Autoencoder_Fashion_MNIST.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### In this notebook, we will see how we can use autoencoder as a classifier with fashion MNIST dataset.\n",
    "- Fashion MNIST dataset is a 28x28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The training set has 60,000 images, and the test set has 10,000 images.\n",
    "- It is a replacement for the original MNIST handwritten digits dataset for producing better results. The Fashion MNIST dataset consists of 10 different classes of fashion accessories like shirts, trousers, sandals, etc \n",
    "- The image dimensions, training and test splits are similar to the original MNIST dataset. \n",
    "- The dataset is freely available on this URL and can be loaded using both tensorflow and keras as a framework without having to download it on your computer."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So, what we want to do is built a convolutional autoencoder and use the encoder part of it combined with fully connected layers to recognize a new sample from the test set correctly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Loading the dependencies\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "import gzip\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "from keras.models import Model, Sequential\n",
    "from keras.optimizers import RMSprop, Adam, SGD, Adadelta\n",
    "from keras.layers import Conv2D, Input, Dense, Flatten,Dropout, merge, Reshape, MaxPooling2D, UpSampling2D, Conv2DTranspose\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.callbacks import ModelCheckpoint, CSVLogger, EarlyStopping\n",
    "from keras import regularizers\n",
    "from keras import backend as k\n",
    "from keras.utils import to_categorical\n",
    "\n",
    "from sklearn.metrics import confusion_matrix, classification_report, accuracy_score\n",
    "import itertools\n",
    "\n",
    "# import os\n",
    "# os.environ[\"TF_CPP_MIN_LOG_LEVEL\"]=\"3\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The way we read the data is as follows:\n",
    "- We open and read the gzip file as bytestream\n",
    "- We then convert the string stored in buffer using np.frombuffer() to numpy array of type float32.\n",
    "- Data is then reshaped into a 3-dimentional array."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loading the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_data(filename, num_images):\n",
    "    with gzip.open(filename) as bytestream:\n",
    "        bytestream.read(16)\n",
    "        buffer = bytestream.read(28 * 28 * num_images)\n",
    "        data = np.frombuffer(buffer, dtype = np.uint8).astype(np.float32)\n",
    "        data = data.reshape(num_images, 28,28)\n",
    "        return data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data = extract_data('train-images-idx3-ubyte.gz' ,60000)\n",
    "test_data = extract_data('t10k-images-idx3-ubyte.gz',10000)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The way we read the labels is the same as we read the data:\n",
    "- We define a function which opens the label file as a bytestream using bytestream.read() to which you pass the label dimension as 1 and the total number of images.\n",
    "- We then convert the string stored in the buffer to a numpy array of type int64.\n",
    "- We do not need to reshape the array since the variable labels will return a column vector of dimension 60,000 x 1."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_labels(filename, num_images):\n",
    "    with gzip.open(filename) as bytestream:\n",
    "        bytestream.read(8)\n",
    "        buffer = bytestream.read(1 * num_images)\n",
    "        labels = np.frombuffer(buffer, dtype = np.uint8).astype(np.uint64)\n",
    "        return labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_labels = extract_labels('train-labels-idx1-ubyte.gz', 60000)\n",
    "test_labels = extract_labels('t10k-labels-idx1-ubyte.gz',10000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([9, 0, 0, 3, 0, 2, 7, 2, 5, 5], dtype=uint64)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_labels[:10]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data Exploration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set (images) shape: {shape} (60000, 28, 28)\n",
      "Test set (image) shape: {shape} (10000, 28, 28)\n"
     ]
    }
   ],
   "source": [
    "# To see the dimension of images in the training set\n",
    "print(\"Training set (images) shape: {shape}\", format(train_data.shape))\n",
    "\n",
    "# To see the dimension of images in the test set\n",
    "print(\"Test set (image) shape: {shape}\", format(test_data.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We don't need training an testing labels for convolutional encoder as the task at hand is to reconstruct the images. However, we do need them for classification task and for data exploration purposes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# we will now create a dictionary of class names with their corresponding class labels:\n",
    "label_dict = {\n",
    "    0: 'A',\n",
    "    1: 'B',\n",
    "    2: 'C',\n",
    "    3: 'D',\n",
    "    4: 'E',\n",
    "    5: 'F',\n",
    "    6: 'G',\n",
    "    7: 'H',\n",
    "    8: 'I',\n",
    "    9: 'J' \n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let us now look at few of the images in the dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'(Label: E)')"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8e7ce32a20>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=[5,5])\n",
    "\n",
    "# display an image from the training data\n",
    "plt.subplot(121)\n",
    "curr_img = np.reshape(train_data[10], (28,28))\n",
    "curr_label = train_labels[10]\n",
    "plt.imshow(curr_img, cmap = 'gray')\n",
    "plt.title(\"(Label: \" + str(label_dict[curr_label]) + \")\")\n",
    "\n",
    "# display an image from the testing data\n",
    "plt.subplot(122)\n",
    "curr_img = np.reshape(test_data[10], (28,28))\n",
    "curr_label = test_labels[10]\n",
    "plt.imshow(curr_img, cmap= 'gray')\n",
    "plt.title(\"(Label: \" + str(label_dict[curr_label])+ \")\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These images are assigned class labels as 0 or A / 4 or E. This means all the alphabets having class of E will have the class label of 4.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data Preprocessing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We need to do the preprocessing of the data before feeding it to the model as our images are gray scale images having pixel values ranging from 0 to 255 with dimension of 28 * 28.\n",
    "So we will convert each 28 * 28 image into a matrix of size 28 * 28 * 1 which can be fed to the network."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train data shape (60000, 28, 28, 1)\n",
      "Test data shape (10000, 28, 28, 1)\n"
     ]
    }
   ],
   "source": [
    "train_data = train_data.reshape(-1, 28, 28, 1)\n",
    "test_data = test_data.reshape(-1, 28, 28, 1)\n",
    "print(\"Train data shape\", train_data.shape)\n",
    "print(\"Test data shape\", test_data.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(dtype('float32'), dtype('float32'))"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# checking the datatype of the numpy arrays\n",
    "train_data.dtype, test_data.dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# rescaling the pixel values to 0-1 scale by rescaling with maximum pixel value of the training and testing data\n",
    "train_data = train_data / np.max(train_data)\n",
    "test_data = test_data / np.max(test_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1.0, 1.0)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# verify the maximum value of the training and the testing data\n",
    "np.max(train_data), np.max(test_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Splitting the data into train and validation sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# We will now split the data into 80% training and 20% validation. This will help us to reduce the chances of overfitting\n",
    "# as you would be validating the model on the data it has not seen in the training.\n",
    "X_train, X_val, y_train, y_val = train_test_split(train_data, train_data, test_size = 0.2, random_state = 12345)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((48000, 28, 28, 1), (12000, 28, 28, 1))"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape, X_val.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Now the data is ready to be fed to the network. \n",
    "batch_size = 128\n",
    "epochs = 200\n",
    "channel = 1\n",
    "x, y = 28, 28\n",
    "input_image = Input(shape = (x, y, channel))\n",
    "num_classes = 10"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We will build encoder and a decoder using convolutional layers.\n",
    "Batch Normalization layer is used"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Building an encoder\n",
    "def encoder(input_image):\n",
    "    # input dimension - 28 * 28 * 1\n",
    "    conv1 = Conv2D(32, (3,3), activation = 'relu', padding = 'same') (input_image)  # 28 *28 * 32\n",
    "    conv1 = BatchNormalization()(conv1)\n",
    "    conv1 = Conv2D(32, (3,3), activation = 'relu', padding = 'same') (conv1)\n",
    "    conv1 = BatchNormalization()(conv1)\n",
    "    pool1 = MaxPooling2D(pool_size = (2,2))(conv1)  # 14 * 14 * 32\n",
    "    conv2 = Conv2D(64, (3,3), activation = 'relu', padding = 'same') (pool1) # 14 * 14 * 64\n",
    "    conv2 = BatchNormalization()(conv2)\n",
    "    conv2 = Conv2D(64, (3,3), activation = 'relu', padding = 'same') (conv2)\n",
    "    conv2 = BatchNormalization()(conv2)\n",
    "    pool2 = MaxPooling2D(pool_size = (2,2)) (conv2) # 7 * 7 * 64\n",
    "    conv3 = Conv2D(128, (3,3), activation = 'relu', padding = 'same') (pool2) # 7 * 7 * 128\n",
    "    conv3 = BatchNormalization()(conv3)\n",
    "    conv3 = Conv2D(128, (3,3), activation = 'relu', padding = 'same') (conv3)\n",
    "    conv3 = BatchNormalization()(conv3)\n",
    "    conv4 = Conv2D(256,(3,3), activation = 'relu', padding = 'same') (conv3) # 7 * 7 * 256\n",
    "    conv4 = BatchNormalization()(conv4)\n",
    "    conv4 = Conv2D(256, (3,3), activation = 'relu', padding = 'same') (conv4)\n",
    "    conv4 = BatchNormalization()(conv4)\n",
    "    return conv4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Building a decoder\n",
    "def decoder(conv4):\n",
    "    conv5 = Conv2D(128, (3,3), activation = 'relu', padding = 'same') (conv4) # 7 * 7 * 128\n",
    "    conv5 = BatchNormalization()(conv5)\n",
    "    conv5 = Conv2D(128, (3,3), activation = 'relu', padding = 'same')(conv5)\n",
    "    conv5 = BatchNormalization()(conv5)\n",
    "    conv6 = Conv2D(64, (3,3), activation= 'relu', padding ='same') (conv5)# 7 * 7 * 64\n",
    "    conv6 = BatchNormalization()(conv6)\n",
    "    conv6 = Conv2D(64, (3,3), activation= 'relu', padding = 'same')(conv6)\n",
    "    conv6 = BatchNormalization()(conv6)\n",
    "    up1 = UpSampling2D((2,2)) (conv6)  # 14 * 14 * 64\n",
    "    conv7 = Conv2D(32, (3,3), activation = 'relu', padding = 'same') (up1)# 14 * 14 * 32\n",
    "    conv7 = BatchNormalization()(conv7)\n",
    "    conv7 = Conv2D(32, (3,3), activation ='relu', padding = 'same')(conv7)\n",
    "    conv7 = BatchNormalization()(conv7)\n",
    "    up2 = UpSampling2D((2,2))(conv7) # 28 * 28 * 32\n",
    "    decoded = Conv2D(1,(3, 3),  activation = 'sigmoid', padding ='same') (up2) # 28 * 28 * 1\n",
    "    return decoded"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Compiling the model using RMSProp optimizer (helps in reducing overfitting by \n",
    "# changing learning rate while the model is training) and specifying the loss as mean squared loss\n",
    "autoencoder = Model(input_image, decoder(encoder(input_image)))\n",
    "autoencoder.compile(loss='mean_squared_error', optimizer = RMSprop())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         (None, 28, 28, 1)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 28, 28, 32)        320       \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 28, 28, 32)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 28, 28, 32)        9248      \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 28, 28, 32)        128       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 14, 14, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 14, 14, 64)        18496     \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 14, 14, 64)        256       \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 14, 14, 64)        36928     \n",
      "_________________________________________________________________\n",
      "batch_normalization_4 (Batch (None, 14, 14, 64)        256       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 7, 7, 64)          0         \n",
      "_________________________________________________________________\n",
      "conv2d_5 (Conv2D)            (None, 7, 7, 128)         73856     \n",
      "_________________________________________________________________\n",
      "batch_normalization_5 (Batch (None, 7, 7, 128)         512       \n",
      "_________________________________________________________________\n",
      "conv2d_6 (Conv2D)            (None, 7, 7, 128)         147584    \n",
      "_________________________________________________________________\n",
      "batch_normalization_6 (Batch (None, 7, 7, 128)         512       \n",
      "_________________________________________________________________\n",
      "conv2d_7 (Conv2D)            (None, 7, 7, 256)         295168    \n",
      "_________________________________________________________________\n",
      "batch_normalization_7 (Batch (None, 7, 7, 256)         1024      \n",
      "_________________________________________________________________\n",
      "conv2d_8 (Conv2D)            (None, 7, 7, 256)         590080    \n",
      "_________________________________________________________________\n",
      "batch_normalization_8 (Batch (None, 7, 7, 256)         1024      \n",
      "_________________________________________________________________\n",
      "conv2d_9 (Conv2D)            (None, 7, 7, 128)         295040    \n",
      "_________________________________________________________________\n",
      "batch_normalization_9 (Batch (None, 7, 7, 128)         512       \n",
      "_________________________________________________________________\n",
      "conv2d_10 (Conv2D)           (None, 7, 7, 128)         147584    \n",
      "_________________________________________________________________\n",
      "batch_normalization_10 (Batc (None, 7, 7, 128)         512       \n",
      "_________________________________________________________________\n",
      "conv2d_11 (Conv2D)           (None, 7, 7, 64)          73792     \n",
      "_________________________________________________________________\n",
      "batch_normalization_11 (Batc (None, 7, 7, 64)          256       \n",
      "_________________________________________________________________\n",
      "conv2d_12 (Conv2D)           (None, 7, 7, 64)          36928     \n",
      "_________________________________________________________________\n",
      "batch_normalization_12 (Batc (None, 7, 7, 64)          256       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_1 (UpSampling2 (None, 14, 14, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_13 (Conv2D)           (None, 14, 14, 32)        18464     \n",
      "_________________________________________________________________\n",
      "batch_normalization_13 (Batc (None, 14, 14, 32)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_14 (Conv2D)           (None, 14, 14, 32)        9248      \n",
      "_________________________________________________________________\n",
      "batch_normalization_14 (Batc (None, 14, 14, 32)        128       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_2 (UpSampling2 (None, 28, 28, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_15 (Conv2D)           (None, 28, 28, 1)         289       \n",
      "=================================================================\n",
      "Total params: 1,758,657\n",
      "Trainable params: 1,755,841\n",
      "Non-trainable params: 2,816\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Let us see the autoencoder summary\n",
    "autoencoder.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training the model "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 48000 samples, validate on 12000 samples\n",
      "Epoch 1/200\n",
      "48000/48000 [==============================] - 11s 227us/step - loss: 0.0264 - val_loss: 0.0457\n",
      "Epoch 2/200\n",
      "48000/48000 [==============================] - 9s 196us/step - loss: 0.0111 - val_loss: 0.0221\n",
      "Epoch 3/200\n",
      "48000/48000 [==============================] - 9s 197us/step - loss: 0.0084 - val_loss: 0.0101\n",
      "Epoch 4/200\n",
      "48000/48000 [==============================] - 9s 197us/step - loss: 0.0070 - val_loss: 0.0077\n",
      "Epoch 5/200\n",
      "48000/48000 [==============================] - 9s 197us/step - loss: 0.0062 - val_loss: 0.0071\n",
      "Epoch 6/200\n",
      "48000/48000 [==============================] - 9s 197us/step - loss: 0.0058 - val_loss: 0.0104\n",
      "Epoch 7/200\n",
      "48000/48000 [==============================] - 10s 198us/step - loss: 0.0051 - val_loss: 0.0054\n",
      "Epoch 8/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0045 - val_loss: 0.0051\n",
      "Epoch 9/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0041 - val_loss: 0.0051\n",
      "Epoch 10/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0039 - val_loss: 0.0036\n",
      "Epoch 11/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0036 - val_loss: 0.0040\n",
      "Epoch 12/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0034 - val_loss: 0.0036\n",
      "Epoch 13/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0032 - val_loss: 0.0036\n",
      "Epoch 14/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0031 - val_loss: 0.0030\n",
      "Epoch 15/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0030 - val_loss: 0.0033\n",
      "Epoch 16/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0029 - val_loss: 0.0030\n",
      "Epoch 17/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0028 - val_loss: 0.0028\n",
      "Epoch 18/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0027 - val_loss: 0.0024\n",
      "Epoch 19/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0026 - val_loss: 0.0045\n",
      "Epoch 20/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0026 - val_loss: 0.0028\n",
      "Epoch 21/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0025 - val_loss: 0.0028\n",
      "Epoch 22/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0024 - val_loss: 0.0023\n",
      "Epoch 23/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0023 - val_loss: 0.0030\n",
      "Epoch 24/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0023 - val_loss: 0.0025\n",
      "Epoch 25/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0023 - val_loss: 0.0022\n",
      "Epoch 26/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0022 - val_loss: 0.0026\n",
      "Epoch 27/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0021 - val_loss: 0.0024\n",
      "Epoch 28/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0021 - val_loss: 0.0023\n",
      "Epoch 29/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0021 - val_loss: 0.0023\n",
      "Epoch 30/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0021 - val_loss: 0.0021\n",
      "Epoch 31/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0020 - val_loss: 0.0023\n",
      "Epoch 32/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0020 - val_loss: 0.0019\n",
      "Epoch 33/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0019 - val_loss: 0.0023\n",
      "Epoch 34/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0019 - val_loss: 0.0020\n",
      "Epoch 35/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0019 - val_loss: 0.0019\n",
      "Epoch 36/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0019 - val_loss: 0.0024\n",
      "Epoch 37/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0018 - val_loss: 0.0018\n",
      "Epoch 38/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0018 - val_loss: 0.0029\n",
      "Epoch 39/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0018 - val_loss: 0.0023\n",
      "Epoch 40/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0017 - val_loss: 0.0016\n",
      "Epoch 41/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0017 - val_loss: 0.0019\n",
      "Epoch 42/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0017 - val_loss: 0.0020\n",
      "Epoch 43/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0017 - val_loss: 0.0019\n",
      "Epoch 44/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0017 - val_loss: 0.0017\n",
      "Epoch 45/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0017 - val_loss: 0.0017\n",
      "Epoch 46/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0016 - val_loss: 0.0018\n",
      "Epoch 47/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0016 - val_loss: 0.0017\n",
      "Epoch 48/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0016 - val_loss: 0.0016\n",
      "Epoch 49/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0016 - val_loss: 0.0020\n",
      "Epoch 50/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0016 - val_loss: 0.0016\n",
      "Epoch 51/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0016 - val_loss: 0.0015\n",
      "Epoch 52/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0015 - val_loss: 0.0019\n",
      "Epoch 53/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0015 - val_loss: 0.0017\n",
      "Epoch 54/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0015 - val_loss: 0.0016\n",
      "Epoch 55/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0015 - val_loss: 0.0016\n",
      "Epoch 56/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0015 - val_loss: 0.0018\n",
      "Epoch 57/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0015 - val_loss: 0.0017\n",
      "Epoch 58/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0015\n",
      "Epoch 59/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0014\n",
      "Epoch 60/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0015\n",
      "Epoch 61/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0015\n",
      "Epoch 62/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0016\n",
      "Epoch 63/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0014\n",
      "Epoch 64/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0014\n",
      "Epoch 65/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0018\n",
      "Epoch 66/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0014 - val_loss: 0.0017\n",
      "Epoch 67/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0013 - val_loss: 0.0016\n",
      "Epoch 68/200\n",
      "48000/48000 [==============================] - 10s 198us/step - loss: 0.0013 - val_loss: 0.0015\n",
      "Epoch 69/200\n",
      "48000/48000 [==============================] - 10s 198us/step - loss: 0.0013 - val_loss: 0.0016\n",
      "Epoch 70/200\n",
      "48000/48000 [==============================] - 9s 198us/step - loss: 0.0013 - val_loss: 0.0013\n",
      "Epoch 71/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0013 - val_loss: 0.0017\n",
      "Epoch 72/200\n",
      "48000/48000 [==============================] - 10s 198us/step - loss: 0.0013 - val_loss: 0.0014\n",
      "Epoch 73/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0013 - val_loss: 0.0014\n",
      "Epoch 74/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0013 - val_loss: 0.0013\n",
      "Epoch 75/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0013 - val_loss: 0.0014\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 76/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0013 - val_loss: 0.0018\n",
      "Epoch 77/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0013 - val_loss: 0.0013\n",
      "Epoch 78/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0013 - val_loss: 0.0015\n",
      "Epoch 79/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0012 - val_loss: 0.0013\n",
      "Epoch 80/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0012 - val_loss: 0.0013\n",
      "Epoch 81/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0012 - val_loss: 0.0012\n",
      "Epoch 82/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0012 - val_loss: 0.0014\n",
      "Epoch 83/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0012 - val_loss: 0.0013\n",
      "Epoch 84/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0012 - val_loss: 0.0013\n",
      "Epoch 85/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0012 - val_loss: 0.0015\n",
      "Epoch 86/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0012 - val_loss: 0.0012\n",
      "Epoch 87/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0012 - val_loss: 0.0015\n",
      "Epoch 88/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0012 - val_loss: 0.0013\n",
      "Epoch 89/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0012 - val_loss: 0.0014\n",
      "Epoch 90/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0012 - val_loss: 0.0015\n",
      "Epoch 91/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0012 - val_loss: 0.0014\n",
      "Epoch 92/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0012 - val_loss: 0.0013\n",
      "Epoch 93/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0011 - val_loss: 0.0013\n",
      "Epoch 94/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0013\n",
      "Epoch 95/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0021\n",
      "Epoch 96/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 97/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 98/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0011 - val_loss: 0.0014\n",
      "Epoch 99/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 100/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 101/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 102/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 103/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0015\n",
      "Epoch 104/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 105/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0015\n",
      "Epoch 106/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 107/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 0.0011 - val_loss: 0.0013\n",
      "Epoch 108/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 109/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0011 - val_loss: 0.0013\n",
      "Epoch 110/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0011 - val_loss: 0.0012\n",
      "Epoch 111/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0011\n",
      "Epoch 112/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0012\n",
      "Epoch 113/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0012\n",
      "Epoch 114/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0013\n",
      "Epoch 115/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0011\n",
      "Epoch 116/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0012\n",
      "Epoch 117/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0015\n",
      "Epoch 118/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0011\n",
      "Epoch 119/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0011\n",
      "Epoch 120/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0012\n",
      "Epoch 121/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0012\n",
      "Epoch 122/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 0.0010 - val_loss: 0.0013\n",
      "Epoch 123/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 9.9999e-04 - val_loss: 0.0011\n",
      "Epoch 124/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 9.9639e-04 - val_loss: 0.0011\n",
      "Epoch 125/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 9.9016e-04 - val_loss: 0.0011\n",
      "Epoch 126/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 9.9421e-04 - val_loss: 0.0012\n",
      "Epoch 127/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 9.7994e-04 - val_loss: 0.0011\n",
      "Epoch 128/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 9.7226e-04 - val_loss: 0.0011\n",
      "Epoch 129/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 9.7160e-04 - val_loss: 0.0012\n",
      "Epoch 130/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.6938e-04 - val_loss: 0.0011\n",
      "Epoch 131/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.6925e-04 - val_loss: 0.0011\n",
      "Epoch 132/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.5569e-04 - val_loss: 0.0011\n",
      "Epoch 133/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.6097e-04 - val_loss: 0.0012\n",
      "Epoch 134/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.5917e-04 - val_loss: 0.0011\n",
      "Epoch 135/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.5380e-04 - val_loss: 0.0011\n",
      "Epoch 136/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.5319e-04 - val_loss: 0.0012\n",
      "Epoch 137/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.4862e-04 - val_loss: 0.0011\n",
      "Epoch 138/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.4575e-04 - val_loss: 0.0011\n",
      "Epoch 139/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.4296e-04 - val_loss: 0.0013\n",
      "Epoch 140/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 9.3923e-04 - val_loss: 0.0011\n",
      "Epoch 141/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.2895e-04 - val_loss: 0.0014\n",
      "Epoch 142/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.3119e-04 - val_loss: 0.0012\n",
      "Epoch 143/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.2170e-04 - val_loss: 0.0012\n",
      "Epoch 144/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.2611e-04 - val_loss: 0.0014\n",
      "Epoch 145/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.1583e-04 - val_loss: 0.0011\n",
      "Epoch 146/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.1227e-04 - val_loss: 0.0012\n",
      "Epoch 147/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.2262e-04 - val_loss: 0.0011\n",
      "Epoch 148/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.1036e-04 - val_loss: 0.0011\n",
      "Epoch 149/200\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.1014e-04 - val_loss: 0.0011\n",
      "Epoch 150/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.1767e-04 - val_loss: 0.0011\n",
      "Epoch 151/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.0573e-04 - val_loss: 0.0011\n",
      "Epoch 152/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 9.0388e-04 - val_loss: 0.0010\n",
      "Epoch 153/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.9637e-04 - val_loss: 0.0011\n",
      "Epoch 154/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.9497e-04 - val_loss: 0.0011\n",
      "Epoch 155/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.9222e-04 - val_loss: 0.0011\n",
      "Epoch 156/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.9232e-04 - val_loss: 0.0011\n",
      "Epoch 157/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.8786e-04 - val_loss: 0.0010\n",
      "Epoch 158/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.8429e-04 - val_loss: 0.0010\n",
      "Epoch 159/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.7985e-04 - val_loss: 0.0010\n",
      "Epoch 160/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.8443e-04 - val_loss: 0.0011\n",
      "Epoch 161/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.8219e-04 - val_loss: 0.0011\n",
      "Epoch 162/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.7491e-04 - val_loss: 0.0011\n",
      "Epoch 163/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.7846e-04 - val_loss: 0.0011\n",
      "Epoch 164/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.6518e-04 - val_loss: 0.0011\n",
      "Epoch 165/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.6564e-04 - val_loss: 0.0010\n",
      "Epoch 166/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.6763e-04 - val_loss: 0.0010\n",
      "Epoch 167/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.6561e-04 - val_loss: 0.0011\n",
      "Epoch 168/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.6143e-04 - val_loss: 9.9077e-04\n",
      "Epoch 169/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.5907e-04 - val_loss: 0.0010\n",
      "Epoch 170/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.5629e-04 - val_loss: 0.0011\n",
      "Epoch 171/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.5799e-04 - val_loss: 0.0010\n",
      "Epoch 172/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.5245e-04 - val_loss: 0.0012\n",
      "Epoch 173/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.5089e-04 - val_loss: 0.0010\n",
      "Epoch 174/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.4762e-04 - val_loss: 0.0011\n",
      "Epoch 175/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.4643e-04 - val_loss: 0.0010\n",
      "Epoch 176/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.4362e-04 - val_loss: 0.0011\n",
      "Epoch 177/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.4522e-04 - val_loss: 0.0011\n",
      "Epoch 178/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.4771e-04 - val_loss: 0.0010\n",
      "Epoch 179/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.3505e-04 - val_loss: 0.0011\n",
      "Epoch 180/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.3634e-04 - val_loss: 0.0011\n",
      "Epoch 181/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.3524e-04 - val_loss: 0.0011\n",
      "Epoch 182/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.3632e-04 - val_loss: 0.0010\n",
      "Epoch 183/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.3004e-04 - val_loss: 0.0011\n",
      "Epoch 184/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.2821e-04 - val_loss: 0.0011\n",
      "Epoch 185/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.2451e-04 - val_loss: 0.0011\n",
      "Epoch 186/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.2318e-04 - val_loss: 0.0011\n",
      "Epoch 187/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.2462e-04 - val_loss: 0.0010\n",
      "Epoch 188/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.1728e-04 - val_loss: 0.0010\n",
      "Epoch 189/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.1606e-04 - val_loss: 0.0010\n",
      "Epoch 190/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.2615e-04 - val_loss: 0.0011\n",
      "Epoch 191/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.2301e-04 - val_loss: 0.0011\n",
      "Epoch 192/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.1235e-04 - val_loss: 9.9467e-04\n",
      "Epoch 193/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.0882e-04 - val_loss: 0.0010\n",
      "Epoch 194/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.0533e-04 - val_loss: 0.0010\n",
      "Epoch 195/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.1214e-04 - val_loss: 0.0011\n",
      "Epoch 196/200\n",
      "48000/48000 [==============================] - 10s 200us/step - loss: 8.0601e-04 - val_loss: 0.0012\n",
      "Epoch 197/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.0654e-04 - val_loss: 9.6515e-04\n",
      "Epoch 198/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 8.0266e-04 - val_loss: 0.0011\n",
      "Epoch 199/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 7.9812e-04 - val_loss: 0.0011\n",
      "Epoch 200/200\n",
      "48000/48000 [==============================] - 10s 199us/step - loss: 7.9890e-04 - val_loss: 0.0011\n"
     ]
    }
   ],
   "source": [
    "# we will now train the autoencoder for 200 epochs using keras fit function\n",
    "# We are going to save the history to plot accuracy and loss curves to help in analyzing the model performance.\n",
    "csv_logger = CSVLogger('autoencoder_log.csv', append = True, separator = ';')\n",
    "history = autoencoder.fit(X_train, y_train, batch_size = batch_size, epochs= epochs,\n",
    "                                   verbose =1, validation_data = (X_val, y_val), callbacks = [csv_logger])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# saving the model\n",
    "autoencoder_json = autoencoder.to_json()\n",
    "with open('autoencoder_json','w') as json_file:\n",
    "    json_file.write(autoencoder_json)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'loss'])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8dd41348d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Now we can plot the loss curves to visualize the model performance\n",
    "# list all data in history\n",
    "print(history.history.keys())\n",
    "\n",
    "\n",
    "# summarize history for loss\n",
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can clearly see that the training and validation loss are in sync. The validation loss is decreasing which is a good sign which also mean that model is not overfitting. So, we can say the model's generalization capabity is good.\n",
    "\n",
    "Now, we will use the encoder part of the autoencoder to classify the fashion MNIST images."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# We will need the encoder weights of the autoencoder for classification task. We will now save the \n",
    "# full autoencoder weights and then we will learn to extract the encoder weights.\n",
    "autoencoder.save_weights('autoencoder_weights.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Classifying the fashion MNIST images"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Loading the encoder weights from the trained autoencoder's weight\n",
    "- We will add a few fully connected (dense) layers to the encoder to classify the fashion MNIST images.\n",
    "- We will convert labels into one-hot encoding vectors. In one-hot encoding, we convert the categorical values to\n",
    "    a vector of numbers as machine learning algorithms cannot work with understand categorical data directly.\n",
    "- So, in our case the one hot encoding will be a row vector and for each image it will have a dimension of 1 X 10.\n",
    "- The vector consists of all zeros except for the class that it represents, and for that, it is 1.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Original label: 9\n",
      "One-hot label: [0. 0. 0. 0. 0. 0. 0. 0. 0. 1.]\n"
     ]
    }
   ],
   "source": [
    "X_train_one_hot = to_categorical(train_labels)\n",
    "X_test_one_hot = to_categorical(test_labels)\n",
    "\n",
    "# Display the change for category label using one-hot encoding\n",
    "print('Original label:', train_labels[0])\n",
    "print('One-hot label:', X_train_one_hot[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Splitting the data again into train and test for classification purpose using the one hot labels this time\n",
    "X_trainc, X_valc, y_trainc, y_valc = train_test_split(train_data, X_train_one_hot, test_size = 0.2, random_state = 12345)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((48000, 28, 28, 1), (12000, 28, 28, 1), (48000, 10), (12000, 10))"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# checking the shape\n",
    "X_trainc.shape, X_valc.shape, y_trainc.shape, y_valc.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Now we will be using building the classification model using the encoder part (define encoder) of the autoencoder\n",
    "# defining the fully connected layer for the encoder part\n",
    "def fc(enco):\n",
    "    flat = Flatten() (enco)\n",
    "    fc1 = Dense(128, activation = 'relu') (flat)\n",
    "    fc2 = Dense(num_classes, activation = 'softmax') (fc1)\n",
    "    return fc2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "encode = encoder(input_image)\n",
    "full_model = Model(input_image, fc(encode))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "for l1, l2 in zip(full_model.layers[:19], autoencoder.layers[:19]):\n",
    "    l1.set_weights(l2.get_weights())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[ 0.03484822,  0.54488236,  0.79232454, -0.43435058,\n",
       "         -0.34744975, -0.9587529 , -0.40104276,  0.12997001,\n",
       "          0.35536173,  0.2808614 ,  0.09164336,  0.18004183,\n",
       "          0.16222575, -0.18727636, -0.13956393, -0.44285783,\n",
       "          0.020556  ,  0.14557205,  0.02137145,  0.5783866 ,\n",
       "          0.25203896,  0.1850529 , -0.35790852, -0.10435557,\n",
       "          0.22306763,  0.23999791, -0.18548656,  0.07568264,\n",
       "         -0.4926539 , -0.4113123 , -0.37110335, -0.433332  ]],\n",
       "\n",
       "       [[ 0.2357364 , -0.39489555,  0.4701999 , -0.1704079 ,\n",
       "         -0.44132775,  0.01831277,  0.03301767,  0.41107908,\n",
       "          0.41745707, -0.41281393, -1.4580994 , -0.6703689 ,\n",
       "          0.15900806, -0.03025126, -0.32384643,  0.57313126,\n",
       "          0.4474571 , -0.5015263 ,  0.04522365, -0.16300884,\n",
       "          0.4344801 ,  0.7895104 , -0.20925617,  0.13910852,\n",
       "          0.09706446, -0.24407418, -0.80453765,  0.32705525,\n",
       "         -0.09608769,  0.16976206,  0.01944006, -0.09635919]],\n",
       "\n",
       "       [[-0.3616235 , -0.0912574 , -0.14394717,  0.05616005,\n",
       "         -0.42791054, -1.0513803 ,  0.38196707, -0.25973085,\n",
       "          0.22198543, -0.9439066 , -0.7405361 ,  0.27127096,\n",
       "         -0.11148115,  0.18584691, -0.16645628,  0.115528  ,\n",
       "          0.30022016,  0.52147233,  0.02483999, -0.15558524,\n",
       "          0.30153415,  0.1964361 ,  0.5217982 ,  0.43939018,\n",
       "          0.2390272 , -0.02010331, -0.30281052, -0.3867128 ,\n",
       "          0.5658195 ,  0.34132668,  0.0546862 , -0.18588355]]],\n",
       "      dtype=float32)"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Since, we are using the encoder part of the autoencoder for classification purpose, we should make sure that\n",
    "# the weights of the encoder part of the autoencoder are are similar to the weights you loaded to the \n",
    "# encoder function of the classification model. If they are not similar, there is no pointin using the autoencoder classification\n",
    "# strategy.\n",
    "\n",
    "# First layer weight of autoencoder\n",
    "autoencoder.get_weights()[0][1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[ 0.03484822,  0.54488236,  0.79232454, -0.43435058,\n",
       "         -0.34744975, -0.9587529 , -0.40104276,  0.12997001,\n",
       "          0.35536173,  0.2808614 ,  0.09164336,  0.18004183,\n",
       "          0.16222575, -0.18727636, -0.13956393, -0.44285783,\n",
       "          0.020556  ,  0.14557205,  0.02137145,  0.5783866 ,\n",
       "          0.25203896,  0.1850529 , -0.35790852, -0.10435557,\n",
       "          0.22306763,  0.23999791, -0.18548656,  0.07568264,\n",
       "         -0.4926539 , -0.4113123 , -0.37110335, -0.433332  ]],\n",
       "\n",
       "       [[ 0.2357364 , -0.39489555,  0.4701999 , -0.1704079 ,\n",
       "         -0.44132775,  0.01831277,  0.03301767,  0.41107908,\n",
       "          0.41745707, -0.41281393, -1.4580994 , -0.6703689 ,\n",
       "          0.15900806, -0.03025126, -0.32384643,  0.57313126,\n",
       "          0.4474571 , -0.5015263 ,  0.04522365, -0.16300884,\n",
       "          0.4344801 ,  0.7895104 , -0.20925617,  0.13910852,\n",
       "          0.09706446, -0.24407418, -0.80453765,  0.32705525,\n",
       "         -0.09608769,  0.16976206,  0.01944006, -0.09635919]],\n",
       "\n",
       "       [[-0.3616235 , -0.0912574 , -0.14394717,  0.05616005,\n",
       "         -0.42791054, -1.0513803 ,  0.38196707, -0.25973085,\n",
       "          0.22198543, -0.9439066 , -0.7405361 ,  0.27127096,\n",
       "         -0.11148115,  0.18584691, -0.16645628,  0.115528  ,\n",
       "          0.30022016,  0.52147233,  0.02483999, -0.15558524,\n",
       "          0.30153415,  0.1964361 ,  0.5217982 ,  0.43939018,\n",
       "          0.2390272 , -0.02010331, -0.30281052, -0.3867128 ,\n",
       "          0.5658195 ,  0.34132668,  0.0546862 , -0.18588355]]],\n",
       "      dtype=float32)"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "full_model.get_weights()[0][1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The weights are similar, so we can move further and compile the model and start the training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Since the encoder part is already trained, we do not need to train it again. So, we will make those \n",
    "# 19 layers of the model as trainable false. We will only train the fully connected layer part\n",
    "for layer in full_model.layers[:19]:\n",
    "    layer.trainable = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Now, we will compile the model\n",
    "full_model.compile(loss ='categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         (None, 28, 28, 1)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_16 (Conv2D)           (None, 28, 28, 32)        320       \n",
      "_________________________________________________________________\n",
      "batch_normalization_15 (Batc (None, 28, 28, 32)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_17 (Conv2D)           (None, 28, 28, 32)        9248      \n",
      "_________________________________________________________________\n",
      "batch_normalization_16 (Batc (None, 28, 28, 32)        128       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2 (None, 14, 14, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_18 (Conv2D)           (None, 14, 14, 64)        18496     \n",
      "_________________________________________________________________\n",
      "batch_normalization_17 (Batc (None, 14, 14, 64)        256       \n",
      "_________________________________________________________________\n",
      "conv2d_19 (Conv2D)           (None, 14, 14, 64)        36928     \n",
      "_________________________________________________________________\n",
      "batch_normalization_18 (Batc (None, 14, 14, 64)        256       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2 (None, 7, 7, 64)          0         \n",
      "_________________________________________________________________\n",
      "conv2d_20 (Conv2D)           (None, 7, 7, 128)         73856     \n",
      "_________________________________________________________________\n",
      "batch_normalization_19 (Batc (None, 7, 7, 128)         512       \n",
      "_________________________________________________________________\n",
      "conv2d_21 (Conv2D)           (None, 7, 7, 128)         147584    \n",
      "_________________________________________________________________\n",
      "batch_normalization_20 (Batc (None, 7, 7, 128)         512       \n",
      "_________________________________________________________________\n",
      "conv2d_22 (Conv2D)           (None, 7, 7, 256)         295168    \n",
      "_________________________________________________________________\n",
      "batch_normalization_21 (Batc (None, 7, 7, 256)         1024      \n",
      "_________________________________________________________________\n",
      "conv2d_23 (Conv2D)           (None, 7, 7, 256)         590080    \n",
      "_________________________________________________________________\n",
      "batch_normalization_22 (Batc (None, 7, 7, 256)         1024      \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 12544)             0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 128)               1605760   \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                1290      \n",
      "=================================================================\n",
      "Total params: 2,782,570\n",
      "Trainable params: 1,607,050\n",
      "Non-trainable params: 1,175,520\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# summary of the model\n",
    "full_model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training the classification model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 48000 samples, validate on 12000 samples\n",
      "Epoch 1/100\n",
      "48000/48000 [==============================] - 5s 105us/step - loss: 0.3715 - acc: 0.8717 - val_loss: 0.2606 - val_acc: 0.9070\n",
      "Epoch 2/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.2247 - acc: 0.9181 - val_loss: 0.2682 - val_acc: 0.9090\n",
      "Epoch 3/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.1811 - acc: 0.9323 - val_loss: 0.2732 - val_acc: 0.9077\n",
      "Epoch 4/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.1537 - acc: 0.9404 - val_loss: 0.2651 - val_acc: 0.9137\n",
      "Epoch 5/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.1343 - acc: 0.9497 - val_loss: 0.2641 - val_acc: 0.9183\n",
      "Epoch 6/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.1141 - acc: 0.9580 - val_loss: 0.2794 - val_acc: 0.9172\n",
      "Epoch 7/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.1065 - acc: 0.9605 - val_loss: 0.3272 - val_acc: 0.9129\n",
      "Epoch 8/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0877 - acc: 0.9668 - val_loss: 0.3329 - val_acc: 0.9113\n",
      "Epoch 9/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0776 - acc: 0.9715 - val_loss: 0.3365 - val_acc: 0.9179\n",
      "Epoch 10/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0667 - acc: 0.9749 - val_loss: 0.3663 - val_acc: 0.9169\n",
      "Epoch 11/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0663 - acc: 0.9755 - val_loss: 0.3758 - val_acc: 0.9178\n",
      "Epoch 12/100\n",
      "48000/48000 [==============================] - 4s 91us/step - loss: 0.0568 - acc: 0.9785 - val_loss: 0.3729 - val_acc: 0.9209\n",
      "Epoch 13/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0487 - acc: 0.9822 - val_loss: 0.4374 - val_acc: 0.9131\n",
      "Epoch 14/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0480 - acc: 0.9826 - val_loss: 0.4172 - val_acc: 0.9193\n",
      "Epoch 15/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0429 - acc: 0.9844 - val_loss: 0.4454 - val_acc: 0.9173\n",
      "Epoch 16/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0432 - acc: 0.9846 - val_loss: 0.4489 - val_acc: 0.9167\n",
      "Epoch 17/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0383 - acc: 0.9856 - val_loss: 0.4646 - val_acc: 0.9190\n",
      "Epoch 18/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0338 - acc: 0.9881 - val_loss: 0.4670 - val_acc: 0.9240\n",
      "Epoch 19/100\n",
      "48000/48000 [==============================] - 4s 94us/step - loss: 0.0326 - acc: 0.9875 - val_loss: 0.4513 - val_acc: 0.9208\n",
      "Epoch 20/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0322 - acc: 0.9890 - val_loss: 0.5287 - val_acc: 0.9181\n",
      "Epoch 21/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0345 - acc: 0.9884 - val_loss: 0.5058 - val_acc: 0.9203\n",
      "Epoch 22/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0275 - acc: 0.9903 - val_loss: 0.5003 - val_acc: 0.9206\n",
      "Epoch 23/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0295 - acc: 0.9900 - val_loss: 0.5098 - val_acc: 0.9195\n",
      "Epoch 24/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0268 - acc: 0.9908 - val_loss: 0.5530 - val_acc: 0.9217\n",
      "Epoch 25/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0248 - acc: 0.9916 - val_loss: 0.5748 - val_acc: 0.9182\n",
      "Epoch 26/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0282 - acc: 0.9904 - val_loss: 0.5891 - val_acc: 0.9147\n",
      "Epoch 27/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0300 - acc: 0.9900 - val_loss: 0.5466 - val_acc: 0.9200\n",
      "Epoch 28/100\n",
      "48000/48000 [==============================] - 4s 94us/step - loss: 0.0271 - acc: 0.9915 - val_loss: 0.5432 - val_acc: 0.9189\n",
      "Epoch 29/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0166 - acc: 0.9940 - val_loss: 0.5851 - val_acc: 0.9162\n",
      "Epoch 30/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0228 - acc: 0.9925 - val_loss: 0.5744 - val_acc: 0.9183\n",
      "Epoch 31/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0227 - acc: 0.9922 - val_loss: 0.5854 - val_acc: 0.9187\n",
      "Epoch 32/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0211 - acc: 0.9931 - val_loss: 0.6011 - val_acc: 0.9181\n",
      "Epoch 33/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0274 - acc: 0.9914 - val_loss: 0.5855 - val_acc: 0.9224\n",
      "Epoch 34/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0166 - acc: 0.9946 - val_loss: 0.6188 - val_acc: 0.9192\n",
      "Epoch 35/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0257 - acc: 0.9920 - val_loss: 0.6302 - val_acc: 0.9188\n",
      "Epoch 36/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0165 - acc: 0.9945 - val_loss: 0.6731 - val_acc: 0.9103\n",
      "Epoch 37/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0205 - acc: 0.9935 - val_loss: 0.6505 - val_acc: 0.9166\n",
      "Epoch 38/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0164 - acc: 0.9945 - val_loss: 0.6260 - val_acc: 0.9207\n",
      "Epoch 39/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0195 - acc: 0.9939 - val_loss: 0.6459 - val_acc: 0.9183\n",
      "Epoch 40/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0242 - acc: 0.9925 - val_loss: 0.6301 - val_acc: 0.9185\n",
      "Epoch 41/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0180 - acc: 0.9940 - val_loss: 0.7275 - val_acc: 0.9130\n",
      "Epoch 42/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0137 - acc: 0.9958 - val_loss: 0.6225 - val_acc: 0.9229\n",
      "Epoch 43/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0197 - acc: 0.9937 - val_loss: 0.6446 - val_acc: 0.9199\n",
      "Epoch 44/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0175 - acc: 0.9943 - val_loss: 0.6656 - val_acc: 0.9177\n",
      "Epoch 45/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0186 - acc: 0.9942 - val_loss: 0.6962 - val_acc: 0.9163\n",
      "Epoch 46/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0169 - acc: 0.9945 - val_loss: 0.6765 - val_acc: 0.9166\n",
      "Epoch 47/100\n",
      "48000/48000 [==============================] - 5s 94us/step - loss: 0.0135 - acc: 0.9959 - val_loss: 0.6996 - val_acc: 0.9189\n",
      "Epoch 48/100\n",
      "48000/48000 [==============================] - 4s 94us/step - loss: 0.0261 - acc: 0.9926 - val_loss: 0.7246 - val_acc: 0.9133\n",
      "Epoch 49/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0153 - acc: 0.9950 - val_loss: 0.6528 - val_acc: 0.9221\n",
      "Epoch 50/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0129 - acc: 0.9957 - val_loss: 0.6538 - val_acc: 0.9207\n",
      "Epoch 51/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0152 - acc: 0.9954 - val_loss: 0.6926 - val_acc: 0.9180\n",
      "Epoch 52/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0185 - acc: 0.9948 - val_loss: 0.6632 - val_acc: 0.9228\n",
      "Epoch 53/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0134 - acc: 0.9958 - val_loss: 0.6749 - val_acc: 0.9224\n",
      "Epoch 54/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0144 - acc: 0.9956 - val_loss: 0.7254 - val_acc: 0.9165\n",
      "Epoch 55/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0181 - acc: 0.9950 - val_loss: 0.7358 - val_acc: 0.9143\n",
      "Epoch 56/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0151 - acc: 0.9960 - val_loss: 0.7175 - val_acc: 0.9218\n",
      "Epoch 57/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0165 - acc: 0.9953 - val_loss: 0.7063 - val_acc: 0.9163\n",
      "Epoch 58/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0178 - acc: 0.9946 - val_loss: 0.6935 - val_acc: 0.9193\n",
      "Epoch 59/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "48000/48000 [==============================] - 4s 94us/step - loss: 0.0108 - acc: 0.9969 - val_loss: 0.7031 - val_acc: 0.9205\n",
      "Epoch 60/100\n",
      "48000/48000 [==============================] - 5s 94us/step - loss: 0.0114 - acc: 0.9966 - val_loss: 0.6992 - val_acc: 0.9229\n",
      "Epoch 61/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0219 - acc: 0.9942 - val_loss: 0.7361 - val_acc: 0.9208\n",
      "Epoch 62/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0116 - acc: 0.9968 - val_loss: 0.7707 - val_acc: 0.9162\n",
      "Epoch 63/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0187 - acc: 0.9946 - val_loss: 0.7127 - val_acc: 0.9241\n",
      "Epoch 64/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0135 - acc: 0.9964 - val_loss: 0.7662 - val_acc: 0.9201\n",
      "Epoch 65/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0191 - acc: 0.9949 - val_loss: 0.7786 - val_acc: 0.9188\n",
      "Epoch 66/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0170 - acc: 0.9954 - val_loss: 0.7551 - val_acc: 0.9177\n",
      "Epoch 67/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0136 - acc: 0.9962 - val_loss: 0.7584 - val_acc: 0.9171\n",
      "Epoch 68/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0149 - acc: 0.9965 - val_loss: 0.7486 - val_acc: 0.9203\n",
      "Epoch 69/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0170 - acc: 0.9957 - val_loss: 0.7673 - val_acc: 0.9148\n",
      "Epoch 70/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0114 - acc: 0.9970 - val_loss: 0.7516 - val_acc: 0.9147\n",
      "Epoch 71/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0147 - acc: 0.9960 - val_loss: 0.7936 - val_acc: 0.9153\n",
      "Epoch 72/100\n",
      "48000/48000 [==============================] - 5s 94us/step - loss: 0.0112 - acc: 0.9967 - val_loss: 0.7679 - val_acc: 0.9159\n",
      "Epoch 73/100\n",
      "48000/48000 [==============================] - 4s 94us/step - loss: 0.0161 - acc: 0.9954 - val_loss: 0.7844 - val_acc: 0.9177\n",
      "Epoch 74/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0095 - acc: 0.9972 - val_loss: 0.7772 - val_acc: 0.9197\n",
      "Epoch 75/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0189 - acc: 0.9950 - val_loss: 0.8062 - val_acc: 0.9171\n",
      "Epoch 76/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0173 - acc: 0.9958 - val_loss: 0.8122 - val_acc: 0.9133\n",
      "Epoch 77/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0128 - acc: 0.9964 - val_loss: 0.7841 - val_acc: 0.9191\n",
      "Epoch 78/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0104 - acc: 0.9969 - val_loss: 0.8137 - val_acc: 0.9153\n",
      "Epoch 79/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0162 - acc: 0.9958 - val_loss: 0.7617 - val_acc: 0.9189\n",
      "Epoch 80/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0143 - acc: 0.9965 - val_loss: 0.8117 - val_acc: 0.9187\n",
      "Epoch 81/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0151 - acc: 0.9961 - val_loss: 0.8305 - val_acc: 0.9135\n",
      "Epoch 82/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0135 - acc: 0.9967 - val_loss: 0.7885 - val_acc: 0.9224\n",
      "Epoch 83/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0140 - acc: 0.9962 - val_loss: 0.7784 - val_acc: 0.9214\n",
      "Epoch 84/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0132 - acc: 0.9964 - val_loss: 0.8182 - val_acc: 0.9216\n",
      "Epoch 85/100\n",
      "48000/48000 [==============================] - 5s 94us/step - loss: 0.0089 - acc: 0.9977 - val_loss: 0.8130 - val_acc: 0.9202\n",
      "Epoch 86/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0182 - acc: 0.9954 - val_loss: 0.8093 - val_acc: 0.9196\n",
      "Epoch 87/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0100 - acc: 0.9973 - val_loss: 0.8451 - val_acc: 0.9194\n",
      "Epoch 88/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0126 - acc: 0.9969 - val_loss: 0.8373 - val_acc: 0.9189\n",
      "Epoch 89/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0139 - acc: 0.9966 - val_loss: 0.8321 - val_acc: 0.9194\n",
      "Epoch 90/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0139 - acc: 0.9967 - val_loss: 0.8514 - val_acc: 0.9147\n",
      "Epoch 91/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0157 - acc: 0.9962 - val_loss: 0.8218 - val_acc: 0.9191\n",
      "Epoch 92/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0082 - acc: 0.9976 - val_loss: 0.8061 - val_acc: 0.9196\n",
      "Epoch 93/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0123 - acc: 0.9969 - val_loss: 0.8923 - val_acc: 0.9125\n",
      "Epoch 94/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0164 - acc: 0.9960 - val_loss: 0.8640 - val_acc: 0.9177\n",
      "Epoch 95/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0122 - acc: 0.9970 - val_loss: 0.8239 - val_acc: 0.9175\n",
      "Epoch 96/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0109 - acc: 0.9974 - val_loss: 0.8240 - val_acc: 0.9197\n",
      "Epoch 97/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0171 - acc: 0.9964 - val_loss: 0.8706 - val_acc: 0.9155\n",
      "Epoch 98/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0107 - acc: 0.9975 - val_loss: 0.8322 - val_acc: 0.9212\n",
      "Epoch 99/100\n",
      "48000/48000 [==============================] - 4s 93us/step - loss: 0.0111 - acc: 0.9971 - val_loss: 0.8252 - val_acc: 0.9205\n",
      "Epoch 100/100\n",
      "48000/48000 [==============================] - 4s 92us/step - loss: 0.0093 - acc: 0.9973 - val_loss: 0.8419 - val_acc: 0.9191\n"
     ]
    }
   ],
   "source": [
    "# Training the model for 100 epochs using the keras fit function and visualizing the accuracy and loss curves\n",
    "\n",
    "history_classification = full_model.fit(X_train, y_trainc, batch_size = 64, epochs= 100,\n",
    "                                   verbose =1, validation_data = (X_valc, y_valc))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Saving the classification model's weights\n",
    "full_model.save_weights('full_model_weights.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Retraining the full model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Now we will retrain the full model along with the other 19 layers\n",
    "for layer in full_model.layers[:19]:\n",
    "    layer.trainable = True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Now, we will compile the model\n",
    "full_model.compile(loss ='categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 48000 samples, validate on 12000 samples\n",
      "Epoch 1/100\n",
      "48000/48000 [==============================] - 10s 218us/step - loss: 0.2806 - acc: 0.9544 - val_loss: 0.8130 - val_acc: 0.8846\n",
      "Epoch 2/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.1700 - acc: 0.9629 - val_loss: 0.6177 - val_acc: 0.8960\n",
      "Epoch 3/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.1220 - acc: 0.9693 - val_loss: 0.5479 - val_acc: 0.9109\n",
      "Epoch 4/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0920 - acc: 0.9751 - val_loss: 0.5407 - val_acc: 0.9066\n",
      "Epoch 5/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0748 - acc: 0.9789 - val_loss: 0.5832 - val_acc: 0.9129\n",
      "Epoch 6/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0791 - acc: 0.9777 - val_loss: 0.5490 - val_acc: 0.9111\n",
      "Epoch 7/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0891 - acc: 0.9760 - val_loss: 0.5278 - val_acc: 0.9118\n",
      "Epoch 8/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0807 - acc: 0.9778 - val_loss: 0.4718 - val_acc: 0.9203\n",
      "Epoch 9/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0721 - acc: 0.9801 - val_loss: 0.5328 - val_acc: 0.9133\n",
      "Epoch 10/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0542 - acc: 0.9844 - val_loss: 0.4773 - val_acc: 0.9165\n",
      "Epoch 11/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0510 - acc: 0.9845 - val_loss: 0.4677 - val_acc: 0.9222\n",
      "Epoch 12/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0372 - acc: 0.9881 - val_loss: 0.5023 - val_acc: 0.9203\n",
      "Epoch 13/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0570 - acc: 0.9835 - val_loss: 0.4406 - val_acc: 0.9176\n",
      "Epoch 14/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0486 - acc: 0.9855 - val_loss: 0.4553 - val_acc: 0.9202\n",
      "Epoch 15/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0367 - acc: 0.9882 - val_loss: 0.5276 - val_acc: 0.9139\n",
      "Epoch 16/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0455 - acc: 0.9874 - val_loss: 0.4875 - val_acc: 0.9155\n",
      "Epoch 17/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0386 - acc: 0.9883 - val_loss: 0.5237 - val_acc: 0.9178\n",
      "Epoch 18/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0357 - acc: 0.9897 - val_loss: 0.5353 - val_acc: 0.9204\n",
      "Epoch 19/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0394 - acc: 0.9888 - val_loss: 0.4841 - val_acc: 0.9232\n",
      "Epoch 20/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0293 - acc: 0.9913 - val_loss: 0.5035 - val_acc: 0.9197\n",
      "Epoch 21/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0343 - acc: 0.9899 - val_loss: 0.5196 - val_acc: 0.9233\n",
      "Epoch 22/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0229 - acc: 0.9924 - val_loss: 0.4998 - val_acc: 0.9235\n",
      "Epoch 23/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0220 - acc: 0.9929 - val_loss: 0.5426 - val_acc: 0.9224\n",
      "Epoch 24/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0298 - acc: 0.9905 - val_loss: 0.5246 - val_acc: 0.9202\n",
      "Epoch 25/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0419 - acc: 0.9882 - val_loss: 0.5090 - val_acc: 0.9182\n",
      "Epoch 26/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0236 - acc: 0.9927 - val_loss: 0.5114 - val_acc: 0.9273\n",
      "Epoch 27/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0204 - acc: 0.9941 - val_loss: 0.5271 - val_acc: 0.9262\n",
      "Epoch 28/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0188 - acc: 0.9947 - val_loss: 0.6044 - val_acc: 0.9160\n",
      "Epoch 29/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0323 - acc: 0.9905 - val_loss: 0.5270 - val_acc: 0.9238\n",
      "Epoch 30/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0193 - acc: 0.9942 - val_loss: 0.5293 - val_acc: 0.9230\n",
      "Epoch 31/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0255 - acc: 0.9925 - val_loss: 0.5049 - val_acc: 0.9270\n",
      "Epoch 32/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0166 - acc: 0.9950 - val_loss: 0.5270 - val_acc: 0.9263\n",
      "Epoch 33/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0199 - acc: 0.9938 - val_loss: 0.5718 - val_acc: 0.9219\n",
      "Epoch 34/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0257 - acc: 0.9922 - val_loss: 0.5663 - val_acc: 0.9205\n",
      "Epoch 35/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0185 - acc: 0.9941 - val_loss: 0.6011 - val_acc: 0.9200\n",
      "Epoch 36/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0234 - acc: 0.9939 - val_loss: 0.5625 - val_acc: 0.9213\n",
      "Epoch 37/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0225 - acc: 0.9934 - val_loss: 0.5394 - val_acc: 0.9220\n",
      "Epoch 38/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0153 - acc: 0.9956 - val_loss: 0.5358 - val_acc: 0.9242\n",
      "Epoch 39/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0163 - acc: 0.9951 - val_loss: 0.5471 - val_acc: 0.9227\n",
      "Epoch 40/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0220 - acc: 0.9939 - val_loss: 0.5814 - val_acc: 0.9231\n",
      "Epoch 41/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0169 - acc: 0.9952 - val_loss: 0.5347 - val_acc: 0.9268\n",
      "Epoch 42/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0111 - acc: 0.9967 - val_loss: 0.5489 - val_acc: 0.9243\n",
      "Epoch 43/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0122 - acc: 0.9965 - val_loss: 0.5446 - val_acc: 0.9278\n",
      "Epoch 44/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0213 - acc: 0.9938 - val_loss: 0.5565 - val_acc: 0.9257\n",
      "Epoch 45/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0152 - acc: 0.9954 - val_loss: 0.5488 - val_acc: 0.9231\n",
      "Epoch 46/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0174 - acc: 0.9950 - val_loss: 0.5959 - val_acc: 0.9224\n",
      "Epoch 47/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0133 - acc: 0.9959 - val_loss: 0.5976 - val_acc: 0.9207\n",
      "Epoch 48/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0161 - acc: 0.9956 - val_loss: 0.5781 - val_acc: 0.9263\n",
      "Epoch 49/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0164 - acc: 0.9958 - val_loss: 0.5783 - val_acc: 0.9201\n",
      "Epoch 50/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0151 - acc: 0.9959 - val_loss: 0.6142 - val_acc: 0.9250\n",
      "Epoch 51/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0134 - acc: 0.9960 - val_loss: 0.5954 - val_acc: 0.9226\n",
      "Epoch 52/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0131 - acc: 0.9960 - val_loss: 0.6082 - val_acc: 0.9212\n",
      "Epoch 53/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0185 - acc: 0.9950 - val_loss: 0.6427 - val_acc: 0.9202\n",
      "Epoch 54/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0162 - acc: 0.9959 - val_loss: 0.5963 - val_acc: 0.9207\n",
      "Epoch 55/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0099 - acc: 0.9971 - val_loss: 0.6136 - val_acc: 0.9280\n",
      "Epoch 56/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0112 - acc: 0.9967 - val_loss: 0.6520 - val_acc: 0.9227\n",
      "Epoch 57/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0143 - acc: 0.9961 - val_loss: 0.6374 - val_acc: 0.9217\n",
      "Epoch 58/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0179 - acc: 0.9955 - val_loss: 0.6429 - val_acc: 0.9222\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 59/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0118 - acc: 0.9964 - val_loss: 0.6668 - val_acc: 0.9224\n",
      "Epoch 60/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0114 - acc: 0.9970 - val_loss: 0.6157 - val_acc: 0.9250\n",
      "Epoch 61/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0112 - acc: 0.9968 - val_loss: 0.6240 - val_acc: 0.9260\n",
      "Epoch 62/100\n",
      "48000/48000 [==============================] - 10s 204us/step - loss: 0.0154 - acc: 0.9955 - val_loss: 0.6530 - val_acc: 0.9223\n",
      "Epoch 63/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0088 - acc: 0.9971 - val_loss: 0.6149 - val_acc: 0.9267\n",
      "Epoch 64/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0119 - acc: 0.9965 - val_loss: 0.6687 - val_acc: 0.9190\n",
      "Epoch 65/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0115 - acc: 0.9966 - val_loss: 0.5873 - val_acc: 0.9296\n",
      "Epoch 66/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0107 - acc: 0.9975 - val_loss: 0.6746 - val_acc: 0.9249\n",
      "Epoch 67/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0186 - acc: 0.9957 - val_loss: 0.6106 - val_acc: 0.9242\n",
      "Epoch 68/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0100 - acc: 0.9973 - val_loss: 0.6170 - val_acc: 0.9227\n",
      "Epoch 69/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0107 - acc: 0.9970 - val_loss: 0.7122 - val_acc: 0.9148\n",
      "Epoch 70/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0193 - acc: 0.9953 - val_loss: 0.6682 - val_acc: 0.9236\n",
      "Epoch 71/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0155 - acc: 0.9964 - val_loss: 0.5945 - val_acc: 0.9242\n",
      "Epoch 72/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0076 - acc: 0.9981 - val_loss: 0.6048 - val_acc: 0.9275\n",
      "Epoch 73/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0115 - acc: 0.9970 - val_loss: 0.7171 - val_acc: 0.9191\n",
      "Epoch 74/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0103 - acc: 0.9973 - val_loss: 0.5945 - val_acc: 0.9233\n",
      "Epoch 75/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0081 - acc: 0.9979 - val_loss: 0.6262 - val_acc: 0.9237\n",
      "Epoch 76/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0109 - acc: 0.9970 - val_loss: 0.6340 - val_acc: 0.9268\n",
      "Epoch 77/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0122 - acc: 0.9965 - val_loss: 0.6089 - val_acc: 0.9225\n",
      "Epoch 78/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0127 - acc: 0.9965 - val_loss: 0.6111 - val_acc: 0.9219\n",
      "Epoch 79/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0063 - acc: 0.9982 - val_loss: 0.6112 - val_acc: 0.9249\n",
      "Epoch 80/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0120 - acc: 0.9965 - val_loss: 0.6312 - val_acc: 0.9254\n",
      "Epoch 81/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0082 - acc: 0.9977 - val_loss: 0.6340 - val_acc: 0.9253\n",
      "Epoch 82/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0097 - acc: 0.9973 - val_loss: 0.5918 - val_acc: 0.9262\n",
      "Epoch 83/100\n",
      "48000/48000 [==============================] - 10s 203us/step - loss: 0.0095 - acc: 0.9974 - val_loss: 0.6127 - val_acc: 0.9263\n",
      "Epoch 84/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0062 - acc: 0.9981 - val_loss: 0.6394 - val_acc: 0.9247\n",
      "Epoch 85/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0095 - acc: 0.9978 - val_loss: 0.6199 - val_acc: 0.9233\n",
      "Epoch 86/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0115 - acc: 0.9970 - val_loss: 0.6872 - val_acc: 0.9226\n",
      "Epoch 87/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0095 - acc: 0.9972 - val_loss: 0.6431 - val_acc: 0.9261\n",
      "Epoch 88/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0103 - acc: 0.9970 - val_loss: 0.5833 - val_acc: 0.9266\n",
      "Epoch 89/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0032 - acc: 0.9989 - val_loss: 0.6515 - val_acc: 0.9276\n",
      "Epoch 90/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0053 - acc: 0.9984 - val_loss: 0.6297 - val_acc: 0.9244\n",
      "Epoch 91/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0148 - acc: 0.9963 - val_loss: 0.6433 - val_acc: 0.9250\n",
      "Epoch 92/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0133 - acc: 0.9968 - val_loss: 0.6077 - val_acc: 0.9289\n",
      "Epoch 93/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0069 - acc: 0.9982 - val_loss: 0.6324 - val_acc: 0.9282\n",
      "Epoch 94/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0052 - acc: 0.9984 - val_loss: 0.6531 - val_acc: 0.9240\n",
      "Epoch 95/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0096 - acc: 0.9973 - val_loss: 0.5882 - val_acc: 0.9214\n",
      "Epoch 96/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0089 - acc: 0.9978 - val_loss: 0.6732 - val_acc: 0.9220\n",
      "Epoch 97/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0107 - acc: 0.9972 - val_loss: 0.6309 - val_acc: 0.9252\n",
      "Epoch 98/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0099 - acc: 0.9975 - val_loss: 0.6235 - val_acc: 0.9253\n",
      "Epoch 99/100\n",
      "48000/48000 [==============================] - 10s 202us/step - loss: 0.0045 - acc: 0.9987 - val_loss: 0.6292 - val_acc: 0.9278\n",
      "Epoch 100/100\n",
      "48000/48000 [==============================] - 10s 201us/step - loss: 0.0083 - acc: 0.9976 - val_loss: 0.7161 - val_acc: 0.9234\n"
     ]
    }
   ],
   "source": [
    "history_classification = full_model.fit(X_train, y_trainc, batch_size = 64, epochs= 100,\n",
    "                                   verbose =1, validation_data = (X_valc, y_valc))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Saving the complete model now\n",
    "full_model.save_weights('full_model_complete.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'val_acc', 'loss', 'acc'])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8a845a9940>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8a841b7588>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# plotting the accuracy and loss curves\n",
    "# Now we can plot the loss curves to visualize the model performance\n",
    "# list all data in history\n",
    "print(history_classification.history.keys())\n",
    "\n",
    "# summarize history for loss\n",
    "plt.plot(history_classification.history['acc'])\n",
    "plt.plot(history_classification.history['val_acc'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper right')\n",
    "plt.show()\n",
    "\n",
    "# summarize history for loss\n",
    "plt.plot(history_classification.history['loss'])\n",
    "plt.plot(history_classification.history['val_loss'])\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Model evaluation on the test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 1s 93us/step\n"
     ]
    }
   ],
   "source": [
    "test_eval = full_model.evaluate(test_data, X_test_one_hot, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Accuracy 0.743126664990908\n",
      "Test Loss 0.919\n"
     ]
    }
   ],
   "source": [
    "# Let us see the test accuracy and loss\n",
    "print('Test Accuracy', test_eval[0])\n",
    "print('Test Loss', test_eval[1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1.1257468e-22 0.0000000e+00 1.0000000e+00 2.0259230e-35 2.9935192e-32\n",
      " 0.0000000e+00 1.1228510e-31 0.0000000e+00 0.0000000e+00 0.0000000e+00]\n"
     ]
    }
   ],
   "source": [
    "predicted_classes = full_model.predict(test_data)\n",
    "print(predicted_classes[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "predicted_classes = np.argmax(np.round(predicted_classes), axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2\n"
     ]
    }
   ],
   "source": [
    "print(predicted_classes[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((10000,), (10000,))"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predicted_classes.shape, test_labels.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Correctly identified samples are:  9189\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8a84565cf8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# finding the correct predicted labels\n",
    "correct = np.where(predicted_classes == test_labels)[0]\n",
    "\n",
    "print(\"Correctly identified samples are: \", len(correct))\n",
    "for i,correct in enumerate(correct[:9]):\n",
    "    plt.subplot(3,3,i+1)\n",
    "    plt.imshow(test_data[correct].reshape(28,28), cmap = 'gray', interpolation = 'none')\n",
    "    plt.title(\"Predicted {}, Class {}\".format(predicted_classes[correct], test_labels[correct]))\n",
    "    plt.tight_layout()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Incorrectly identified samples are:  811\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8a5db5cf28>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# finding the incorrect predictions\n",
    "incorrect = np.where(predicted_classes != test_labels)[0]\n",
    "\n",
    "print(\"Incorrectly identified samples are: \", len(incorrect))\n",
    "for i, incorrect in enumerate(incorrect[:9]):\n",
    "    plt.subplot(3,3,i+1)\n",
    "    plt.imshow(test_data[incorrect].reshape(28,28), cmap = 'gray', interpolation = 'none')\n",
    "    plt.title(\"Predicted {} class {}\".format(predicted_classes[incorrect], test_labels[incorrect]))\n",
    "    plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy:  0.9189\n"
     ]
    }
   ],
   "source": [
    "# Accuracy\n",
    "print(\"Accuracy: \", accuracy_score(test_labels, predicted_classes))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "             precision    recall  f1-score   support\n",
      "\n",
      "    Class 0       0.85      0.88      0.86      1000\n",
      "    Class 1       0.99      0.98      0.99      1000\n",
      "    Class 2       0.88      0.90      0.89      1000\n",
      "    Class 3       0.89      0.93      0.91      1000\n",
      "    Class 4       0.84      0.90      0.87      1000\n",
      "    Class 5       0.99      0.98      0.99      1000\n",
      "    Class 6       0.82      0.71      0.76      1000\n",
      "    Class 7       0.96      0.97      0.97      1000\n",
      "    Class 8       0.99      0.98      0.98      1000\n",
      "    Class 9       0.97      0.96      0.97      1000\n",
      "\n",
      "avg / total       0.92      0.92      0.92     10000\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Classification report\n",
    "target_names = [\"Class {}\".format(i) for i in range(num_classes)]\n",
    "print(classification_report(test_labels, predicted_classes, target_names=target_names))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_confusion_matrix(cm, classes,\n",
    "                          normalize=False,\n",
    "                          title='Confusion matrix',\n",
    "                          cmap=plt.cm.Blues):\n",
    "    \"\"\"\n",
    "    This function prints and plots the confusion matrix.\n",
    "    Normalization can be applied by setting `normalize=True`.\n",
    "    \"\"\"\n",
    "#     if normalize:\n",
    "#         cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "#         print(\"Normalized confusion matrix\")\n",
    "#     else:\n",
    "#         print('Confusion matrix, without normalization')\n",
    "\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "    tick_marks = np.arange(len(classes))\n",
    "    plt.xticks(tick_marks, classes, rotation=45)\n",
    "    plt.yticks(tick_marks, classes)\n",
    "\n",
    "    fmt = '.2f' if normalize else 'd'\n",
    "    thresh = cm.max() / 2.\n",
    "    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n",
    "        plt.text(j, i, format(cm[i, j], fmt),\n",
    "                 horizontalalignment=\"center\",\n",
    "                 color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "\n",
    "    plt.tight_layout()\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], dtype=uint64)"
      ]
     },
     "execution_count": 81,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.unique(test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8989531b70>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f898b8e2b38>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "class_names = np.unique(test_labels)\n",
    "\n",
    "# Compute confusion matrix\n",
    "cnf_matrix = confusion_matrix(test_labels, predicted_classes)\n",
    "np.set_printoptions(precision=2)\n",
    "\n",
    "# Plot non-normalized confusion matrix\n",
    "plt.figure()\n",
    "plot_confusion_matrix(cnf_matrix, classes=class_names,\n",
    "                      title='Confusion matrix, without normalization')\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "plt.figure()\n",
    "plot_confusion_matrix(cnf_matrix, classes=class_names, normalize=True,\n",
    "                      title='Normalized confusion matrix')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We learned how to use an autoencoder to classify images with python and keras. "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}