Transfer Learning - VGG.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import time\n",
    "import keras as k \n",
    "from keras_preprocessing.image import ImageDataGenerator\n",
    "from keras.models import Sequential\n",
    "from keras.applications import vgg16\n",
    "from keras.models import Model\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Activation, Flatten, Dropout,InputLayer\n",
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras import regularizers, optimizers\n",
    "from keras.metrics import categorical_accuracy\n",
    "from keras.callbacks import EarlyStopping, ReduceLROnPlateau"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.random.seed(1)\n",
    "\n",
    "df = pd.read_csv(r'C:\\Users\\casocha\\Downloads\\histopathologic-cancer-detection/train_labels.csv')\n",
    "\n",
    "def append_ext(fn):\n",
    "    return fn+\".tif\"\n",
    "\n",
    "df[\"id\"]=df[\"id\"].apply(append_ext)\n",
    "                   \n",
    "\n",
    "df[\"label\"]=df[\"label\"].astype(str)\n",
    "\n",
    "train_path = r'C:\\Users\\casocha\\Downloads\\histopathologic-cancer-detection\\train'\n",
    "valid_path = r'C:\\Users\\casocha\\Downloads\\histopathologic-cancer-detection\\train'\n",
    "\n",
    "testdf= pd.read_csv(r'C:\\Users\\casocha\\Downloads\\histopathologic-cancer-detection/sample_submission.csv')\n",
    "\n",
    "def append_ext2(fn):\n",
    "    return r'C:\\Users\\casocha\\Downloads\\histopathologic-cancer-detection\\test/' + fn+\".tif\"\n",
    "\n",
    "testdf[\"id\"]=testdf[\"id\"].apply(append_ext2)\n",
    "                   \n",
    "                   \n",
    "testdf = testdf.drop(['label'],axis=1)\n",
    "\n",
    "test_path = r'C:\\Users\\casocha\\Downloads\\histopathologic-cancer-detection\\test/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_datagen = ImageDataGenerator(rescale=1./255)\n",
    "\n",
    "test_generator=test_datagen.flow_from_dataframe(\n",
    "dataframe=testdf,\n",
    "directory=\"test_path\",\n",
    "x_col=\"id\",\n",
    "y_col=None,\n",
    "batch_size=64,\n",
    "seed=1,\n",
    "shuffle=False,\n",
    "class_mode=None,\n",
    "use_mutiprocessing = True,\n",
    "target_size=(96,96))\n",
    "\n",
    "train_datagen = ImageDataGenerator(\n",
    "       # horizontal_flip=True,\n",
    "       #vertical_flip=True,\n",
    "       #brightness_range=[0.5, 1.5],\n",
    "       #fill_mode='reflect',                               \n",
    "        #rotation_range=15,\n",
    "        rescale=1./255,\n",
    "        #shear_range=0.2,\n",
    "        #zoom_range=0.2\n",
    "        validation_split=0.15\n",
    "    \n",
    ")\n",
    "\n",
    "\n",
    "train_generator = train_datagen.flow_from_dataframe(\n",
    "                dataframe=df,\n",
    "                directory=train_path,\n",
    "                x_col = 'id',\n",
    "                y_col = 'label',\n",
    "                has_ext=False,\n",
    "                subset='training',\n",
    "                target_size=(96, 96),\n",
    "                batch_size=64,\n",
    "                seed=1,\n",
    "                class_mode='binary',\n",
    "                use_mutiprocessing = True\n",
    "                )\n",
    "\n",
    "validation_generator = train_datagen.flow_from_dataframe(\n",
    "                dataframe=df,\n",
    "                directory=valid_path,\n",
    "                x_col = 'id',\n",
    "                y_col = 'label',\n",
    "                has_ext=False,\n",
    "                subset='validation', # This is the trick to properly separate train and validation dataset\n",
    "                target_size=(96, 96),\n",
    "                batch_size=64,\n",
    "                seed=1,\n",
    "                shuffle=False,\n",
    "                class_mode='binary',\n",
    "                use_mutiprocessing = True\n",
    "                )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_shape = (96, 96, 3)\n",
    "\n",
    "vgg_model = vgg16.VGG16(include_top=False, weights='imagenet', \n",
    "                                     input_shape=input_shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "vgg_model.trainable = False\n",
    "for layer in vgg_model.layers:\n",
    "    layer.trainable = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "pd.set_option('max_colwidth', -1)\n",
    "layers = [(layer, layer.name, layer.trainable) for layer in vgg_model.layers]\n",
    "pd.DataFrame(layers, columns=['Layer Type', 'Layer Name', 'Layer Trainable'])  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(vgg_model)\n",
    "model.add(Flatten())\n",
    "model.add(Dense(64,activation='relu'))\n",
    "model.add(Dense(1,activation = 'sigmoid'))\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss='binary_crossentropy',\n",
    "              optimizer=optimizers.RMSprop(lr=1e-4),\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "STEP_SIZE_TRAIN=train_generator.n//train_generator.batch_size\n",
    "STEP_SIZE_VALID=validation_generator.n//validation_generator.batch_size\n",
    "STEP_SIZE_TEST=test_generator.n//test_generator.batch_size\n",
    "\n",
    "earlystopper = EarlyStopping(monitor='val_loss', patience=2, verbose=1, restore_best_weights=True)\n",
    "reducel = ReduceLROnPlateau(monitor='val_loss', patience=1, verbose=1, factor=0.1)\n",
    "\n",
    "start = time.time()\n",
    "history = model.fit_generator(generator=train_generator,\n",
    "                    steps_per_epoch=STEP_SIZE_TRAIN,\n",
    "                    validation_data=validation_generator,\n",
    "                    validation_steps=STEP_SIZE_VALID,\n",
    "                    epochs=6,callbacks=[reducel, earlystopper]\n",
    ")\n",
    "\n",
    "end = time.time()\n",
    "print(\"Training time: \", (end - start)/60, \" minutes\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.evaluate_generator(generator=validation_generator,\n",
    "steps=STEP_SIZE_TEST)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "plt.figure(figsize=(12,6))\n",
    "# summarize history for accuracy\n",
    "plt.subplot(121)\n",
    "plt.plot(history.history['acc'])\n",
    "plt.plot(history.history['val_acc'])\n",
    "plt.title('Transfer Learning')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "# summarize history for loss\n",
    "plt.subplot(122)\n",
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('Transfer Learning')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper right')\n",
    "plt.show()"
   ]
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}