Using a pretrained Convolution model on ImageNet for feature extraction.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5\n",
      "58892288/58889256 [==============================] - 9s 0us/step\n",
      "58900480/58889256 [==============================] - 9s 0us/step\n"
     ]
    }
   ],
   "source": [
    "from keras.applications import VGG16\n",
    "\n",
    "conv_base = VGG16(weights='imagenet', include_top=False, input_shape=(150, 150, 3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         (None, 150, 150, 3)       0         \n",
      "_________________________________________________________________\n",
      "block1_conv1 (Conv2D)        (None, 150, 150, 64)      1792      \n",
      "_________________________________________________________________\n",
      "block1_conv2 (Conv2D)        (None, 150, 150, 64)      36928     \n",
      "_________________________________________________________________\n",
      "block1_pool (MaxPooling2D)   (None, 75, 75, 64)        0         \n",
      "_________________________________________________________________\n",
      "block2_conv1 (Conv2D)        (None, 75, 75, 128)       73856     \n",
      "_________________________________________________________________\n",
      "block2_conv2 (Conv2D)        (None, 75, 75, 128)       147584    \n",
      "_________________________________________________________________\n",
      "block2_pool (MaxPooling2D)   (None, 37, 37, 128)       0         \n",
      "_________________________________________________________________\n",
      "block3_conv1 (Conv2D)        (None, 37, 37, 256)       295168    \n",
      "_________________________________________________________________\n",
      "block3_conv2 (Conv2D)        (None, 37, 37, 256)       590080    \n",
      "_________________________________________________________________\n",
      "block3_conv3 (Conv2D)        (None, 37, 37, 256)       590080    \n",
      "_________________________________________________________________\n",
      "block3_pool (MaxPooling2D)   (None, 18, 18, 256)       0         \n",
      "_________________________________________________________________\n",
      "block4_conv1 (Conv2D)        (None, 18, 18, 512)       1180160   \n",
      "_________________________________________________________________\n",
      "block4_conv2 (Conv2D)        (None, 18, 18, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block4_conv3 (Conv2D)        (None, 18, 18, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block4_pool (MaxPooling2D)   (None, 9, 9, 512)         0         \n",
      "_________________________________________________________________\n",
      "block5_conv1 (Conv2D)        (None, 9, 9, 512)         2359808   \n",
      "_________________________________________________________________\n",
      "block5_conv2 (Conv2D)        (None, 9, 9, 512)         2359808   \n",
      "_________________________________________________________________\n",
      "block5_conv3 (Conv2D)        (None, 9, 9, 512)         2359808   \n",
      "_________________________________________________________________\n",
      "block5_pool (MaxPooling2D)   (None, 4, 4, 512)         0         \n",
      "=================================================================\n",
      "Total params: 14,714,688\n",
      "Trainable params: 14,714,688\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "conv_base.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "\n",
    "base_dir = '/home/geist/Desktop/projects/kaggle/cats_dogs/cats_and_dogs_small'\n",
    "train_dir = os.path.join(base_dir, 'train')\n",
    "validation_dir = os.path.join(base_dir, 'validation')\n",
    "test_dir = os.path.join(base_dir, 'test')\n",
    "\n",
    "datagen = ImageDataGenerator(rescale=1./255)\n",
    "batch_size = 20\n",
    "\n",
    "def extract_features(directory, sample_count):\n",
    "    features = np.zeros(shape=(sample_count, 4, 4, 512))\n",
    "    labels= np.zeros(shape=(sample_count))\n",
    "    generator = datagen.flow_from_directory(\n",
    "        directory,\n",
    "        target_size=(150,150),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "    i=0\n",
    "    for input_batch, labels_batch in generator:\n",
    "        features_batch = conv_base.predict(input_batch)\n",
    "        features[i*batch_size : (i+1)*batch_size] = features_batch\n",
    "        labels[i*batch_size : (i+1)*batch_size] = labels_batch\n",
    "        i += 1\n",
    "        if i*batch_size >= sample_count:\n",
    "            break\n",
    "    return features, labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 2000 images belonging to 2 classes.\n",
      "Found 1000 images belonging to 2 classes.\n",
      "Found 1000 images belonging to 2 classes.\n"
     ]
    }
   ],
   "source": [
    "train_features, train_labels = extract_features(train_dir, 2000)\n",
    "validation_features, validation_labels = extract_features(validation_dir, 1000)\n",
    "test_features, test_labels = extract_features(test_dir, 1000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_features = np.reshape(train_features, (2000, 4*4*512))\n",
    "validation_features = np.reshape(validation_features, (1000, 4*4*512))\n",
    "test_features = np.reshape(test_features, (1000, 4*4*512))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 2000 samples, validate on 1000 samples\n",
      "Epoch 1/30\n",
      "2000/2000 [==============================] - 3s 1ms/step - loss: 0.6178 - acc: 0.6500 - val_loss: 0.4463 - val_acc: 0.8370\n",
      "Epoch 2/30\n",
      "2000/2000 [==============================] - 0s 213us/step - loss: 0.4361 - acc: 0.7990 - val_loss: 0.3600 - val_acc: 0.8560\n",
      "Epoch 3/30\n",
      "2000/2000 [==============================] - 0s 216us/step - loss: 0.3521 - acc: 0.8525 - val_loss: 0.3230 - val_acc: 0.8760\n",
      "Epoch 4/30\n",
      "2000/2000 [==============================] - 0s 214us/step - loss: 0.3229 - acc: 0.8590 - val_loss: 0.3142 - val_acc: 0.8680\n",
      "Epoch 5/30\n",
      "2000/2000 [==============================] - 0s 213us/step - loss: 0.2886 - acc: 0.8830 - val_loss: 0.2821 - val_acc: 0.8890\n",
      "Epoch 6/30\n",
      "2000/2000 [==============================] - 0s 213us/step - loss: 0.2749 - acc: 0.8850 - val_loss: 0.2705 - val_acc: 0.8940\n",
      "Epoch 7/30\n",
      "2000/2000 [==============================] - 0s 214us/step - loss: 0.2479 - acc: 0.9015 - val_loss: 0.2632 - val_acc: 0.8970\n",
      "Epoch 8/30\n",
      "2000/2000 [==============================] - 0s 227us/step - loss: 0.2262 - acc: 0.9165 - val_loss: 0.2570 - val_acc: 0.9020\n",
      "Epoch 9/30\n",
      "2000/2000 [==============================] - 0s 223us/step - loss: 0.2189 - acc: 0.9100 - val_loss: 0.2541 - val_acc: 0.9000\n",
      "Epoch 10/30\n",
      "2000/2000 [==============================] - 0s 216us/step - loss: 0.2035 - acc: 0.9265 - val_loss: 0.2466 - val_acc: 0.9010\n",
      "Epoch 11/30\n",
      "2000/2000 [==============================] - 0s 219us/step - loss: 0.1958 - acc: 0.9265 - val_loss: 0.2540 - val_acc: 0.8900\n",
      "Epoch 12/30\n",
      "2000/2000 [==============================] - 0s 216us/step - loss: 0.1878 - acc: 0.9315 - val_loss: 0.2513 - val_acc: 0.8930\n",
      "Epoch 13/30\n",
      "2000/2000 [==============================] - 0s 216us/step - loss: 0.1797 - acc: 0.9350 - val_loss: 0.2411 - val_acc: 0.8990\n",
      "Epoch 14/30\n",
      "2000/2000 [==============================] - 0s 215us/step - loss: 0.1696 - acc: 0.9385 - val_loss: 0.2569 - val_acc: 0.8920\n",
      "Epoch 15/30\n",
      "2000/2000 [==============================] - 0s 216us/step - loss: 0.1667 - acc: 0.9420 - val_loss: 0.2363 - val_acc: 0.9050\n",
      "Epoch 16/30\n",
      "2000/2000 [==============================] - 0s 218us/step - loss: 0.1616 - acc: 0.9445 - val_loss: 0.2376 - val_acc: 0.9030\n",
      "Epoch 17/30\n",
      "2000/2000 [==============================] - 0s 225us/step - loss: 0.1454 - acc: 0.9515 - val_loss: 0.2355 - val_acc: 0.9020\n",
      "Epoch 18/30\n",
      "2000/2000 [==============================] - 0s 218us/step - loss: 0.1358 - acc: 0.9535 - val_loss: 0.2391 - val_acc: 0.9010\n",
      "Epoch 19/30\n",
      "2000/2000 [==============================] - 0s 214us/step - loss: 0.1392 - acc: 0.9530 - val_loss: 0.2366 - val_acc: 0.9040\n",
      "Epoch 20/30\n",
      "2000/2000 [==============================] - 0s 216us/step - loss: 0.1300 - acc: 0.9565 - val_loss: 0.2375 - val_acc: 0.9030\n",
      "Epoch 21/30\n",
      "2000/2000 [==============================] - 0s 228us/step - loss: 0.1308 - acc: 0.9545 - val_loss: 0.2353 - val_acc: 0.9030\n",
      "Epoch 22/30\n",
      "2000/2000 [==============================] - 0s 216us/step - loss: 0.1237 - acc: 0.9590 - val_loss: 0.2361 - val_acc: 0.9010\n",
      "Epoch 23/30\n",
      "2000/2000 [==============================] - 0s 218us/step - loss: 0.1157 - acc: 0.9615 - val_loss: 0.2382 - val_acc: 0.8990\n",
      "Epoch 24/30\n",
      "2000/2000 [==============================] - 0s 215us/step - loss: 0.1144 - acc: 0.9655 - val_loss: 0.2372 - val_acc: 0.9000\n",
      "Epoch 25/30\n",
      "2000/2000 [==============================] - 0s 216us/step - loss: 0.1097 - acc: 0.9640 - val_loss: 0.2452 - val_acc: 0.9010\n",
      "Epoch 26/30\n",
      "2000/2000 [==============================] - 0s 228us/step - loss: 0.1069 - acc: 0.9645 - val_loss: 0.2429 - val_acc: 0.8980\n",
      "Epoch 27/30\n",
      "2000/2000 [==============================] - 0s 218us/step - loss: 0.0988 - acc: 0.9690 - val_loss: 0.2387 - val_acc: 0.9010\n",
      "Epoch 28/30\n",
      "2000/2000 [==============================] - 0s 217us/step - loss: 0.0938 - acc: 0.9710 - val_loss: 0.2383 - val_acc: 0.9020\n",
      "Epoch 29/30\n",
      "2000/2000 [==============================] - 0s 218us/step - loss: 0.0912 - acc: 0.9740 - val_loss: 0.2434 - val_acc: 0.8980\n",
      "Epoch 30/30\n",
      "2000/2000 [==============================] - 0s 218us/step - loss: 0.0897 - acc: 0.9750 - val_loss: 0.2529 - val_acc: 0.8980\n"
     ]
    }
   ],
   "source": [
    "from keras import models\n",
    "from keras import layers\n",
    "from keras import optimizers\n",
    "\n",
    "model = models.Sequential()\n",
    "model.add(layers.Dense(256, activation='relu', input_dim=4*4*512))\n",
    "model.add(layers.Dropout(0.5))\n",
    "model.add(layers.Dense(1, activation='sigmoid'))\n",
    "\n",
    "model.compile(optimizer = optimizers.RMSprop(lr=2e-5),\n",
    "             loss = 'binary_crossentropy',\n",
    "             metrics=['acc'])\n",
    "\n",
    "history = model.fit(train_features, train_labels, epochs=30, batch_size=20, \n",
    "                    validation_data=(validation_features,validation_labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fd50c05f650>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fd4f8700bd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "acc = history.history['acc']\n",
    "val_acc = history.history['val_acc']\n",
    "loss = history.history['loss']\n",
    "val_loss = history.history['val_loss']\n",
    "\n",
    "epochs = range(1, len(acc)+1)\n",
    "\n",
    "plt.plot(epochs, acc, 'bo', label='Training_acc')\n",
    "plt.plot(epochs, val_acc, 'b', label='Validation acc')\n",
    "plt.title('Training and validation acc')\n",
    "plt.legend()\n",
    "\n",
    "plt.figure()\n",
    "\n",
    "plt.plot(epochs, loss, 'bo', label='Training_loss')\n",
    "plt.plot(epochs, val_loss, 'b', label='Validation loss')\n",
    "plt.title('Training and validation loss')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Accuracy of ~90%"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.15rc1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}