source": [
    "# save best weights\n",
    "checkpoint = ModelCheckpoint(filepath='cnn_feats_3', verbose=1, save_best_only=True)\n",
    "\n",
    "# num epochs\n",
    "epochs = 12\n",
    "\n",
    "# run model\n",
    "model_history = model.fit([x_train, x_train_features], y_train, epochs=epochs,\n",
    "                 shuffle=True,\n",
    "                 batch_size=100, validation_data=([x_validation, x_validation_features], y_validation),\n",
    "                 callbacks=[checkpoint], verbose=2)\n",
    "\n",
    "# save model to json\n",
    "model_json = model.to_json()\n",
    "with open(\"cnn_feats3.json\", \"w\") as json_file:\n",
    "    json_file.write(model_json)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(12,4))\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.suptitle('CNN on raw pixel data', fontsize=10)\n",
    "plt.ylabel('Loss', fontsize=16)\n",
    "plt.plot(model_history.history['loss'], color='b', label='Training Loss')\n",
    "plt.plot(model_history.history['val_loss'], color='r', label='Validation Loss')\n",
    "plt.legend(loc='upper right')\n",
    "\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.ylabel('Accuracy', fontsize=16)\n",
    "plt.plot(model_history.history['acc'], color='b', label='Training Accuracy')\n",
    "plt.plot(model_history.history['val_acc'], color='r', label='Validation Accuracy')\n",
    "plt.legend(loc='lower right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3589/3589 [==============================] - 34s 10ms/step\n"
     ]
    }
   ],
   "source": [
    "y_predicted_probs = model.predict([x_test, x_test_features], verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3589/3589 [==============================] - 18s 5ms/step\n"
     ]
    }
   ],
   "source": [
    "score = model.evaluate([x_test, x_test_features], y_test, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Metrics:  ['loss', 'acc']\n",
      "Loss on test data 1.4949886020959071\n",
      "Test Accuracy 0.44246308164249126\n"
     ]
    }
   ],
   "source": [
    "print(\"Metrics: \", model.metrics_names)\n",
    "print(\"Loss on test data\", score[0])\n",
    "print(\"Test Accuracy\", score[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def convert_probs_to_labels(y_probs, labels):\n",
    "    predicted_label = lambda x: labels[x.argmax()]\n",
    "    return [predicted_label(each) for each in y_probs]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_predicted_labels = convert_probs_to_labels(y_predicted_probs, labels)\n",
    "y_actual_labels = convert_probs_to_labels(y_test, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# https://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html\n",
    "def plot_confusion_matrix(cm, classes,\n",
    "                          normalize=False,\n",
    "                          title='Confusion matrix',\n",
    "                          cmap=plt.cm.Blues):\n",
    "    \"\"\"\n",
    "    This function prints and plots the confusion matrix.\n",
    "    Normalization can be applied by setting `normalize=True`.\n",
    "    \"\"\"\n",
    "    if normalize:\n",
    "        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "        print(\"Normalized confusion matrix\")\n",
    "    else:\n",
    "        print('Confusion matrix, without normalization')\n",
    "\n",
    "    print(cm)\n",
    "\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "    tick_marks = np.arange(len(classes))\n",
    "    plt.xticks(tick_marks, classes, rotation=45)\n",
    "    plt.yticks(tick_marks, classes)\n",
    "\n",
    "    fmt = '.2f' if normalize else 'd'\n",
    "    thresh = cm.max() / 2.\n",
    "    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n",
    "        plt.text(j, i, format(cm[i, j], fmt),\n",
    "                 horizontalalignment=\"center\",\n",
    "                 color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')\n",
    "    plt.tight_layout()\n",
    "    \n",
    "cnf_matrix = confusion_matrix(y_actual_labels, y_predicted_labels)\n",
    "np.set_printoptions(precision=2)\n",
    "\n",
    "# Plot non-normalized confusion matrix\n",
    "fig = plt.figure(figsize=(8,6))\n",
    "plot_confusion_matrix(cnf_matrix, classes=labels,\n",
    "                      title='Confusion Matrix for Test Data with Raw Pixels and HoG features')\n",
    "plt.show()\n",
    "fig.savefig('confusion_matrix_hog.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def test_for_single_image(file_name, model, labels):\n",
    "    image_pixels = cv2.imread(file_name, 0)\n",
    "#     print(len(image_pixels))\n",
    "    image_input = image_pixels.reshape(-1, 48, 48, 1)\n",
    "    predicted_emotion = model.predict(image_input, verbose=0)\n",
    "#     print(predicted_emotion)\n",
    "    print(\"This image is classified to be {} with {:.2f}% confidence\".format(\n",
    "        labels[predicted_emotion.argmax()],predicted_emotion.max()*100))\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_for_single_image('100th.jpg', model, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_for_single_image('ashok.jpg', model, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_for_single_image('ashok2.png', model, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_for_single_image('rose2.jpg', model, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_for_single_image('rose3.jpg', model, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_for_single_image('jose.png', model, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_for_single_image('rohit_happy.jpeg', model, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_for_single_image('ankita.png', model, labels)"
   ]
  }
 ],
 "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": "3.6.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}