Model_baseline_keras.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 256,
   "metadata": {},
   "outputs": [],
   "source": [
    "import keras\n",
    "from keras.layers import Input, Activation, Dense, Conv2D, MaxPooling2D,Flatten,Lambda\n",
    "from keras.models import Model\n",
    "import keras.backend as K\n",
    "from keras.optimizers import SGD\n",
    "from keras.callbacks import LearningRateScheduler\n",
    "from keras.preprocessing.image import ImageDataGenerator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "digit_input = Input(shape=(105,105,1))\n",
    "\n",
    "x = Conv2D(64, (10, 10),activation='relu',input_shape=(105,105,1))(digit_input)\n",
    "x = MaxPooling2D()(x)\n",
    "x = Conv2D(128, (8, 8),activation='relu')(x)\n",
    "x = MaxPooling2D()(x)\n",
    "x = Conv2D(128, (6, 6),activation ='relu')(x)\n",
    "x = MaxPooling2D()(x)\n",
    "x = Conv2D(256, (4, 4),activation='relu')(x)\n",
    "out = Flatten()(x)\n",
    "#out = Dense(4096,activation='sigmoid')(x)\n",
    "vision_model = Model(digit_input,out)\n",
    "\n",
    "letter1 = Input(shape=(105,105,1))\n",
    "letter2 = Input(shape=(105,105,1))\n",
    "\n",
    "out_1 = vision_model(letter1)\n",
    "out_2 = vision_model(letter2)\n",
    "\n",
    "def m_dist(A,B):\n",
    "    return K.sum(K.abs(A-B),axis=1,keepdims=True)\n",
    "\n",
    "merged_vector = Lambda(lambda x:m_dist(x[0],x[1]), output_shape=lambda inp_shp:(inp_shp[0][0],1))([out_1,out_2])\n",
    "\n",
    "out_fin = Dense(4096,activation=\"sigmoid\")(merged_vector)\n",
    "final_output = Dense(1,activation=\"sigmoid\")(out_fin)\n",
    "\n",
    "final_model = keras.models.Model(inputs=[letter1, letter2], outputs=final_output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 477,
   "metadata": {},
   "outputs": [],
   "source": [
    "import math\n",
    "from keras.optimizers import Adam\n",
    "from keras.callbacks import TensorBoard\n",
    "class LossHistory(keras.callbacks.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",
    "def step_decay(epoch):\n",
    "    if epoch%3:\n",
    "        return K.get_value(optimizer.lr)*0.99\n",
    "    return K.get_value(optimizer.lr)\n",
    "  \n",
    "loss_history = LossHistory()\n",
    "lrate = LearningRateScheduler(step_decay)\n",
    "filepath=\"checkpoints/weight1.hdf5\"\n",
    "checkpoint = ModelCheckpoint(filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='min')\n",
    "early_stopper = EarlyStopping(monitor='loss', min_delta=1e-5, patience=25, verbose=1)\n",
    "\n",
    "callbacks_list = [lrate,loss_history,checkpoint,early_stopper]\n",
    "\n",
    "optimizer = Adam(lr=3e-4,amsgrad=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 494,
   "metadata": {},
   "outputs": [],
   "source": [
    "final_model.compile(optimizer=optimizer,\n",
    "                   loss='binary_crossentropy',\n",
    "                   metrics=['binary_accuracy'],)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 495,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_7 (InputLayer)         (None, 105, 105, 1)       0         \n",
      "_________________________________________________________________\n",
      "conv2d_9 (Conv2D)            (None, 96, 96, 64)        6464      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_7 (MaxPooling2 (None, 48, 48, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_10 (Conv2D)           (None, 41, 41, 128)       524416    \n",
      "_________________________________________________________________\n",
      "max_pooling2d_8 (MaxPooling2 (None, 20, 20, 128)       0         \n",
      "_________________________________________________________________\n",
      "conv2d_11 (Conv2D)           (None, 15, 15, 128)       589952    \n",
      "_________________________________________________________________\n",
      "max_pooling2d_9 (MaxPooling2 (None, 7, 7, 128)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_12 (Conv2D)           (None, 4, 4, 256)         524544    \n",
      "_________________________________________________________________\n",
      "flatten_3 (Flatten)          (None, 4096)              0         \n",
      "=================================================================\n",
      "Total params: 1,645,376\n",
      "Trainable params: 1,645,376\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "final_model.layers[2].summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 496,
   "metadata": {},
   "outputs": [],
   "source": [
    "from matplotlib import pyplot as plt \n",
    "%matplotlib inline\n",
    "import glob\n",
    "import cv2\n",
    "import numpy as np\n",
    "from collections import Counter\n",
    "from IPython.core.display import display, HTML\n",
    "import random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 497,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset_train_path , dataset_evaluation_path = (\"python/images_background/\",\"python/images_evaluation/\")\n",
    "\n",
    "languages = [path.split(\"/\")[-1] for path in glob.glob(dataset_train_path+\"*\")]\n",
    "languages_evaluation = [path.split(\"/\")[-1] for path in glob.glob(dataset_evaluation_path+\"*\")]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_alphabets = {language:[cv2.cvtColor(cv2.imread(image),cv2.COLOR_RGB2GRAY) for image in glob.glob(dataset_train_path+language+\"/*/*\")]\\\n",
    "                                                                  for language in languages }\n",
    "all_images =  np.vstack([np.array(all_alphabets[key]) for key in all_alphabets])\n",
    "\n",
    "all_evaluation_alphabets = {language:[cv2.cvtColor(cv2.imread(image),cv2.COLOR_RGB2GRAY) for image in glob.glob(dataset_evaluation_path+language+\"/*/*\")]\\\n",
    "                                                                  for language in languages_evaluation[5:] }\n",
    "all_evaluation_images =  np.vstack([np.array(all_evaluation_alphabets[key]) for key in all_evaluation_alphabets])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 491,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_figures(figures, nrows = 1, ncols=1,figsize=(15,15),cmap=None):\n",
    "    \"\"\"Plot a dictionary of figures.\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    figures : <title, figure> dictionary\n",
    "    ncols : number of columns of subplots wanted in the display\n",
    "    nrows : number of rows of subplots wanted in the figure\n",
    "    \"\"\"\n",
    "\n",
    "    fig, axeslist = plt.subplots(ncols=ncols, nrows=nrows,figsize=figsize)\n",
    "\n",
    "    for ind,title in enumerate(figures):\n",
    "        axeslist.ravel()[ind].imshow(figures[ind],cmap=cmap)\n",
    "        axeslist.ravel()[ind].set_title(ind)\n",
    "        axeslist.ravel()[ind].set_axis_off()\n",
    "    plt.tight_layout() # optional"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 232,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x182a839710>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x182bfbc438>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "for image in (batch[0][0][4],batch[0][1][4]) :\n",
    "    plt.imshow(image.reshape(105,105),cmap=\"bone\")\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 498,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data_gen = ImageDataGenerator(\n",
    "    samplewise_center=True,\n",
    "    samplewise_std_normalization = True,\n",
    "    rotation_range =10,\n",
    "    width_shift_range = 0.05,\n",
    "    height_shift_range = 0.05,\n",
    "    shear_range = 0.05,\n",
    "    zoom_range = 0.05,\n",
    "    #rescale = 1./255         \n",
    ")\n",
    "\n",
    "def flip(p):\n",
    "    return random.random() < p\n",
    "\n",
    "def augmented_generator(features,batch_size,distribution,validation):\n",
    "    \n",
    "    batch_features = np.zeros((batch_size, 2,105, 105,1))\n",
    "    batch_labels = np.zeros((batch_size,1))\n",
    "    macro_dict = all_evaluation_alphabets if validation else all_alphabets\n",
    "    lengthes = np.insert(np.cumsum([len(macro_dict[key]) for key in macro_dict]),0,0)\n",
    "    \n",
    "    while True:\n",
    "\n",
    "        for i in range(batch_size):\n",
    "            if flip(distribution):\n",
    "                not_insame_class = True\n",
    "                while not_insame_class:\n",
    "                    indecies= np.random.choice(len(features),2)\n",
    "                    not_insame_class = np.argmax(lengthes-indecies[0]>0) != np.argmax(lengthes-indecies[1]>0)\n",
    "                batch_labels[i] = not_insame_class\n",
    "            else:\n",
    "                \n",
    "                first = np.searchsorted(lengthes,np.random.randint(1,len(features)))\n",
    "                indecies = np.zeros(2)\n",
    "                while abs(indecies[0]-indecies[1]) < 20 : indecies = np.random.choice(range(lengthes[first-1],lengthes[first]),2)\n",
    "                batch_labels[i] = True\n",
    "    \n",
    "            image1,image2 = features[indecies[0]].reshape(105,105,1),features[indecies[1]].reshape(105,105,1)\n",
    "            if not validation and flip(0.5): \n",
    "                generator1 , generator2 = train_data_gen.flow(image1[np.newaxis,:],batch_size=1) , train_data_gen.flow(image2[np.newaxis,:],batch_size=1)\n",
    "                image1,image2 = next(generator1),next(generator2)\n",
    "            batch_features[i] = (image1,image2)\n",
    "            \n",
    "        yield [batch_features[:,0],batch_features[:,1]], batch_labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 499,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      " - 35s - loss: 0.9387 - binary_accuracy: 0.5187\n",
      "lr: 9.900000000000001e-05\n",
      "Epoch 2/2\n",
      " - 37s - loss: 0.7141 - binary_accuracy: 0.5156\n",
      "lr: 9.801e-05\n"
     ]
    }
   ],
   "source": [
    "history2 = final_model.fit_generator(augmented_generator(all_images,128,0.5,False),\n",
    "        validation_data= augmented_generator(all_evaluation_images,128,0.5,True),\n",
    "        validation_steps = 50,\n",
    "        steps_per_epoch=50,\n",
    "        verbose=2,\n",
    "        epochs=1000,          \n",
    "        callbacks=callbacks_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 509,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x184a209a90>]"
      ]
     },
     "execution_count": 509,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x184b916470>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['loss'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 554,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_kernels(model,layer_num):\n",
    "    \n",
    "    plt.imshow(model.get_weights()[layer_num][:, :, :, 0].squeeze(), cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 553,
   "metadata": {},
   "outputs": [],
   "source": [
    "layer=final_model.get_layer(index=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 523,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'binary_accuracy': [0.51875, 0.515625],\n",
       " 'loss': [0.9386626899242401, 0.7140816032886506]}"
      ]
     },
     "execution_count": 523,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "plot_kernels()"
   ]
  },
  {
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}