{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os, cv2, random\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline \n",
    "from keras.models import Sequential, load_model\n",
    "from keras.layers import Dropout, Flatten, Convolution2D, MaxPooling2D, Dense, Activation\n",
    "from keras.optimizers import Adam\n",
    "from keras.callbacks import Callback, EarlyStopping\n",
    "from keras.callbacks import BaseLogger, TensorBoard"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Constants definition"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "TRAIN_DIR = '/home/dlab-user/train/'\n",
    "TEST_DIR = '/home/dlab-user/test/'\n",
    "ROWS = 128\n",
    "COLS = 128\n",
    "CHANNELS = 3\n",
    "TRAIN_IMAGES_COUNT = 1000\n",
    "PATH_TO_LOGS = '/home/dlab-user/logs'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Reading and adjusting images for training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_images = [TRAIN_DIR+i for i in os.listdir(TRAIN_DIR)[:TRAIN_IMAGES_COUNT]]\n",
    "test_images =  [TEST_DIR+i for i in os.listdir(TEST_DIR)]\n",
    "random.shuffle(all_images)\n",
    "test_coeff = int(len (all_images) * .9)\n",
    "\n",
    "train_images, test_images = all_images[:test_coeff], all_images[test_coeff:]\n",
    "\n",
    "def read_image(file_path):\n",
    "    img = cv2.imread(file_path, cv2.IMREAD_COLOR)\n",
    "    return cv2.resize(img, (ROWS, COLS), interpolation=cv2.INTER_CUBIC).reshape(ROWS, COLS, CHANNELS)\n",
    "\n",
    "def prepare_data(images):\n",
    "    count = len(images)\n",
    "    data = np.ndarray((count, ROWS, COLS, CHANNELS), dtype=np.uint8)\n",
    "\n",
    "    for i, image_file in enumerate(images):\n",
    "        image = read_image(image_file)\n",
    "        data[i] = image#.T\n",
    "    return data\n",
    "\n",
    "train = prepare_data(train_images)\n",
    "test = prepare_data(test_images)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Image counts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(\"Train shape: {}\".format(train.shape))\n",
    "print(\"Test shape: {}\".format(test.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Assigning labels to training images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "labels = []\n",
    "for i in train_images:\n",
    "    if 'dog' in i.split(\"/\")[-1] :\n",
    "        labels.append(1)\n",
    "    else:\n",
    "        labels.append(0)\n",
    "        \n",
    "labels_test = []\n",
    "for i in test_images:\n",
    "    if 'dog' in i.split(\"/\")[-1] :\n",
    "        labels_test.append(1)\n",
    "    else:\n",
    "        labels_test.append(0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Building a convnet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizer = Adam(lr=1e-6)\n",
    "objective = 'binary_crossentropy'\n",
    "\n",
    "def build_model():\n",
    "    \n",
    "    model = Sequential()\n",
    "\n",
    "    model.add(Convolution2D(32, 3, 3, border_mode='same', input_shape=(ROWS, COLS, 3), activation='relu'))\n",
    "    model.add(Convolution2D(32, 3, 3, border_mode='same', activation='relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Convolution2D(64, 3, 3, border_mode='same', activation='relu'))\n",
    "    model.add(Convolution2D(64, 3, 3, border_mode='same', activation='relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    model.add(Convolution2D(128, 3, 3, border_mode='same', activation='relu'))\n",
    "    model.add(Convolution2D(128, 3, 3, border_mode='same', activation='relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(256, activation='relu'))\n",
    "    model.add(Dropout(0.5))\n",
    "    \n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "    \n",
    "    model.compile(loss=objective, optimizer=optimizer, metrics=['accuracy'])\n",
    "    return model\n",
    "\n",
    "\n",
    "model = build_model()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training the model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This block takes about 2.5-3 hours to execute if training on whole dataset of 22500 images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "nb_epoch = 10\n",
    "batch_size = 16\n",
    "\n",
    "class LossHistory(Callback):\n",
    "    def on_train_begin(self, logs={}):\n",
    "        self.losses = []\n",
    "        self.val_losses = []\n",
    "        \n",
    "    def on_epoch_end(self, batch, logs={}):\n",
    "        self.losses.append(logs.get('loss'))\n",
    "        self.val_losses.append(logs.get('val_loss'))\n",
    "\n",
    "early_stopping = EarlyStopping(monitor='val_loss', patience=5, verbose=1, mode='auto')        \n",
    "        \n",
    "def train_and_test_model():\n",
    "    history = LossHistory()\n",
    "    tensorboard = TensorBoard(log_dir=PATH_TO_LOGS)\n",
    "    model.fit(train, labels, batch_size=batch_size, nb_epoch=nb_epoch,\n",
    "              validation_split=0.25, verbose=2, shuffle=True, callbacks=[history, early_stopping, tensorboard])\n",
    "    \n",
    "\n",
    "    predictions = model.predict(test, verbose=2)\n",
    "    return predictions, history\n",
    "\n",
    "predictions, history = train_and_test_model()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Saving the model and weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "path_to_model = '/home/dlab-user/model_1000.json'\n",
    "path_to_weights = '/home/dlab-user/weigths_1000.h5'\n",
    "\n",
    "model.save(path_to_model)\n",
    "model.save_weights(path_to_weights)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Plotting learning curves"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "loss = history.losses\n",
    "val_loss = history.val_losses\n",
    "\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.title('VGG-16 Loss Trend')\n",
    "plt.plot(loss, 'blue', label='Training Loss')\n",
    "plt.plot(val_loss, 'green', label='Validation Loss')\n",
    "plt.xticks(range(0,len(loss))[0::2])\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "KERNEL_NAME"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}