5.3-using-a-pretrained-convnet.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'2.4.3'"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import keras\n",
    "keras.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Using a pre-trained convnet\n",
    "\n",
    "This notebook contains the code sample found in Chapter 5, Section 3 of [Deep Learning with Python](https://www.manning.com/books/deep-learning-with-python?a_aid=keras&a_bid=76564dff). Note that the original text features far more content, in particular further explanations and figures: in this notebook, you will only find source code and related comments.\n",
    "\n",
    "----\n",
    "\n",
    "A common and highly effective approach to deep learning on small image datasets is to leverage a pre-trained network. A pre-trained network \n",
    "is simply a saved network previously trained on a large dataset, typically on a large-scale image classification task. If this original \n",
    "dataset is large enough and general enough, then the spatial feature hierarchy learned by the pre-trained network can effectively act as a \n",
    "generic model of our visual world, and hence its features can prove useful for many different computer vision problems, even though these \n",
    "new problems might involve completely different classes from those of the original task. For instance, one might train a network on \n",
    "ImageNet (where classes are mostly animals and everyday objects) and then re-purpose this trained network for something as remote as \n",
    "identifying furniture items in images. Such portability of learned features across different problems is a key advantage of deep learning \n",
    "compared to many older shallow learning approaches, and it makes deep learning very effective for small-data problems.\n",
    "\n",
    "In our case, we will consider a large convnet trained on the ImageNet dataset (1.4 million labeled images and 1000 different classes). \n",
    "ImageNet contains many animal classes, including different species of cats and dogs, and we can thus expect to perform very well on our cat \n",
    "vs. dog classification problem.\n",
    "\n",
    "We will use the VGG16 architecture, developed by Karen Simonyan and Andrew Zisserman in 2014, a simple and widely used convnet architecture \n",
    "for ImageNet. Although it is a bit of an older model, far from the current state of the art and somewhat heavier than many other recent \n",
    "models, we chose it because its architecture is similar to what you are already familiar with, and easy to understand without introducing \n",
    "any new concepts. This may be your first encounter with one of these cutesie model names -- VGG, ResNet, Inception, Inception-ResNet, \n",
    "Xception... you will get used to them, as they will come up frequently if you keep doing deep learning for computer vision.\n",
    "\n",
    "There are two ways to leverage a pre-trained network: *feature extraction* and *fine-tuning*. We will cover both of them. Let's start with \n",
    "feature extraction."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature extraction\n",
    "\n",
    "Feature extraction consists of using the representations learned by a previous network to extract interesting features from new samples. \n",
    "These features are then run through a new classifier, which is trained from scratch.\n",
    "\n",
    "As we saw previously, convnets used for image classification comprise two parts: they start with a series of pooling and convolution \n",
    "layers, and they end with a densely-connected classifier. The first part is called the \"convolutional base\" of the model. In the case of \n",
    "convnets, \"feature extraction\" will simply consist of taking the convolutional base of a previously-trained network, running the new data \n",
    "through it, and training a new classifier on top of the output.\n",
    "\n",
    "![swapping FC classifiers](https://s3.amazonaws.com/book.keras.io/img/ch5/swapping_fc_classifier.png)\n",
    "\n",
    "Why only reuse the convolutional base? Could we reuse the densely-connected classifier as well? In general, it should be avoided. The \n",
    "reason is simply that the representations learned by the convolutional base are likely to be more generic and therefore more reusable: the \n",
    "feature maps of a convnet are presence maps of generic concepts over a picture, which is likely to be useful regardless of the computer \n",
    "vision problem at hand. On the other end, the representations learned by the classifier will necessarily be very specific to the set of \n",
    "classes that the model was trained on -- they will only contain information about the presence probability of this or that class in the \n",
    "entire picture. Additionally, representations found in densely-connected layers no longer contain any information about _where_ objects are \n",
    "located in the input image: these layers get rid of the notion of space, whereas the object location is still described by convolutional \n",
    "feature maps. For problems where object location matters, densely-connected features would be largely useless.\n",
    "\n",
    "Note that the level of generality (and therefore reusability) of the representations extracted by specific convolution layers depends on \n",
    "the depth of the layer in the model. Layers that come earlier in the model extract local, highly generic feature maps (such as visual \n",
    "edges, colors, and textures), while layers higher-up extract more abstract concepts (such as \"cat ear\" or \"dog eye\"). So if your new \n",
    "dataset differs a lot from the dataset that the original model was trained on, you may be better off using only the first few layers of the \n",
    "model to do feature extraction, rather than using the entire convolutional base.\n",
    "\n",
    "In our case, since the ImageNet class set did contain multiple dog and cat classes, it is likely that it would be beneficial to reuse the \n",
    "information contained in the densely-connected layers of the original model. However, we will chose not to, in order to cover the more \n",
    "general case where the class set of the new problem does not overlap with the class set of the original model."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's put this in practice by using the convolutional base of the VGG16 network, trained on ImageNet, to extract interesting features from \n",
    "our cat and dog images, and then training a cat vs. dog classifier on top of these features.\n",
    "\n",
    "The VGG16 model, among others, comes pre-packaged with Keras. You can import it from the `keras.applications` module. Here's the list of \n",
    "image classification models (all pre-trained on the ImageNet dataset) that are available as part of `keras.applications`:\n",
    "\n",
    "* Xception\n",
    "* InceptionV3\n",
    "* ResNet50\n",
    "* VGG16\n",
    "* VGG19\n",
    "* MobileNet\n",
    "\n",
    "Let's instantiate the VGG16 model:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.applications import VGG16\n",
    "\n",
    "conv_base = VGG16(weights='imagenet',\n",
    "                  include_top=False,\n",
    "                  input_shape=(150, 150, 3))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We passed three arguments to the constructor:\n",
    "\n",
    "* `weights`, to specify which weight checkpoint to initialize the model from\n",
    "* `include_top`, which refers to including or not the densely-connected classifier on top of the network. By default, this \n",
    "densely-connected classifier would correspond to the 1000 classes from ImageNet. Since we intend to use our own densely-connected \n",
    "classifier (with only two classes, cat and dog), we don't need to include it.\n",
    "* `input_shape`, the shape of the image tensors that we will feed to the network. This argument is purely optional: if we don't pass it, \n",
    "then the network will be able to process inputs of any size.\n",
    "\n",
    "Here's the detail of the architecture of the VGG16 convolutional base: it's very similar to the simple convnets that you are already \n",
    "familiar with."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"vgg16\"\n",
      "_________________________________________________________________\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": "markdown",
   "metadata": {},
   "source": [
    "The final feature map has shape `(4, 4, 512)`. That's the feature on top of which we will stick a densely-connected classifier.\n",
    "\n",
    "At this point, there are two ways we could proceed: \n",
    "\n",
    "* Running the convolutional base over our dataset, recording its output to a Numpy array on disk, then using this data as input to a \n",
    "standalone densely-connected classifier similar to those you have seen in the first chapters of this book. This solution is very fast and \n",
    "cheap to run, because it only requires running the convolutional base once for every input image, and the convolutional base is by far the \n",
    "most expensive part of the pipeline. However, for the exact same reason, this technique would not allow us to leverage data augmentation at \n",
    "all.\n",
    "* Extending the model we have (`conv_base`) by adding `Dense` layers on top, and running the whole thing end-to-end on the input data. This \n",
    "allows us to use data augmentation, because every input image is going through the convolutional base every time it is seen by the model. \n",
    "However, for this same reason, this technique is far more expensive than the first one.\n",
    "\n",
    "We will cover both techniques. Let's walk through the code required to set-up the first one: recording the output of `conv_base` on our \n",
    "data and using these outputs as inputs to a new model.\n",
    "\n",
    "We will start by simply running instances of the previously-introduced `ImageDataGenerator` to extract images as Numpy arrays as well as \n",
    "their labels. We will extract features from these images simply by calling the `predict` method of the `conv_base` model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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": [
    "import os\n",
    "import numpy as np\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "\n",
    "base_dir = '/home/vi/sources/nn/data/DogsCats/small'\n",
    "\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 inputs_batch, labels_batch in generator:\n",
    "        features_batch = conv_base.predict(inputs_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",
    "            # Note that since generators yield data indefinitely in a loop,\n",
    "            # we must `break` after every image has been seen once.\n",
    "            break\n",
    "    return features, labels\n",
    "\n",
    "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": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((2000, 4, 4, 512), (2000,))"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_features.shape, train_labels.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The extracted features are currently of shape `(samples, 4, 4, 512)`. We will feed them to a densely-connected classifier, so first we must \n",
    "flatten them to `(samples, 8192)`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "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": "markdown",
   "metadata": {},
   "source": [
    "At this point, we can define our densely-connected classifier (note the use of dropout for regularization), and train it on the data and \n",
    "labels that we just recorded:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "100/100 [==============================] - 1s 8ms/step - loss: 0.6807 - acc: 0.6017 - val_loss: 0.4524 - val_acc: 0.8450\n",
      "Epoch 2/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.4621 - acc: 0.7965 - val_loss: 0.3711 - val_acc: 0.8590\n",
      "Epoch 3/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.3920 - acc: 0.8264 - val_loss: 0.3373 - val_acc: 0.8560\n",
      "Epoch 4/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.3117 - acc: 0.8863 - val_loss: 0.3020 - val_acc: 0.8880\n",
      "Epoch 5/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.2842 - acc: 0.8835 - val_loss: 0.3011 - val_acc: 0.8710\n",
      "Epoch 6/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.2655 - acc: 0.8907 - val_loss: 0.2752 - val_acc: 0.8970\n",
      "Epoch 7/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.2481 - acc: 0.9051 - val_loss: 0.2668 - val_acc: 0.8990\n",
      "Epoch 8/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.2372 - acc: 0.9088 - val_loss: 0.2635 - val_acc: 0.8990\n",
      "Epoch 9/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.2149 - acc: 0.9143 - val_loss: 0.2542 - val_acc: 0.9000\n",
      "Epoch 10/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1988 - acc: 0.9391 - val_loss: 0.2554 - val_acc: 0.8980\n",
      "Epoch 11/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.2042 - acc: 0.9233 - val_loss: 0.2480 - val_acc: 0.8980\n",
      "Epoch 12/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1795 - acc: 0.9336 - val_loss: 0.2481 - val_acc: 0.9040\n",
      "Epoch 13/30\n",
      "100/100 [==============================] - 0s 5ms/step - loss: 0.1897 - acc: 0.9296 - val_loss: 0.2465 - val_acc: 0.9030\n",
      "Epoch 14/30\n",
      "100/100 [==============================] - 0s 5ms/step - loss: 0.1512 - acc: 0.9497 - val_loss: 0.2411 - val_acc: 0.9080\n",
      "Epoch 15/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1735 - acc: 0.9382 - val_loss: 0.2619 - val_acc: 0.8880\n",
      "Epoch 16/30\n",
      "100/100 [==============================] - 0s 5ms/step - loss: 0.1557 - acc: 0.9466 - val_loss: 0.2465 - val_acc: 0.9050\n",
      "Epoch 17/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1501 - acc: 0.9508 - val_loss: 0.2489 - val_acc: 0.9010\n",
      "Epoch 18/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1347 - acc: 0.9556 - val_loss: 0.2382 - val_acc: 0.9040\n",
      "Epoch 19/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1437 - acc: 0.9464 - val_loss: 0.2378 - val_acc: 0.9020\n",
      "Epoch 20/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1393 - acc: 0.9559 - val_loss: 0.2404 - val_acc: 0.9050\n",
      "Epoch 21/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1338 - acc: 0.9566 - val_loss: 0.2425 - val_acc: 0.9080\n",
      "Epoch 22/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1211 - acc: 0.9540 - val_loss: 0.2362 - val_acc: 0.9030\n",
      "Epoch 23/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1177 - acc: 0.9606 - val_loss: 0.2358 - val_acc: 0.9030\n",
      "Epoch 24/30\n",
      "100/100 [==============================] - 0s 5ms/step - loss: 0.0909 - acc: 0.9773 - val_loss: 0.2375 - val_acc: 0.9030\n",
      "Epoch 25/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.1114 - acc: 0.9607 - val_loss: 0.2384 - val_acc: 0.9030\n",
      "Epoch 26/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.0936 - acc: 0.9720 - val_loss: 0.2378 - val_acc: 0.9030\n",
      "Epoch 27/30\n",
      "100/100 [==============================] - 0s 5ms/step - loss: 0.0998 - acc: 0.9663 - val_loss: 0.2402 - val_acc: 0.9020\n",
      "Epoch 28/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.0996 - acc: 0.9685 - val_loss: 0.2464 - val_acc: 0.9020\n",
      "Epoch 29/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.0835 - acc: 0.9716 - val_loss: 0.2573 - val_acc: 0.8970\n",
      "Epoch 30/30\n",
      "100/100 [==============================] - 0s 4ms/step - loss: 0.0915 - acc: 0.9717 - val_loss: 0.2598 - val_acc: 0.8960\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,\n",
    "                    epochs=30,\n",
    "                    batch_size=20,\n",
    "                    validation_data=(validation_features, validation_labels))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Training is very fast, since we only have to deal with two `Dense` layers -- an epoch takes less than one second even on CPU.\n",
    "\n",
    "Let's take a look at the loss and accuracy curves during training:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "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(len(acc))\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 accuracy')\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": [
    "\n",
    "We reach a validation accuracy of about 90%, much better than what we could achieve in the previous section with our small model trained from \n",
    "scratch. However, our plots also indicate that we are overfitting almost from the start -- despite using dropout with a fairly large rate. \n",
    "This is because this technique does not leverage data augmentation, which is essential to preventing overfitting with small image datasets.\n",
    "\n",
    "Now, let's review the second technique we mentioned for doing feature extraction, which is much slower and more expensive, but which allows \n",
    "us to leverage data augmentation during training: extending the `conv_base` model and running it end-to-end on the inputs. Note that this \n",
    "technique is in fact so expensive that you should only attempt it if you have access to a GPU: it is absolutely intractable on CPU. If you \n",
    "cannot run your code on GPU, then the previous technique is the way to go.\n",
    "\n",
    "Because models behave just like layers, you can add a model (like our `conv_base`) to a `Sequential` model just like you would add a layer. \n",
    "So you can do the following:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras import models\n",
    "from keras import layers\n",
    "\n",
    "model = models.Sequential()\n",
    "model.add(conv_base)\n",
    "model.add(layers.Flatten())\n",
    "model.add(layers.Dense(256, activation='relu'))\n",
    "model.add(layers.Dense(1, activation='sigmoid'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is what our model looks like now:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "vgg16 (Functional)           (None, 4, 4, 512)         14714688  \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 8192)              0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 256)               2097408   \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 1)                 257       \n",
      "=================================================================\n",
      "Total params: 16,812,353\n",
      "Trainable params: 16,812,353\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As you can see, the convolutional base of VGG16 has 14,714,688 parameters, which is very large. The classifier we are adding on top has 2 \n",
    "million parameters.\n",
    "\n",
    "Before we compile and train our model, a very important thing to do is to freeze the convolutional base. \"Freezing\" a layer or set of \n",
    "layers means preventing their weights from getting updated during training. If we don't do this, then the representations that were \n",
    "previously learned by the convolutional base would get modified during training. Since the `Dense` layers on top are randomly initialized, \n",
    "very large weight updates would be propagated through the network, effectively destroying the representations previously learned.\n",
    "\n",
    "In Keras, freezing a network is done by setting its `trainable` attribute to `False`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "This is the number of trainable weights before freezing the conv base: 30\n"
     ]
    }
   ],
   "source": [
    "print('This is the number of trainable weights '\n",
    "      'before freezing the conv base:', len(model.trainable_weights))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "conv_base.trainable = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "This is the number of trainable weights after freezing the conv base: 4\n"
     ]
    }
   ],
   "source": [
    "print('This is the number of trainable weights '\n",
    "      'after freezing the conv base:', len(model.trainable_weights))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "With this setup, only the weights from the two `Dense` layers that we added will be trained. That's a total of four weight tensors: two per \n",
    "layer (the main weight matrix and the bias vector). Note that in order for these changes to take effect, we must first compile the model. \n",
    "If you ever modify weight trainability after compilation, you should then re-compile the model, or these changes would be ignored.\n",
    "\n",
    "Now we can start training our model, with the same data augmentation configuration that we used in our previous example:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 2000 images belonging to 2 classes.\n",
      "Found 1000 images belonging to 2 classes.\n",
      "Epoch 1/30\n",
      "100/100 - 17s - loss: 0.3705 - acc: 0.9525 - val_loss: 0.4503 - val_acc: 0.9690\n",
      "Epoch 2/30\n",
      "100/100 - 16s - loss: 0.3659 - acc: 0.9565 - val_loss: 0.4569 - val_acc: 0.9690\n",
      "Epoch 3/30\n",
      "100/100 - 17s - loss: 0.3702 - acc: 0.9565 - val_loss: 0.4405 - val_acc: 0.9690\n",
      "Epoch 4/30\n",
      "100/100 - 17s - loss: 0.3608 - acc: 0.9535 - val_loss: 0.4236 - val_acc: 0.9680\n",
      "Epoch 5/30\n",
      "100/100 - 17s - loss: 0.2955 - acc: 0.9620 - val_loss: 0.4113 - val_acc: 0.9670\n",
      "Epoch 6/30\n",
      "100/100 - 16s - loss: 0.2599 - acc: 0.9515 - val_loss: 0.3821 - val_acc: 0.9670\n",
      "Epoch 7/30\n",
      "100/100 - 16s - loss: 0.3290 - acc: 0.9560 - val_loss: 0.4345 - val_acc: 0.9660\n",
      "Epoch 8/30\n",
      "100/100 - 16s - loss: 0.2529 - acc: 0.9565 - val_loss: 0.3603 - val_acc: 0.9690\n",
      "Epoch 9/30\n",
      "100/100 - 16s - loss: 0.2389 - acc: 0.9605 - val_loss: 0.3601 - val_acc: 0.9710\n",
      "Epoch 10/30\n",
      "100/100 - 16s - loss: 0.2285 - acc: 0.9695 - val_loss: 0.3610 - val_acc: 0.9670\n",
      "Epoch 11/30\n",
      "100/100 - 16s - loss: 0.2679 - acc: 0.9595 - val_loss: 0.3633 - val_acc: 0.9660\n",
      "Epoch 12/30\n",
      "100/100 - 16s - loss: 0.2441 - acc: 0.9565 - val_loss: 0.3850 - val_acc: 0.9650\n",
      "Epoch 13/30\n",
      "100/100 - 17s - loss: 0.2288 - acc: 0.9565 - val_loss: 0.4045 - val_acc: 0.9650\n",
      "Epoch 14/30\n",
      "100/100 - 16s - loss: 0.2184 - acc: 0.9650 - val_loss: 0.3869 - val_acc: 0.9650\n",
      "Epoch 15/30\n",
      "100/100 - 16s - loss: 0.2289 - acc: 0.9685 - val_loss: 0.3687 - val_acc: 0.9670\n",
      "Epoch 16/30\n",
      "100/100 - 16s - loss: 0.2475 - acc: 0.9615 - val_loss: 0.3564 - val_acc: 0.9680\n",
      "Epoch 17/30\n",
      "100/100 - 16s - loss: 0.2079 - acc: 0.9630 - val_loss: 0.3705 - val_acc: 0.9670\n",
      "Epoch 18/30\n",
      "100/100 - 17s - loss: 0.1981 - acc: 0.9620 - val_loss: 0.3466 - val_acc: 0.9690\n",
      "Epoch 19/30\n",
      "100/100 - 17s - loss: 0.2399 - acc: 0.9565 - val_loss: 0.3759 - val_acc: 0.9660\n",
      "Epoch 20/30\n",
      "100/100 - 16s - loss: 0.2015 - acc: 0.9590 - val_loss: 0.3648 - val_acc: 0.9680\n",
      "Epoch 21/30\n",
      "100/100 - 17s - loss: 0.1682 - acc: 0.9640 - val_loss: 0.3162 - val_acc: 0.9710\n",
      "Epoch 22/30\n",
      "100/100 - 16s - loss: 0.1570 - acc: 0.9685 - val_loss: 0.3215 - val_acc: 0.9720\n",
      "Epoch 23/30\n",
      "100/100 - 17s - loss: 0.1918 - acc: 0.9640 - val_loss: 0.3138 - val_acc: 0.9720\n",
      "Epoch 24/30\n",
      "100/100 - 17s - loss: 0.1790 - acc: 0.9640 - val_loss: 0.3023 - val_acc: 0.9700\n",
      "Epoch 25/30\n",
      "100/100 - 17s - loss: 0.1773 - acc: 0.9670 - val_loss: 0.3037 - val_acc: 0.9710\n",
      "Epoch 26/30\n",
      "100/100 - 16s - loss: 0.1650 - acc: 0.9630 - val_loss: 0.3221 - val_acc: 0.9720\n",
      "Epoch 27/30\n",
      "100/100 - 17s - loss: 0.1674 - acc: 0.9685 - val_loss: 0.3291 - val_acc: 0.9690\n",
      "Epoch 28/30\n",
      "100/100 - 16s - loss: 0.1606 - acc: 0.9680 - val_loss: 0.3086 - val_acc: 0.9710\n",
      "Epoch 29/30\n",
      "100/100 - 16s - loss: 0.1669 - acc: 0.9680 - val_loss: 0.3086 - val_acc: 0.9700\n",
      "Epoch 30/30\n",
      "100/100 - 16s - loss: 0.1857 - acc: 0.9660 - val_loss: 0.3003 - val_acc: 0.9700\n"
     ]
    }
   ],
   "source": [
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.applications.vgg16 import preprocess_input\n",
    "\n",
    "train_datagen = ImageDataGenerator(\n",
    "      preprocessing_function=preprocess_input,\n",
    "      rotation_range=40,\n",
    "      width_shift_range=0.2,\n",
    "      height_shift_range=0.2,\n",
    "      shear_range=0.2,\n",
    "      zoom_range=0.2,\n",
    "      horizontal_flip=True,\n",
    "      fill_mode='nearest')\n",
    "\n",
    "# Note that the validation data should not be augmented!\n",
    "test_datagen = ImageDataGenerator(preprocessing_function=preprocess_input)\n",
    "\n",
    "train_generator = train_datagen.flow_from_directory(\n",
    "        # This is the target directory\n",
    "        train_dir,\n",
    "        # All images will be resized to 150x150\n",
    "        target_size=(150, 150),\n",
    "        batch_size=20,\n",
    "        # Since we use binary_crossentropy loss, we need binary labels\n",
    "        class_mode='binary')\n",
    "\n",
    "validation_generator = test_datagen.flow_from_directory(\n",
    "        validation_dir,\n",
    "        target_size=(150, 150),\n",
    "        batch_size=20,\n",
    "        class_mode='binary')\n",
    "\n",
    "model.compile(loss='binary_crossentropy',\n",
    "              optimizer=optimizers.RMSprop(lr=2e-5),\n",
    "              metrics=['acc'])\n",
    "\n",
    "history = model.fit_generator(\n",
    "      train_generator,\n",
    "      steps_per_epoch=100,\n",
    "      epochs=30,\n",
    "      validation_data=validation_generator,\n",
    "      validation_steps=50,\n",
    "      verbose=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save('cats_and_dogs_small_3.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's plot our results again:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEICAYAAABS0fM3AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Z1A+gAAAACXBIWXMAAAsTAAALEwEAmpwYAAA0mklEQVR4nO3dd5hURdbA4d9hSA6gSFIkDSgKukoacQUVDCioi6AYYETQ3QUxrLqLKKKCKIrKfiaUFRcQASUYEHZRFBRzYMQBlSTqICCwSBIkM+f7o+4MzcTunu65Hc77PP1M9w11q7p77ulbdatKVBVjjDHJq5zfGTDGGOMvCwTGGJPkLBAYY0ySs0BgjDFJzgKBMcYkOQsExhiT5CwQmDwi8paI9In0tn4SkWwRuSAK6aqInOA9/5eI3BfMtmEcJ0NE3gk3n8YEQ6wfQXwTkZ0BL1OBvcBB73V/VZ1S9rmKHSKSDfxFVedFOF0FmqrqqkhtKyJpwE9ABVU9EJGMGhOE8n5nwJSOqlbNfV7cSU9EytvJxcQK+z7GFqsaSlAi0lFE1orIXSKyAZggIkeLyH9EZJOIbPWe1w/YZ4GI/MV73ldEPhaRUd62P4lIlzC3bSwiH4rIDhGZJyLPisjkIvIdTB4fFJFPvPTeEZFaAet7i8hqEdksIkOKeX/OEJENIpISsKy7iCzxnrcVkc9EZJuIrBeR0SJSsYi0XhSRhwJe3+nt84uI3JBv20tE5GsR+U1E1ojIsIDVH3p/t4nIThE5M/e9Ddi/nYgsFJHt3t92wb43Ib7PNURkgleGrSIyM2DdZSKS5ZXhBxHp7C0/rBpORIblfs4ikuZVkf1ZRH4G3vOWz/A+h+3ed+SUgP2PEJF/ep/ndu87doSI/FdEbs1XniUi0r2wspqSWSBIbMcCNYBGQD/c5z3Be90Q2A2MLmb/M4AVQC3gMWCciEgY274MfAnUBIYBvYs5ZjB57AVcD9QBKgIDAUTkZGCMl/5x3vHqUwhV/QL4HTgvX7ove88PAnd45TkTOB+4qZh84+Whs5efTkBTIH/7xO/AdUB14BJggIh089ad4/2trqpVVfWzfGnXAP4LPO2V7f+A/4pIzXxlKPDeFKKk93kSrqrxFC+tJ7w8tAVeAu70ynAOkF3EMQrTAWgOXOS9fgv3PtUBFgGBVZmjgDZAO9z3eBCQA0wErs3dSERaAPVw740Jh6raI0EeuH/IC7znHYF9QOVitm8JbA14vQBXtQTQF1gVsC4VUODYULbFnWQOAKkB6ycDk4MsU2F5vDfg9U3A297z+4GpAeuqeO/BBUWk/RAw3nteDXeSblTEtrcDbwS8VuAE7/mLwEPe8/HAyIDtTgzctpB0nwSe8J6neduWD1jfF/jYe94b+DLf/p8BfUt6b0J5n4G6uBPu0YVs93xufov7/nmvh+V+zgFla1JMHqp72xyFC1S7gRaFbFcZ2IprdwEXMJ6Lxv9UsjzsiiCxbVLVPbkvRCRVRJ73LrV/w1VFVA+sHslnQ+4TVd3lPa0a4rbHAVsClgGsKSrDQeZxQ8DzXQF5Oi4wbVX9Hdhc1LFwv/4vF5FKwOXAIlVd7eXjRK+6ZIOXj4dxVwclOSwPwOp85TtDRN73qmS2AzcGmW5u2qvzLVuN+zWcq6j35jAlvM8NcJ/Z1kJ2bQD8EGR+C5P33ohIioiM9KqXfuPQlUUt71G5sGN53+lpwLUiUg7oibuCMWGyQJDY8t8S9g/gJOAMVT2SQ1URRVX3RMJ6oIaIpAYsa1DM9qXJ4/rAtL1j1ixqY1VdijuRduHwaiFwVUzLcb86jwTuCScPuCuiQC8Ds4AGqnoU8K+AdEu6he8XXFVOoIbAuiDylV9x7/Ma3GdWvZD91gDHF5Hm77irwVzHFrJNYBl7AZfhqs+Owl015ObhV2BPMceaCGTgqux2ab5qNBMaCwTJpRrucnubV988NNoH9H5hZwLDRKSiiJwJ/ClKeXwVuFREzvIadodT8nf8ZeA23IlwRr58/AbsFJFmwIAg8zAd6CsiJ3uBKH/+q+F+be/x6tt7BazbhKuSaVJE2nOAE0Wkl4iUF5GrgZOB/wSZt/z5KPR9VtX1uLr757xG5QoikhsoxgHXi8j5IlJOROp57w9AFnCNt3060COIPOzFXbWl4q66cvOQg6tm+z8ROc67ejjTu3rDO/HnAP/ErgZKzQJBcnkSOAL3a+tz4O0yOm4GrsF1M65efhruBFCYJwkzj6r6HXAz7uS+HlePvLaE3V7BNWC+p6q/BiwfiDtJ7wBe8PIcTB7e8srwHrDK+xvoJmC4iOzAtWlMD9h3FzAC+ETc3Up/zJf2ZuBS3K/5zbjG00vz5TtYT1L8+9wb2I+7Kvofro0EVf0S1xj9BLAd+IBDVyn34X7BbwUe4PArrMK8hLsiWwcs9fIRaCDwDbAQ2AI8yuHnrJeAU3FtTqYUrEOZKXMiMg1YrqpRvyIxiUtErgP6qepZfucl3tkVgYk6ETldRI73qhI64+qFZ/qcLRPHvGq3m4CxfuclEVggMGXhWNytjTtx98APUNWvfc2RiVsichGuPWUjJVc/mSBY1ZAxxiQ5uyIwxpgkF1eDztWqVUvT0tL8zoYxxsSVr7766ldVrV3U+rgKBGlpaWRmZvqdDWOMiSsikr9H+mGsasgYY5KcBQJjjElyFgiMMSbJWSAwxpgkZ4HAGGOSnAUCY4xJchYIjDEmycVVPwJjTOzbvBnGjoXdu4PbvkcPOO206OapMJs2wZw5cO21kFLUHH1JwgKBMSZi1qyBCy+E5ctBgpjPTRXGjYPvv4fU1JK3j5TVq10+V66Eo46Cbt3K7tixyKqGjDERsXw5tG8Pv/wCCxZATk7Jj48+cts/+WTZ5fO771w+N26EGjVgks1vZoHAGFN6CxfCWWfB3r0uCHToENx+Z50FXbvCo4/Cr+HMsxaizz+Hs8+Ggwfhww/huuvgP/+BrVujf+xYZoHAGFMq8+bBuefCkUfCJ59Aq1ah7T9yJOzcCQ89FJ385Zo7F84/310FfPKJa5fo3Rv27YPp00veP5FZIDDGhO3VV+GSS6BJE3dyPeGE0NNo3hz+/Gd47jn48cfI5xFg6lT405+gaVP4+GOXX3BB6+STrXrIAoExJizPPw9XXQWnnw4ffAB164af1rBhUL48DBkSsezlefZZ6NULzjzT5fPYYw+tE3FXBZ98Er0gFA8sEBhjQqLqqnFuvBEuvhjeeQeOPrp0aR53HPzjH+6Xe6RGmld1AeaWW9zVwNtvuzuE8svIcAFh8uTIHDceWSAwxgQtJwfuuAPuu8/9kn7jjcjd9nnnnVCrFgwa5E7ipc3nrbfCAw9A377w2mtwxBGFb9ugAXTs6KqHknXmXgsExpig7N8PffrAU0+5YPDii1ChQuTSP/JIuP9+eP999+s9XPv2uaqgZ5+FgQNh/HhX7VSc3r1h1Sr44ovwjxvPLBCYkN1zDwwd6ncuTFn7y19c9cnDD8M//wnlonD26N8fjj8e7rrL3eIZqt9/d7ejTpvmbkl9/PHgOrZdcQVUrlz6RuOvvnJXF4sWlS6dMqeqcfNo06aNGn8tWqQKqkcdpbp/v9+5MWVl/XrVcuVUb789+seaNs19x158MbT9fv1V9YwzXD7//e/Qj3v11ao1a6ru3Rv6vqqqOTmqZ57p8l6tmur774eXTjQAmVrMudWuCExIBg1yf7dvd52ITHJ45RVX796vX/SPdeWV7k6ke+8NfryitWtdR7GsLNce8Oc/h37c3r3dOEnhVku98QZ89plroG7QADp3hpkzw0urrFkgMEF75x3Xeei++9zl9jvv+J0jU1YmTYI2bdw9/9EmAo895k7uzzxT8vYrVrghI9audSfxcMcNuvBCqF07vOqh/fth8GD3/gwZ4nott2rlqpzGjQsvP2XJAoEJSk6Ouxpo3Nh90dPT4d13/c6VKQvffQdff+1+MZeVjh1dR7WHH3a/0ouSmemGqdi92w1t0bFj+MesUAF69oTZs2HbttD2HTfODWA3cqRrmK5Z0/1o6tTJta089lj4+SoLFghMUKZMgcWLYcQIqFTJ/Xr6/HNXRWQS2+TJbpjmnj3L9rgjR8KOHS4YFOa999zQFlWquA5hrVuX/pi9e7vxkmbMCH6fnTtdddBZZ7n+CrmqVIFZs+Caa1zj9513xvDtqcU1IMTawxqL/bF7t2rDhqpt2qgePOiWLVjgGsXeeMPXrJkoO3hQtUED1Ysv9uf4N9ygWrGi6k8/Hb781Vfd8lNOUV27NnLHy8lRbdZM9eyzg99n2DD3v/DZZ4WvP3hQ9eab3TZ9+vhzkwUlNBYHdQIGOgMrgFXA3YWsbwTMB5YAC4D63vJzgayAxx6gm7euMfCFl+Y0oGJJ+bBA4I9Ro9w3Zf78Q8v27lWtUkX1ppv8y5eJvvfec5/9K6/4c/w1a1QrV1a99tpDy8aOdXcGtWununlz5I85YoQrc/7gU5gNG9z/wRVXFL9dTs6hgNG1q+quXRHJatBKHQiAFOAHoAlQEVgMnJxvmxlAH+/5ecCkQtKpAWwBUr3X04FrvOf/AgaUlBcLBGVvyxbVo49W7dy54LpLL1U94YSyz5MpO9df726F/P13//IweLCqiLt1+eGH3VmrSxfVnTujc7zsbHeMBx8sedubblItX151xYrg0h492pXlnHNUt20rXT5DUVIgEC2h0kpEzgSGqepF3uvBXpXSIwHbfAd0VtU1IiLAdlU9Ml86/YAOqprhbbMJOFZVD+Q/RlHS09M1M1IDkZigDBoEo0a52/LyTyf49NNw221usK7GjX3JnomiXbvcAG1XXAETJviXj+3bXSezcuXc9JK9ekW+V3N+HTrAhg3Fz7S2cqUbubR/f9eLOVivvOLmQTjlFPc/FGw5WrYsepiMkojIV6qaXuQGxUUJL0j0AP4d8Lo3MDrfNi8Dt3nPLwcUqJlvm/eAS73ntYBVAesaAN+WlBe7Iihbq1erVqrk6jULs2yZ++X0/PNlmi1TRqZOLVgl6JennnJ5+dvfDrVTRdMLL7jjffFF0dtccYVq1aqueihUb7+tmprqjhHsY9my8MtDCVcEkZqzeCAwWkT6Ah8C64C8DuIiUhc4FZgbasLelUQ/gIYNG0YiryZI993n/g4fXvj6k05yHWfeeadsOhqZsjVpEtSvX7pbMiPl1lvhggvcffrBDBlRWj16uFFLJ0+Gtm0Lrv/8c9dx7YEH4JhjQk//ootg6VJ3xRGs+vVDP07QiosSLpBwJjA34PVgYHAx21cF1uZbdhswNuC1AL8C5Qs7RlEPuyIoO1lZri7zzjuL3+6GG1SrV1c9cKDgusmTVRs1cuk0auRem/iwcaNqSorqXXf5nRP/XHmlau3aqvv2Hb48J0f1rLNUjzlGdccOf/IWKiIwxMRCoKmINBaRisA1wKzADUSklojkpjUYGJ8vjZ7AKwHBR4H3cdVOAH2AN4PIiykjd98N1au73pLFufBC1/kmf9PNlCnuKmH1andhu3q1ez1lSrRybCJp6lQ36FtZdiKLNb17uzaJufnqMWbPdrOcDRsGVav6krXIKy5K6KFf8BcDK3F3Dw3xlg0HuuqhdoTvvW3+DVQK2DcNV1VULl+aTYAvcbePzgjcp6iHXRGUjXnzXJ3kqFElb7tpk/vFP3z44csbNSq8nrNRo2jk2ERaerpqy5Z+58Jf+/ap1qqletVVh5bt36/avLnqSScVvFKIZZT2rqFYYncNRV9Ojhvw69df3RgulSuXvE96upuc5MMPDy0rV67wXpQi7hgmdi1f7uri//lP+Pvf/c6Nv269FV54ATZudLObvfCCu7J9/XXo3t3v3AWvpLuGbIgJc5hp09xY6iNGBBcEwFUPffYZ/PbboWVFtetbe3/smzTJBfKyHlIiFl17rRty4tVX3VwHQ4dCu3bhD2wXqywQmDx797pJZ1q2dPdqB6tTJzhwwA36lWvEiIJTGKamuuUmduXkuHacTp1KNxl9omjbFpo2dcHxySdh/Xo3gFxZ3LlUliwQmDxjxkB2tpvZKZTZp9q1cyf5wNFIMzJg7Fho1Mj90zRq5F5nZEQ82yaCPv7YNewncyNxIBH3XnzwATzyiLsSaN/e71xFXlK0EYwZ44bSjbSzznIjCyaC3N6brVuHN8/AJZe4OV9XrIh83oyzbRs895ybLzjcHqYl+etfXc/XjRvd6JkGfvoJmjRxI7B++y00a+Z3jkJXUhtBpDqUxbQPP4z82Pn797tu5cuXu3rDeL9UfPllN+77Qw+Ft3+nTjBnjvs12ahRZPNmnPvug9GjXZXN9ddHPv09e9zwy1dcYUEgUOPG0KeP69AVj0EgGEkRCF55peRtQnXggLt74IEH3L3GzzwTncm8y8qkSfCHP7g7hsJx4YXu77vvuok4TGStWgX/+pd7PmlSdALB7NnuyvDaayOfdrx78UW/cxBdcXzq8lf58m5WojvvdJfrvXrBvn1+5yo8q1a5u3569w7/yqZ5c6hXL7Gnr5wyBdLSXMBPSyvbznFDhrgJgW680TXKr1kT+WNMmgTHHQfnnRd+Gn6+R9GSiGUqoLhOBrH2iNUOZY895jpLXXhh/HQ5DzR0qOsUtmZN6dLp29cNWV3YcBPxbvLkgoOEpaaWzbAZX3zhjnf//aqrVrnnjzwS2WNs2uSGUx44MPw0/HyPoiVRykQkJqaJlUesBgJV1fHj3WQZZ5yh+uuvfucmeDk5qscfr3reeaVP6+WX3Tfqyy9Ln1as8aundE6OaocOqnXqqP72m1vWrp3qySe7dZEyerQrz+LF4aeRiL3JE6VMJQUCqxqKkOuvd6MRZmXB2WfD2rV+5yg4n30GP/wQmdsFL7jA/U3E6qGffw5teaTMmeNuXRw6FKpVc8t693YjV379deSOM2mSm28i/5wTofDrPYqmRCxTYSwQRFC3bvD22y4ItG8fH7dSTp7sbkW84orSp1W7NrRqlZiBwI+e0gcPuknPmzZ1t3XmuuoqqFjRnbwjYeVK+OKL0v8YSMTe5IlYpsJYIIiwjh3dL7g9e1w/g1geGmnfPjekRLduh35tllbucBM7dkQmvVjhR0/piRNd/5eHHz58FqsaNVy/jVdecXevldbkye4mgdIOKZGIvclDKVMojcox1wBdXL1RrD1iuY0gv5UrVdPS3AxG8+b5nZvCvfGGq++cMydyac6f79KcPTtyacaKspxf4fffVevVc21OhbUFvP66e5/feqt0x8nJUW3cWPWCC0qXTq5EnIMimDKF0qjsRwM01ljsn3XrVP/wB9WKFVVnzPA7NwVdfrmbXGP//siluWeP6hFHuCkFk9mrr6o2axZ+kH3kEfff+eGHha/fs8fdodWrV/h5VFV97z13nIkTS5dOsgulUdmPBmgLBD7bvNnd5SESW3P7btniAtTtt0c+7c6d3UkwWT3/vPu8K1Z0t2ROmRLa/ps2qR55pOqf/lT8dv37u6CbezdRqHJyVP/4R9XjjnNXICZ8IoWf3EVKt22klBQIrI0gymrUcL1tO3eG/v1dfa/GwPBO06e7NoJoDC524YVu6I1odHqKZaqu7rh/f+jSxQ230b69G2jvmWeCT2fECNi5E0aOLH673r1h9243Nn443njDzb37wAMF68FNaEJpVI7JBujiokSsPeLxiiDXvn2qGRku8t9+u+rBg/7mp317N9NSJO9Fz/XNN66c48ZFPu1YdfCg+1xB9dprD81etXu3ardubvl995X8fv/4o2qFCqp/+UvJx8zJUW3SRPX880PP7759qiee6PojRLJqMFlZG4EFgqAdPOjqzvOfLMraDz+4PDz8cHTSz8lRrVtX9eqro5N+rNm3z32eoHrbbQWD/P79qjfc4NYPGFB8z+uePV11z7p1wR37/vvD6xX+3HMuP7NmhbafKVooDeVl3ahugSDG5OSoPvige+cvucSfutnhw93xV6+O3jGuu061Zk3/r3yi7fff3ecIqg89VPQv/pwc1UGD3HZXXeUae/PLzHTrhwwJ/vgrV7p9Hnss+H127HA3CZx9dnSuCE3ssUAQo8aMcb8G2rdX3bq17I6bk6PatKlqx47RPc7kye7blZkZ3eP4acuWQzcCjBkT3D6PP+7el06dDh+XKifHDfNRq5bq9u2h5eOPf1Q99dTgtx82zOXh889DO46JXxYIYtj06a4++NRTVX/5pWyO+fnnWib19xs2aFSrn/yWe2twhQrucwzFhAmqKSmqbdu6O4RUXX8AUH3qqdDz8uyzbt+srJK3Xb9etUoV1R49Qj9OPEjEfgyRYIEgxr37rvvHbNxY9fvvo3+8m29WrVw59F+d4WjRIvpXHn74/nvXWbBKFff5hePNN1UrVXK32WZnq552mmv43bs39LR+/dUFpH/8o+RtBwxwt7SuXBn6cWJdoowUGg0lBYKkmKoy1i1c6G43LF8ennoKqlYteZ8qVdzgdikpwR9n3z433vwFF8DUqeHnN1iDBrkJv7duTZwZr7Ky4KKL3DhAb70V/kQ+4GbO+9Of3ITxO3e6z+Tqq8NLq1s3+PJLd8tuUd+JFSvglFPcnAajR4ed7ZiVluZu2c2vUSM3F3cyK2mqSt9/5YfySMQrglzLlqnWr3/4r5mSHldfHdovyDffdPv95z/RK0egBQvc8TIy/LtDKpL27HHVDfXru88rEr7+2g0x/cc/lq5hfcYM917PnVv0Npdf7oY82bgx/OPEMj86asULSrgiSIqpKuNBs2ZuaOHly4Pbfu5cN4ft1q1u+OtgriImT3YjhOZOKxlt55zjOtDdc4/L54wZ8d1x6dln3S/Od9+N3Ny1LVu6GeJESjfV6aWXwlFHuRFJC/t8P/vMdTx74AGoUyf848Syhg0LvyJItJFCo6K4KBFrj0S+IgjHuHHBT4azdaurk7711jLJ2mGef97ls107d6dNPNqyxY3tc9FFfuekaH/9q6sTzz9LXk6Ouzvt2GPjcwa9YFkbQdGwISYS1w03BD8Zzquvwt690RlSoiT9+rkhLTIz3VXCL7+UfR5Ka+RI2LYNHn3U75wUrXdv2LULZs48fPmsWfDJJzBsWHBXjvEqIwPGjnVtAiLu79ixbrkpnjUWJ4AFC6BrVzj6aDcpzEknFdymQwfYuBGWLQt/gvrSmj/fNWrWquXy2bSpP/kI1c8/w4knuglhXnrJ79wULScHjj/e5XXuXLfswAE49VT3+/jbb90NCSb5lNRYbFcECaBjRxcMdu8ufDKc7Gx3h0rv3v4FAYDzz4f333d3yJx1VmSnWoym++93fx980N98lKRcObj2Wpg3D9avd8vGj3ftTiNHWhAwRbNAkCBat3aX/1Wrwrnnul/fuXJnP4qFS+T0dPjoI6hU6dBsbrFsyRJ3FXDrra6qIdb17u2uDF5+GX7/3c113K4dXHaZ3zkzscwCQQJp2tQFg7Q0uPhi1y6g6u4kOecctzwWNGsGn34K9eq5e/LffNPvHBXt7rvd3TiDB/udk+CceCK0bes+8yeegA0b4PHH/b0SNOErsykti2tJjrWH3TUUnMDJcHJHO33hBb9zVdCvv7o7nsqVUx0/3u/cFJQ77ebjj/udk9A884zLd6VKbghsE58ieRcU1rM4Oe3aBT16uN6vlSq5X4bVq/udq4J27oQrrnCNx489Bnfe6XeOnJwc98t60ybXI7dyZb9zFLxNm1wP8twG4kj1eTBlK5I9pUtqLLbmowSVmuqqXAYOhJo1YzMIgGvTmD0brrvODUmxaZO7RdPvqozp0+Grr2DixPgKAuA6DQ4c6N5bCwLx6+efQ1teGnZFYGLCwYOuQXbMGLj+enf/t193uezdC82bw5FHumAQynhOxkRKWV4RWGOxiQkpKW4Ih6FDYcIEV621e7c/efnXv+Cnn9yViQUB45cRIwoOyZKa6pZHmgUCEzNEXO/XZ55xvWG7dIHt28s2D9u3u/4C559fdmMyGVOYsuwpbW0EJubccotr17juOtfX4O234ZhjyubYjz4Kmze7hmu/2ymMycgom/4/dkVgYlLPnq4ReeVK1wv5p5+if8y1a9299716uQ56xiQLCwQmZnXu7IZL2LwZ2reHb76J7vGGDnW3jT70UHSPY0ysCSoQiEhnEVkhIqtE5O5C1jcSkfkiskREFohI/YB1DUXkHRFZJiJLRSTNW/6iiPwkIlneo2WkCmUSx5lnuiEpRFzv6E8+iWz6uT03Rdy4POedB40bR/YYxsS6EgOBiKQAzwJdgJOBniJycr7NRgEvqeppwHDgkYB1LwGPq2pzoC3wv4B1d6pqS++RFX4xTCI75RQXAGrXhk6d4L//db/cS/uYNAn++tfDb9H74IModuM3JkYFc0XQFlilqj+q6j5gKpB/CKuTgfe85+/nrvcCRnlVfRdAVXeq6q6I5NwklbQ0+Phjd3//pZe62zpL+7juuoK3qO7eDUOG+FJEY3wTzF1D9YA1Aa/XAmfk22YxcDnwFNAdqCYiNYETgW0i8jrQGJgH3K2qB739RojI/cB8b/ne/AcXkX5AP4CGNudcUqtTxw1j/e9/u6EpSmvo0MKXR6PnpjGxLFK3jw4ERotIX+BDYB1w0Ev/bKAV8DMwDegLjAMGAxuAisBY4C5ctdJhVHWst5709PT46QZtouLII+Hvf49MWuPH2xy3xkBwVUPrgAYBr+t7y/Ko6i+qermqtgKGeMu24a4esrxqpQPATKC1t369NzDeXmACrgrKmDJTlj03jYllwQSChUBTEWksIhWBa4BZgRuISC0RyU1rMDA+YN/qIlLbe30esNTbp673V4BuwLelKIcxIbM5bo1xSqwaUtUDInILMBdIAcar6nciMhw3xvUsoCPwiIgormroZm/fgyIyEJjvnfC/Al7wkp7iBQgBsoAbI1oyY4JQVj03jYllNvqoMcYkOBt91BhTatGYMjHYNMtsusYkZoPOGWOKNWUK9OvnZr0Dd6dVv37uebjVasGmGY1jm4KsasgYU6xITpASaprROHYysqohY0ypRGPKxGDTLMvpGpOZBQJjTLGK6mBXmo53waYZjWObgiwQGGOKFY2Od8GmaZ3+yoYFAmNMsaLR8S7YNK3TX9mwxmJjjElw1lhsjDGmWBYIjDEmyVkgMMZH1mvWH/a+H856FhvjE+s16w973wuyxmJjfGK9Zv2RjO+7NRYbE6Os16w/7H0vyAKBMT6xXrP+sPe9IAsExvjEes36w973giwQGOMT6zXrD3vfC7LGYmOMSXDWWGyMMaZYFgiMMSbJWSCIEdbT0RTHvh8mmqxncQywno6mOPb9MNFmjcUxIBl7Oprg2ffDlJY1FscB6+loimPfDxNtFghigPV0TCyRrs+374eJNgsEMcB6OiaO3Pr81atB9VB9fmmCgX0/TLRZIIgB1tMxcQwZcqhRN9euXW55uOz7YaLNGouNiaBy5dyVQH4ikJNT9vkxBqyx2JgyZfX5Jh5ZIDAmgqw+38QjCwTGRJDV55t4ZD2LjYmwjAw78Zv4YlcExhiT5CwQGGNMkrNAYEJio2Aak3isjcAEzUbBNCYx2RWBCVo0es0aY/xngcAEzUbBNCYxWSAwQbNes8YkpqACgYh0FpEVIrJKRO4uZH0jEZkvIktEZIGI1A9Y11BE3hGRZSKyVETSvOWNReQLL81pIlIxYqUyUWG9Zo1JTCUGAhFJAZ4FugAnAz1F5OR8m40CXlLV04DhwCMB614CHlfV5kBb4H/e8keBJ1T1BGAr8OfSFMREn/WaNSYxBXNF0BZYpao/quo+YCpwWb5tTgbe856/n7veCxjlVfVdAFXdqaq7RESA84BXvX0mAt1KUxBTNjIy3PSIOTnurwUBY+JfMIGgHrAm4PVab1mgxcDl3vPuQDURqQmcCGwTkddF5GsRedy7wqgJbFPVA8WkCYCI9BORTBHJ3LRpU3ClMsYYE7RINRYPBDqIyNdAB2AdcBDXT+Fsb/3pQBOgbygJq+pYVU1X1fTatWtHKLvGGGNyBRMI1gENAl7X95blUdVfVPVyVW0FDPGWbcP90s/yqpUOADOB1sBmoLqIlC8qTZM8rLeyMf4KJhAsBJp6d/lUBK4BZgVuICK1RCQ3rcHA+IB9q4tI7k/584Cl6qZFex/o4S3vA7wZfjFMvIrGHL/GmNCUGAi8X/K3AHOBZcB0Vf1ORIaLSFdvs47AChFZCRwDjPD2PYirFpovIt8AArzg7XMX8HcRWYVrMxgXsVKZuGG9lY3xn81ZbHxlc/waE302Z7GJadZb2Rj/WSAwvrLeyv6xRnqTywKB8ZX1VvaHNdKbQNZGYEwSSktzJ//8GjVyPcZNYrE2AmNMATakuAlkgcCYJGSN9CaQBQJjkpA10ptAFgiMSULWSG8C2eT1xiSpjAw78RvHrgiMMSbJWSAwxpgkZ4HAGGOSnAUCY4xJchYIjDEmyVkgMMaYJGeBwBhjkpwFAmOMSXIWCIwxJslZIDDGmCRngcAYY5KcBQJjjElyFgiMMSbJWSAwxpgkZ4HAGGOSnAUCY4xJchYIAkyZAmlpUK6c+ztlSum2iyfxUKZQ8hgP5YmWZC67CZOqxs2jTZs2Gi2TJ6umpqrCoUdqqlseznbxJB7KFEoe46E80ZLMZTdFAzK1mHOruG3iQ3p6umZmZkYl7bQ0WL264PJGjSA7O/Tt4kk8lCmUPMZDeaIlmctuiiYiX6lqepHrLRA45cq530/5iUBOTujbxZN4KFMoeYyH8kRLMpfdFK2kQGBtBJ6GDYNbHux28SQeyhRKHuOhPNGSzGU34bNA4BkxAlJTD1+WmuqWh7NdLAi20TAeyhRKHuOhPNGSzGU3pVBcA0KsPaLZWKzqGtQaNVIVcX+LamALdjs/hdpoGC9lCjaP8VCeaEnmspvCYY3FyckaDY0xuayNIEn9/HNoy40xycsCQYKyRkNjTLAsECQoazQ0xgTLAkGCysiAsWNdm4CI+zt2rFtujDGByvudARM9GRl24jfGlMyuCIwxJskFFQhEpLOIrBCRVSJydyHrG4nIfBFZIiILRKR+wLqDIpLlPWYFLH9RRH4KWNcyIiUyxhgTkhKrhkQkBXgW6ASsBRaKyCxVXRqw2SjgJVWdKCLnAY8Avb11u1W1ZRHJ36mqr4ade2OMMaUWzBVBW2CVqv6oqvuAqcBl+bY5GXjPe/5+IeuNMcbEqGACQT1gTcDrtd6yQIuBy73n3YFqIlLTe11ZRDJF5HMR6ZZvvxFeddITIlKpsIOLSD9v/8xNmzYFkV1jjDGhiFRj8UCgg4h8DXQA1gEHvXWNvK7NvYAnReR4b/lgoBlwOlADuKuwhFV1rKqmq2p67dq1I5RdY4wxuYIJBOuABgGv63vL8qjqL6p6uaq2AoZ4y7Z5f9d5f38EFgCtvNfrvfGQ9gITcFVQxhhjylgwgWAh0FREGotIReAaYFbgBiJSS0Ry0xoMjPeWH51b5SMitYD2wFLvdV3vrwDdgG9LXRpjjDEhK/GuIVU9ICK3AHOBFGC8qn4nIsNxQ5vOAjoCj4iIAh8CN3u7NweeF5EcXNAZGXC30RQRqQ0IkAXcGLliGWOMCZYNQ22MMQmupGGobYgJY0xQ9u/fz9q1a9mzZ4/fWTFFqFy5MvXr16dChQoh7WeBwBgTlLVr11KtWjXS0tJwTXsmlqgqmzdvZu3atTRu3DikfW2sIWNMUPbs2UPNmjUtCMQoEaFmzZphXbFZIDDGBM2CQGwL9/OxQGCMMUnOAoExJiqmTIG0NChXzv2dMqV06W3evJmWLVvSsmVLjj32WOrVq5f3et++fcXum5mZyd/+9rcSj9GuXbvSZTJOWWNxlE2ZAkOGuEnjGzZ0U0XaZDEm0U2ZAv36wa5d7vXq1e41hP/9r1mzJllZWQAMGzaMqlWrMnDgwLz1Bw4coHz5wk9p6enppKcXefdknk8//TS8zMU5uyKIotx/htWrQfXQP0NpfxkZE+uGDDkUBHLt2uWWR1Lfvn258cYbOeOMMxg0aBBffvklZ555Jq1ataJdu3asWLECgAULFnDppZcCLojccMMNdOzYkSZNmvD000/npVe1atW87Tt27EiPHj1o1qwZGRkZ5Pa5mjNnDs2aNaNNmzb87W9/y0s3UHZ2NmeffTatW7emdevWhwWYRx99lFNPPZUWLVpw991uepdVq1ZxwQUX0KJFC1q3bs0PP/wQ2TeqBHZFEEXF/TPYVYFJZD//HNry0li7di2ffvopKSkp/Pbbb3z00UeUL1+eefPmcc899/Daa68V2Gf58uW8//777Nixg5NOOokBAwYUuPf+66+/5rvvvuO4446jffv2fPLJJ6Snp9O/f38+/PBDGjduTM+ePQvNU506dXj33XepXLky33//PT179iQzM5O33nqLN998ky+++ILU1FS2bNkCQEZGBnfffTfdu3dnz5495OTkRP6NKoYFgigqy38GY2JJw4buCriw5ZF25ZVXkpKSAsD27dvp06cP33//PSLC/v37C93nkksuoVKlSlSqVIk6deqwceNG6tevf9g2bdu2zVvWsmVLsrOzqVq1Kk2aNMm7T79nz56MHTu2QPr79+/nlltuISsri5SUFFauXAnAvHnzuP7660lNTQWgRo0a7Nixg3Xr1tG9e3fAdQora1Y1FEVFfemj8c9gTCwZMQK8c12e1FS3PNKqVKmS9/y+++7j3HPP5dtvv2X27NlF3lNfqdKh6U9SUlI4cOBAWNsU5YknnuCYY45h8eLFZGZmltiY7TcLBFFUlv8MxsSSjAwYOxYaNQIR93fs2OhXiW7fvp169dy8WS+++GLE0z/ppJP48ccfyc7OBmDatGlF5qNu3bqUK1eOSZMmcfCgm56lU6dOTJgwgV1enfGWLVuoVq0a9evXZ+bMmQDs3bs3b31ZsUAQRX79MxgTCzIyIDsbcnLc37L43g8aNIjBgwfTqlWrkH7BB+uII47gueeeo3PnzrRp04Zq1apx1FFHFdjupptuYuLEibRo0YLly5fnXbV07tyZrl27kp6eTsuWLRk1ahQAkyZN4umnn+a0006jXbt2bNiwIeJ5L46NPmqMCcqyZcto3ry539nw3c6dO6latSqqys0330zTpk254447/M5WnsI+p5JGH7UrAmOMCcELL7xAy5YtOeWUU9i+fTv9+/f3O0ullvCBINK9G40xye2OO+4gKyuLpUuXMmXKlLw7gOJZQt8+Go3ejcYYk2gS+oqgrHo3GmNMPEvoQGAduowxpmQJHQisQ5cxxpQsoQOBdegyJnGce+65zJ0797BlTz75JAMGDChyn44dO5J7y/nFF1/Mtm3bCmwzbNiwvPv5izJz5kyWLl2a9/r+++9n3rx5IeQ+tiV0ILAOXcYkjp49ezJ16tTDlk2dOrXIgd/ymzNnDtWrVw/r2PkDwfDhw7ngggvCSisWJfRdQ+BO+nbiNyaybr8dvKkBIqZlS3jyyaLX9+jRg3vvvZd9+/ZRsWJFsrOz+eWXXzj77LMZMGAACxcuZPfu3fTo0YMHHnigwP5paWlkZmZSq1YtRowYwcSJE6lTpw4NGjSgTZs2gOsjMHbsWPbt28cJJ5zApEmTyMrKYtasWXzwwQc89NBDvPbaazz44INceuml9OjRg/nz5zNw4EAOHDjA6aefzpgxY6hUqRJpaWn06dOH2bNns3//fmbMmEGzZs0Oy1N2dja9e/fm999/B2D06NF5k+M8+uijTJ48mXLlytGlSxdGjhzJqlWruPHGG9m0aRMpKSnMmDGD448/vtTvfUJfERhjEkeNGjVo27Ytb731FuCuBq666ipEhBEjRpCZmcmSJUv44IMPWLJkSZHpfPXVV0ydOpWsrCzmzJnDwoUL89ZdfvnlLFy4kMWLF9O8eXPGjRtHu3bt6Nq1K48//jhZWVmHnXj37NlD3759mTZtGt988w0HDhxgzJgxeetr1arFokWLGDBgQKHVT7nDVS9atIhp06blzaIWOFz14sWLGTRoEOCGq7755ptZvHgxn376KXXr1i3dm+pJ+CsCY0zkFffLPZpyq4cuu+wypk6dyrhx4wCYPn06Y8eO5cCBA6xfv56lS5dy2mmnFZrGRx99RPfu3fM6gnXt2jVv3bfffsu9997Ltm3b2LlzJxdddFGx+VmxYgWNGzfmxBNPBKBPnz48++yz3H777YALLABt2rTh9ddfL7B/rAxXbVcEcch6S5tkddlllzF//nwWLVrErl27aNOmDT/99BOjRo1i/vz5LFmyhEsuuaTI4adL0rdvX0aPHs0333zD0KFDw04nV+5Q1kUNYx0rw1VbIIgzNv2lSWZVq1bl3HPP5YYbbshrJP7tt9+oUqUKRx11FBs3bsyrOirKOeecw8yZM9m9ezc7duxg9uzZeet27NhB3bp12b9/P1MC/qmqVavGjh07CqR10kknkZ2dzapVqwA3imiHDh2CLk+sDFdtgSDOWG9pk+x69uzJ4sWL8wJBixYtaNWqFc2aNaNXr160b9++2P1bt27N1VdfTYsWLejSpQunn3563roHH3yQM844g/bt2x/WsHvNNdfw+OOP06pVq8PmE65cuTITJkzgyiuv5NRTT6VcuXLceOONQZclVoartmGo40y5cu5KID8RN+67MdFiw1DHBxuGOglYb2ljTKRZIIgz1lvaGBNpFgjijPWWNn6Kp6rkZBTu52P9COKQ9ZY2fqhcuTKbN2+mZs2aiIjf2TH5qCqbN28Oq3+BBQJjTFDq16/P2rVr2bRpk99ZMUWoXLky9evXD3k/CwTGmKBUqFCBxo0b+50NEwXWRmCMMUnOAoExxiQ5CwTGGJPk4qpnsYhsAlaHuXst4NcIZicWJFqZrDyxL9HKlGjlgcLL1EhVaxe1Q1wFgtIQkcziuljHo0Qrk5Un9iVamRKtPBBemaxqyBhjkpwFAmOMSXLJFAjG+p2BKEi0Mll5Yl+ilSnRygNhlClp2giMMcYULpmuCIwxxhTCAoExxiS5pAgEItJZRFaIyCoRudvv/JSWiGSLyDcikiUicTllm4iMF5H/ici3ActqiMi7IvK99/doP/MYiiLKM0xE1nmfU5aIXOxnHkMhIg1E5H0RWSoi34nIbd7yeP6MiipTXH5OIlJZRL4UkcVeeR7wljcWkS+88900EalYYlqJ3kYgIinASqATsBZYCPRU1aW+ZqwURCQbSFfVuO0IIyLnADuBl1T1D96yx4AtqjrSC9hHq+pdfuYzWEWUZxiwU1VH+Zm3cIhIXaCuqi4SkWrAV0A3oC/x+xkVVaariMPPSdxY4FVUdaeIVAA+Bm4D/g68rqpTReRfwGJVHVNcWslwRdAWWKWqP6rqPmAqcJnPeUp6qvohsCXf4suAid7zibh/0rhQRHnilqquV9VF3vMdwDKgHvH9GRVVprikzk7vZQXvocB5wKve8qA+o2QIBPWANQGv1xLHH75HgXdE5CsR6ed3ZiLoGFVd7z3fABzjZ2Yi5BYRWeJVHcVNNUogEUkDWgFfkCCfUb4yQZx+TiKSIiJZwP+Ad4EfgG2qesDbJKjzXTIEgkR0lqq2BroAN3vVEglFXZ1lvNdbjgGOB1oC64F/+pqbMIhIVeA14HZV/S1wXbx+RoWUKW4/J1U9qKotgfq42o9m4aSTDIFgHdAg4HV9b1ncUtV13t//AW/gvgCJYKNXj5tbn/s/n/NTKqq60ftHzQFeIM4+J6/e+TVgiqq+7i2O68+osDLF++cEoKrbgPeBM4HqIpI76VhQ57tkCAQLgaZeS3pF4Bpgls95CpuIVPEauhCRKsCFwLfF7xU3ZgF9vOd9gDd9zEup5Z4wPd2Jo8/Ja4gcByxT1f8LWBW3n1FRZYrXz0lEaotIde/5EbgbYpbhAkIPb7OgPqOEv2sIwLsd7EkgBRivqiP8zVH4RKQJ7ioA3FSjL8djeUTkFaAjbsjcjcBQYCYwHWiIG278KlWNiwbYIsrTEVfdoEA20D+gfj2michZwEfAN0COt/geXJ16vH5GRZWpJ3H4OYnIabjG4BTcj/rpqjrcO0dMBWoAXwPXqureYtNKhkBgjDGmaMlQNWSMMaYYFgiMMSbJWSAwxpgkZ4HAGGOSnAUCY4xJchYIjDEmyVkgMMaYJPf/1PRAa6dmluUAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "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(len(acc))\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 accuracy')\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": [
    "As you can see, we reach a validation accuracy of about 96%. This is much better than our small convnet trained from scratch."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Fine-tuning\n",
    "\n",
    "Another widely used technique for model reuse, complementary to feature extraction, is _fine-tuning_. \n",
    "Fine-tuning consists in unfreezing a few of the top layers \n",
    "of a frozen model base used for feature extraction, and jointly training both the newly added part of the model (in our case, the \n",
    "fully-connected classifier) and these top layers. This is called \"fine-tuning\" because it slightly adjusts the more abstract \n",
    "representations of the model being reused, in order to make them more relevant for the problem at hand.\n",
    "\n",
    "![fine-tuning VGG16](https://s3.amazonaws.com/book.keras.io/img/ch5/vgg16_fine_tuning.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We have stated before that it was necessary to freeze the convolution base of VGG16 in order to be able to train a randomly initialized \n",
    "classifier on top. For the same reason, it is only possible to fine-tune the top layers of the convolutional base once the classifier on \n",
    "top has already been trained. If the classified wasn't already trained, then the error signal propagating through the network during \n",
    "training would be too large, and the representations previously learned by the layers being fine-tuned would be destroyed. Thus the steps \n",
    "for fine-tuning a network are as follow:\n",
    "\n",
    "* 1) Add your custom network on top of an already trained base network.\n",
    "* 2) Freeze the base network.\n",
    "* 3) Train the part you added.\n",
    "* 4) Unfreeze some layers in the base network.\n",
    "* 5) Jointly train both these layers and the part you added.\n",
    "\n",
    "We have already completed the first 3 steps when doing feature extraction. Let's proceed with the 4th step: we will unfreeze our `conv_base`, \n",
    "and then freeze individual layers inside of it.\n",
    "\n",
    "As a reminder, this is what our convolutional base looks like:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"vgg16\"\n",
      "_________________________________________________________________\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: 0\n",
      "Non-trainable params: 14,714,688\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "conv_base.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "We will fine-tune the last 3 convolutional layers, which means that all layers up until `block4_pool` should be frozen, and the layers \n",
    "`block5_conv1`, `block5_conv2` and `block5_conv3` should be trainable.\n",
    "\n",
    "Why not fine-tune more layers? Why not fine-tune the entire convolutional base? We could. However, we need to consider that:\n",
    "\n",
    "* Earlier layers in the convolutional base encode more generic, reusable features, while layers higher up encode more specialized features. It is \n",
    "more useful to fine-tune the more specialized features, as these are the ones that need to be repurposed on our new problem. There would \n",
    "be fast-decreasing returns in fine-tuning lower layers.\n",
    "* The more parameters we are training, the more we are at risk of overfitting. The convolutional base has 15M parameters, so it would be \n",
    "risky to attempt to train it on our small dataset.\n",
    "\n",
    "Thus, in our situation, it is a good strategy to only fine-tune the top 2 to 3 layers in the convolutional base.\n",
    "\n",
    "Let's set this up, starting from where we left off in the previous example:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "conv_base.trainable = True\n",
    "\n",
    "set_trainable = False\n",
    "for layer in conv_base.layers:\n",
    "    if layer.name == 'block5_conv1':\n",
    "        set_trainable = True\n",
    "    if set_trainable:\n",
    "        layer.trainable = True\n",
    "    else:\n",
    "        layer.trainable = False"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we can start fine-tuning our network. We will do this with the RMSprop optimizer, using a very low learning rate. The reason for using \n",
    "a low learning rate is that we want to limit the magnitude of the modifications we make to the representations of the 3 layers that we are \n",
    "fine-tuning. Updates that are too large may harm these representations.\n",
    "\n",
    "Now let's proceed with fine-tuning:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "100/100 [==============================] - 18s 170ms/step - loss: 0.0094 - acc: 0.9957 - val_loss: 0.5007 - val_acc: 0.9790\n",
      "Epoch 2/100\n",
      "100/100 [==============================] - 17s 167ms/step - loss: 0.0144 - acc: 0.9980 - val_loss: 0.5290 - val_acc: 0.9760\n",
      "Epoch 3/100\n",
      "100/100 [==============================] - 17s 170ms/step - loss: 0.0366 - acc: 0.9927 - val_loss: 0.9081 - val_acc: 0.9630\n",
      "Epoch 4/100\n",
      "100/100 [==============================] - 17s 169ms/step - loss: 0.0151 - acc: 0.9955 - val_loss: 0.5252 - val_acc: 0.9780\n",
      "Epoch 5/100\n",
      "100/100 [==============================] - 17s 169ms/step - loss: 0.0061 - acc: 0.9984 - val_loss: 0.5423 - val_acc: 0.9760\n",
      "Epoch 6/100\n",
      "100/100 [==============================] - 18s 175ms/step - loss: 0.0024 - acc: 0.9994 - val_loss: 0.5207 - val_acc: 0.9770\n",
      "Epoch 7/100\n",
      "100/100 [==============================] - 17s 169ms/step - loss: 0.0267 - acc: 0.9955 - val_loss: 0.5045 - val_acc: 0.9780\n",
      "Epoch 8/100\n",
      "100/100 [==============================] - 17s 171ms/step - loss: 0.1261 - acc: 0.9916 - val_loss: 0.4974 - val_acc: 0.9780\n",
      "Epoch 9/100\n",
      "100/100 [==============================] - 17s 168ms/step - loss: 0.0234 - acc: 0.9942 - val_loss: 0.4695 - val_acc: 0.9800\n",
      "Epoch 10/100\n",
      "100/100 [==============================] - 17s 172ms/step - loss: 0.0196 - acc: 0.9942 - val_loss: 0.4407 - val_acc: 0.9790\n",
      "Epoch 11/100\n",
      "100/100 [==============================] - 17s 172ms/step - loss: 0.0700 - acc: 0.9929 - val_loss: 0.5519 - val_acc: 0.9750\n",
      "Epoch 12/100\n",
      "100/100 [==============================] - 17s 172ms/step - loss: 0.0343 - acc: 0.9914 - val_loss: 0.4135 - val_acc: 0.9780\n",
      "Epoch 13/100\n",
      "100/100 [==============================] - 17s 172ms/step - loss: 0.0037 - acc: 0.9992 - val_loss: 0.6950 - val_acc: 0.9710\n",
      "Epoch 14/100\n",
      "100/100 [==============================] - 17s 170ms/step - loss: 0.0432 - acc: 0.9920 - val_loss: 0.4164 - val_acc: 0.9760\n",
      "Epoch 15/100\n",
      "100/100 [==============================] - 17s 171ms/step - loss: 0.0384 - acc: 0.9973 - val_loss: 0.4002 - val_acc: 0.9780\n",
      "Epoch 16/100\n",
      "100/100 [==============================] - 17s 175ms/step - loss: 0.0316 - acc: 0.9939 - val_loss: 0.4920 - val_acc: 0.9710\n",
      "Epoch 17/100\n",
      "100/100 [==============================] - 17s 169ms/step - loss: 0.0143 - acc: 0.9965 - val_loss: 0.4999 - val_acc: 0.9710\n",
      "Epoch 18/100\n",
      "100/100 [==============================] - 17s 170ms/step - loss: 0.0090 - acc: 0.9983 - val_loss: 0.4266 - val_acc: 0.9720\n",
      "Epoch 19/100\n",
      "100/100 [==============================] - 17s 171ms/step - loss: 0.0046 - acc: 0.9994 - val_loss: 0.5191 - val_acc: 0.9760\n",
      "Epoch 20/100\n",
      "100/100 [==============================] - 17s 172ms/step - loss: 0.0114 - acc: 0.9980 - val_loss: 0.5128 - val_acc: 0.9730\n",
      "Epoch 21/100\n",
      "100/100 [==============================] - 17s 168ms/step - loss: 0.0271 - acc: 0.9937 - val_loss: 0.5888 - val_acc: 0.9760\n",
      "Epoch 22/100\n",
      "100/100 [==============================] - 17s 169ms/step - loss: 0.0180 - acc: 0.9968 - val_loss: 0.6151 - val_acc: 0.9760\n",
      "Epoch 23/100\n",
      "100/100 [==============================] - 17s 167ms/step - loss: 0.0097 - acc: 0.9984 - val_loss: 0.6704 - val_acc: 0.9740\n",
      "Epoch 24/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0024 - acc: 0.9984 - val_loss: 0.5848 - val_acc: 0.9760\n",
      "Epoch 25/100\n",
      "100/100 [==============================] - 17s 167ms/step - loss: 0.0176 - acc: 0.9942 - val_loss: 0.5945 - val_acc: 0.9770\n",
      "Epoch 26/100\n",
      "100/100 [==============================] - 18s 175ms/step - loss: 0.0087 - acc: 0.9980 - val_loss: 0.6640 - val_acc: 0.9750\n",
      "Epoch 27/100\n",
      "100/100 [==============================] - 17s 171ms/step - loss: 0.0092 - acc: 0.9972 - val_loss: 0.5794 - val_acc: 0.9750\n",
      "Epoch 28/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0131 - acc: 0.9970 - val_loss: 0.6230 - val_acc: 0.9770\n",
      "Epoch 29/100\n",
      "100/100 [==============================] - 17s 165ms/step - loss: 0.0291 - acc: 0.9952 - val_loss: 0.6221 - val_acc: 0.9750\n",
      "Epoch 30/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0122 - acc: 0.9967 - val_loss: 0.5702 - val_acc: 0.9780\n",
      "Epoch 31/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0293 - acc: 0.9964 - val_loss: 0.5677 - val_acc: 0.9780\n",
      "Epoch 32/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0360 - acc: 0.9952 - val_loss: 0.5433 - val_acc: 0.9760\n",
      "Epoch 33/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0500 - acc: 0.9973 - val_loss: 0.6736 - val_acc: 0.9730\n",
      "Epoch 34/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0183 - acc: 0.9961 - val_loss: 0.6040 - val_acc: 0.9750\n",
      "Epoch 35/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0133 - acc: 0.9959 - val_loss: 1.3636 - val_acc: 0.9490\n",
      "Epoch 36/100\n",
      "100/100 [==============================] - 17s 165ms/step - loss: 0.0424 - acc: 0.9933 - val_loss: 0.6527 - val_acc: 0.9760\n",
      "Epoch 37/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0090 - acc: 0.9977 - val_loss: 0.5502 - val_acc: 0.9770\n",
      "Epoch 38/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0156 - acc: 0.9969 - val_loss: 0.5218 - val_acc: 0.9750\n",
      "Epoch 39/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0188 - acc: 0.9965 - val_loss: 0.5702 - val_acc: 0.9760\n",
      "Epoch 40/100\n",
      "100/100 [==============================] - 17s 174ms/step - loss: 0.0068 - acc: 0.9989 - val_loss: 0.5084 - val_acc: 0.9750\n",
      "Epoch 41/100\n",
      "100/100 [==============================] - 18s 184ms/step - loss: 0.0100 - acc: 0.9965 - val_loss: 0.5620 - val_acc: 0.9770\n",
      "Epoch 42/100\n",
      "100/100 [==============================] - 18s 179ms/step - loss: 0.0368 - acc: 0.9945 - val_loss: 0.5341 - val_acc: 0.9770\n",
      "Epoch 43/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0142 - acc: 0.9980 - val_loss: 0.7765 - val_acc: 0.9720\n",
      "Epoch 44/100\n",
      "100/100 [==============================] - 18s 178ms/step - loss: 0.0065 - acc: 0.9973 - val_loss: 0.5991 - val_acc: 0.9730\n",
      "Epoch 45/100\n",
      "100/100 [==============================] - 18s 182ms/step - loss: 0.0246 - acc: 0.9958 - val_loss: 0.6311 - val_acc: 0.9730\n",
      "Epoch 46/100\n",
      "100/100 [==============================] - 18s 180ms/step - loss: 0.0382 - acc: 0.9947 - val_loss: 0.6475 - val_acc: 0.9730\n",
      "Epoch 47/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0190 - acc: 0.9973 - val_loss: 0.5348 - val_acc: 0.9730\n",
      "Epoch 48/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0086 - acc: 0.9963 - val_loss: 0.5177 - val_acc: 0.9740\n",
      "Epoch 49/100\n",
      "100/100 [==============================] - 18s 179ms/step - loss: 0.0155 - acc: 0.9971 - val_loss: 0.6711 - val_acc: 0.9760\n",
      "Epoch 50/100\n",
      "100/100 [==============================] - 18s 180ms/step - loss: 0.0084 - acc: 0.9978 - val_loss: 0.7135 - val_acc: 0.9750\n",
      "Epoch 51/100\n",
      "100/100 [==============================] - 18s 179ms/step - loss: 0.0016 - acc: 0.9995 - val_loss: 0.6352 - val_acc: 0.9760\n",
      "Epoch 52/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0056 - acc: 0.9990 - val_loss: 0.6451 - val_acc: 0.9750\n",
      "Epoch 53/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0118 - acc: 0.9946 - val_loss: 0.5873 - val_acc: 0.9770\n",
      "Epoch 54/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0061 - acc: 0.9989 - val_loss: 0.6410 - val_acc: 0.9760\n",
      "Epoch 55/100\n",
      "100/100 [==============================] - 18s 179ms/step - loss: 0.0143 - acc: 0.9981 - val_loss: 0.6257 - val_acc: 0.9710\n",
      "Epoch 56/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0099 - acc: 0.9978 - val_loss: 0.6326 - val_acc: 0.9760\n",
      "Epoch 57/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0433 - acc: 0.9971 - val_loss: 0.6160 - val_acc: 0.9740\n",
      "Epoch 58/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0075 - acc: 0.9982 - val_loss: 0.5284 - val_acc: 0.9730\n",
      "Epoch 59/100\n",
      "100/100 [==============================] - 18s 182ms/step - loss: 0.0041 - acc: 0.9986 - val_loss: 0.5146 - val_acc: 0.9740\n",
      "Epoch 60/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0390 - acc: 0.9939 - val_loss: 0.6144 - val_acc: 0.9740\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 61/100\n",
      "100/100 [==============================] - 18s 178ms/step - loss: 0.0187 - acc: 0.9964 - val_loss: 0.5615 - val_acc: 0.9730\n",
      "Epoch 62/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0141 - acc: 0.9966 - val_loss: 0.5332 - val_acc: 0.9770\n",
      "Epoch 63/100\n",
      "100/100 [==============================] - 18s 178ms/step - loss: 0.0139 - acc: 0.9969 - val_loss: 0.5417 - val_acc: 0.9750\n",
      "Epoch 64/100\n",
      "100/100 [==============================] - 18s 176ms/step - loss: 0.0303 - acc: 0.9958 - val_loss: 0.5546 - val_acc: 0.9710\n",
      "Epoch 65/100\n",
      "100/100 [==============================] - 18s 179ms/step - loss: 0.0212 - acc: 0.9971 - val_loss: 0.7562 - val_acc: 0.9710\n",
      "Epoch 66/100\n",
      "100/100 [==============================] - 18s 178ms/step - loss: 0.0073 - acc: 0.9984 - val_loss: 0.4690 - val_acc: 0.9770\n",
      "Epoch 67/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0066 - acc: 0.9991 - val_loss: 0.5484 - val_acc: 0.9730\n",
      "Epoch 68/100\n",
      "100/100 [==============================] - 18s 178ms/step - loss: 0.0197 - acc: 0.9992 - val_loss: 0.8517 - val_acc: 0.9680\n",
      "Epoch 69/100\n",
      "100/100 [==============================] - 18s 176ms/step - loss: 0.0066 - acc: 0.9979 - val_loss: 0.5795 - val_acc: 0.9750\n",
      "Epoch 70/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0065 - acc: 0.9990 - val_loss: 0.6076 - val_acc: 0.9750\n",
      "Epoch 71/100\n",
      "100/100 [==============================] - 18s 176ms/step - loss: 0.0170 - acc: 0.9972 - val_loss: 0.5691 - val_acc: 0.9760\n",
      "Epoch 72/100\n",
      "100/100 [==============================] - 18s 178ms/step - loss: 0.0273 - acc: 0.9972 - val_loss: 0.5898 - val_acc: 0.9740\n",
      "Epoch 73/100\n",
      "100/100 [==============================] - 18s 181ms/step - loss: 0.0068 - acc: 0.9965 - val_loss: 0.5682 - val_acc: 0.9730\n",
      "Epoch 74/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0018 - acc: 0.9993 - val_loss: 0.5890 - val_acc: 0.9780\n",
      "Epoch 75/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0053 - acc: 0.9988 - val_loss: 0.4936 - val_acc: 0.9800\n",
      "Epoch 76/100\n",
      "100/100 [==============================] - 18s 178ms/step - loss: 0.0374 - acc: 0.9963 - val_loss: 0.8526 - val_acc: 0.9690\n",
      "Epoch 77/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0215 - acc: 0.9964 - val_loss: 0.6361 - val_acc: 0.9770\n",
      "Epoch 78/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0235 - acc: 0.9982 - val_loss: 0.7158 - val_acc: 0.9740\n",
      "Epoch 79/100\n",
      "100/100 [==============================] - 18s 178ms/step - loss: 0.0162 - acc: 0.9963 - val_loss: 0.5156 - val_acc: 0.9760\n",
      "Epoch 80/100\n",
      "100/100 [==============================] - 18s 180ms/step - loss: 0.0037 - acc: 0.9994 - val_loss: 0.6266 - val_acc: 0.9720\n",
      "Epoch 81/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0071 - acc: 0.9981 - val_loss: 0.8062 - val_acc: 0.9690\n",
      "Epoch 82/100\n",
      "100/100 [==============================] - 18s 177ms/step - loss: 0.0129 - acc: 0.9962 - val_loss: 0.6886 - val_acc: 0.9740\n",
      "Epoch 83/100\n",
      "100/100 [==============================] - 18s 180ms/step - loss: 0.0041 - acc: 0.9987 - val_loss: 0.8581 - val_acc: 0.9700\n",
      "Epoch 84/100\n",
      "100/100 [==============================] - 18s 180ms/step - loss: 0.0369 - acc: 0.9933 - val_loss: 0.7598 - val_acc: 0.9720\n",
      "Epoch 85/100\n",
      "100/100 [==============================] - 18s 179ms/step - loss: 0.0075 - acc: 0.9976 - val_loss: 0.7141 - val_acc: 0.9730\n",
      "Epoch 86/100\n",
      "100/100 [==============================] - 18s 179ms/step - loss: 0.0017 - acc: 0.9994 - val_loss: 0.6791 - val_acc: 0.9750\n",
      "Epoch 87/100\n",
      "100/100 [==============================] - 19s 187ms/step - loss: 0.0120 - acc: 0.9981 - val_loss: 0.7272 - val_acc: 0.9750\n",
      "Epoch 88/100\n",
      "100/100 [==============================] - 18s 176ms/step - loss: 0.0160 - acc: 0.9947 - val_loss: 0.6133 - val_acc: 0.9760\n",
      "Epoch 89/100\n",
      "100/100 [==============================] - 17s 167ms/step - loss: 0.0227 - acc: 0.9984 - val_loss: 0.7683 - val_acc: 0.9720\n",
      "Epoch 90/100\n",
      "100/100 [==============================] - 17s 168ms/step - loss: 0.0174 - acc: 0.9977 - val_loss: 0.8184 - val_acc: 0.9710\n",
      "Epoch 91/100\n",
      "100/100 [==============================] - 17s 167ms/step - loss: 0.0105 - acc: 0.9969 - val_loss: 0.6377 - val_acc: 0.9730\n",
      "Epoch 92/100\n",
      "100/100 [==============================] - 17s 169ms/step - loss: 0.0093 - acc: 0.9977 - val_loss: 0.5976 - val_acc: 0.9740\n",
      "Epoch 93/100\n",
      "100/100 [==============================] - 17s 169ms/step - loss: 0.0390 - acc: 0.9974 - val_loss: 0.6206 - val_acc: 0.9740\n",
      "Epoch 94/100\n",
      "100/100 [==============================] - 17s 167ms/step - loss: 0.0029 - acc: 0.9994 - val_loss: 0.8689 - val_acc: 0.9710\n",
      "Epoch 95/100\n",
      "100/100 [==============================] - 17s 168ms/step - loss: 0.0423 - acc: 0.9947 - val_loss: 0.6732 - val_acc: 0.9740\n",
      "Epoch 96/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0080 - acc: 0.9989 - val_loss: 0.8798 - val_acc: 0.9670\n",
      "Epoch 97/100\n",
      "100/100 [==============================] - 17s 167ms/step - loss: 0.0126 - acc: 0.9988 - val_loss: 0.6741 - val_acc: 0.9740\n",
      "Epoch 98/100\n",
      "100/100 [==============================] - 17s 167ms/step - loss: 0.0199 - acc: 0.9975 - val_loss: 0.6897 - val_acc: 0.9690\n",
      "Epoch 99/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0017 - acc: 0.9988 - val_loss: 0.8459 - val_acc: 0.9710\n",
      "Epoch 100/100\n",
      "100/100 [==============================] - 17s 166ms/step - loss: 0.0073 - acc: 0.9989 - val_loss: 0.7736 - val_acc: 0.9730\n"
     ]
    }
   ],
   "source": [
    "model.compile(loss='binary_crossentropy',\n",
    "              optimizer=optimizers.RMSprop(lr=1e-5),\n",
    "              metrics=['acc'])\n",
    "\n",
    "history = model.fit_generator(\n",
    "      train_generator,\n",
    "      steps_per_epoch=100,\n",
    "      epochs=100,\n",
    "      validation_data=validation_generator,\n",
    "      validation_steps=50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save('cats_and_dogs_small_4.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's plot our results using the same plotting code as before:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "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(len(acc))\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 accuracy')\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": [
    "\n",
    "These curves look very noisy. To make them more readable, we can smooth them by replacing every loss and accuracy with exponential moving \n",
    "averages of these quantities. Here's a trivial utility function to do this:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def smooth_curve(points, factor=0.8):\n",
    "  smoothed_points = []\n",
    "  for point in points:\n",
    "    if smoothed_points:\n",
    "      previous = smoothed_points[-1]\n",
    "      smoothed_points.append(previous * factor + point * (1 - factor))\n",
    "    else:\n",
    "      smoothed_points.append(point)\n",
    "  return smoothed_points\n",
    "\n",
    "plt.plot(epochs,\n",
    "         smooth_curve(acc), 'bo', label='Smoothed training acc')\n",
    "plt.plot(epochs,\n",
    "         smooth_curve(val_acc), 'b', label='Smoothed validation acc')\n",
    "plt.title('Training and validation accuracy')\n",
    "plt.legend()\n",
    "\n",
    "plt.figure()\n",
    "\n",
    "plt.plot(epochs,\n",
    "         smooth_curve(loss), 'bo', label='Smoothed training loss')\n",
    "plt.plot(epochs,\n",
    "         smooth_curve(val_loss), 'b', label='Smoothed validation loss')\n",
    "plt.title('Training and validation loss')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "These curves look much cleaner and more stable. We are seeing a nice 1% absolute improvement.\n",
    "\n",
    "Note that the loss curve does not show any real improvement (in fact, it is deteriorating). You may wonder, how could accuracy improve if the \n",
    "loss isn't decreasing? The answer is simple: what we display is an average of pointwise loss values, but what actually matters for accuracy \n",
    "is the distribution of the loss values, not their average, since accuracy is the result of a binary thresholding of the class probability \n",
    "predicted by the model. The model may still be improving even if this isn't reflected in the average loss.\n",
    "\n",
    "We can now finally evaluate this model on the test data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 1000 images belonging to 2 classes.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.8/dist-packages/tensorflow/python/keras/engine/training.py:1877: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.\n",
      "  warnings.warn('`Model.evaluate_generator` is deprecated and '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "test acc: 0.9660000205039978\n"
     ]
    }
   ],
   "source": [
    "test_generator = test_datagen.flow_from_directory(\n",
    "        test_dir,\n",
    "        target_size=(150, 150),\n",
    "        batch_size=20,\n",
    "        class_mode='binary')\n",
    "\n",
    "test_loss, test_acc = model.evaluate_generator(test_generator, steps=50)\n",
    "print('test acc:', test_acc)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "Here we get a test accuracy of 97%. In the original Kaggle competition around this dataset, this would have been one of the top results. \n",
    "However, using modern deep learning techniques, we managed to reach this result using only a very small fraction of the training data \n",
    "available (about 10%). There is a huge difference between being able to train on 20,000 samples compared to 2,000 samples!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Take-aways: using convnets with small datasets\n",
    "\n",
    "Here's what you should take away from the exercises of these past two sections:\n",
    "\n",
    "* Convnets are the best type of machine learning models for computer vision tasks. It is possible to train one from scratch even on a very \n",
    "small dataset, with decent results.\n",
    "* On a small dataset, overfitting will be the main issue. Data augmentation is a powerful way to fight overfitting when working with image \n",
    "data.\n",
    "* It is easy to reuse an existing convnet on a new dataset, via feature extraction. This is a very valuable technique for working with \n",
    "small image datasets.\n",
    "* As a complement to feature extraction, one may use fine-tuning, which adapts to a new problem some of the representations previously \n",
    "learned by an existing model. This pushes performance a bit further.\n",
    "\n",
    "Now you have a solid set of tools for dealing with image classification problems, in particular with small datasets."
   ]
  }
 ],
 "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}