subst_LReLU_106x79.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Substitution with smaller LeakyReLu, Conv-s and ZeroPadding\n",
    "### This is very first attempt with LeakyReLU\n",
    "Substitution presented here is very well described under this link:\n",
    "\n",
    "[https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/convolutional_neural_networks.html](https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/convolutional_neural_networks.html)\n",
    "\n",
    "Each image 106x79x1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[name: \"/device:CPU:0\"\n",
       " device_type: \"CPU\"\n",
       " memory_limit: 268435456\n",
       " locality {\n",
       " }\n",
       " incarnation: 16464584643074866572, name: \"/device:GPU:0\"\n",
       " device_type: \"GPU\"\n",
       " memory_limit: 356122624\n",
       " locality {\n",
       "   bus_id: 1\n",
       " }\n",
       " incarnation: 15010918190853181526\n",
       " physical_device_desc: \"device: 0, name: Tesla K80, pci bus id: 0000:00:1e.0, compute capability: 3.7\"]"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow.python.client import device_lib\n",
    "sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))\n",
    "device_lib.list_local_devices()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Conv2D, MaxPooling2D, ZeroPadding2D\n",
    "from keras.layers import Activation, Dropout, Flatten, Dense\n",
    "from keras.layers.advanced_activations import LeakyReLU, PReLU\n",
    "from keras.callbacks import ModelCheckpoint\n",
    "from keras.utils.vis_utils import plot_model\n",
    "from sklearn.utils import shuffle\n",
    "from keras import backend as K\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "width, height = 106, 79\n",
    "train_data_dir = '../gen_x3_3/train'\n",
    "test_data_dir = '../gen_x3_3/test'\n",
    "nb_train_samples = 47232\n",
    "nb_test_samples = 11808\n",
    "epochs = 20\n",
    "batch_size = 1024\n",
    "\n",
    "input_shape = (width, height, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /usr/local/lib/python3.5/dist-packages/keras/backend/tensorflow_backend.py:1264: calling reduce_prod (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "keep_dims is deprecated, use keepdims instead\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "model.add(ZeroPadding2D(padding=(4, 4), name='zeropad1', input_shape=input_shape))\n",
    "model.add(Conv2D(32, (9, 9), strides=(2, 2), padding='same', name='conv1'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same', name='maxpool1'))\n",
    "\n",
    "# same as:\n",
    "# ZeroPadding2D(padding(4, 4))\n",
    "# Conv2D(64, kernel_size(9, 9), strides=(1, 1), padding='same')\n",
    "model.add(ZeroPadding2D(padding=(1, 1), name='block1zeropad1'))\n",
    "model.add(Conv2D(64, kernel_size=(3, 3), strides=(1, 1), padding='same', name='block1_conv1'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(ZeroPadding2D(padding=(1, 1), name='block1zeropad2'))\n",
    "model.add(Conv2D(64, kernel_size=(3, 3), strides=(1, 1), padding='same', name='block1_conv2'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(ZeroPadding2D(padding=(1, 1), name='block1zeropad3'))\n",
    "model.add(Conv2D(64, kernel_size=(3, 3), strides=(1, 1), padding='same', name='block1_conv3'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(ZeroPadding2D(padding=(1, 1), name='block1zeropad4'))\n",
    "model.add(Conv2D(64, kernel_size=(3, 3), strides=(1, 1), padding='same', name='block1_conv4'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same', name='maxpool2'))\n",
    "\n",
    "# same as:\n",
    "# ZeroPadding2D(padding(1, 1))\n",
    "# model.add(Conv2D(128, kernel_size=(3, 3), strides=(1, 1), padding='same'))\n",
    "model.add(Conv2D(64, kernel_size=(1, 1), strides=(1, 1), padding='same', name='block2_conv1'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(ZeroPadding2D(padding=(1, 1), name='block2zeropad2'))\n",
    "model.add(Conv2D(64, kernel_size=(3, 3), strides=(1, 1), padding='valid', name='block2_conv2'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(Conv2D(128, kernel_size=(1, 1), strides=(1, 1), padding='same', name='block2_conv3'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same', name='maxpool3'))\n",
    "\n",
    "# same as:\n",
    "# ZeroPadding2D(padding(1, 1))\n",
    "# model.add(Conv2D(128, kernel_size=(3, 3), strides=(1, 1), padding='same'))\n",
    "model.add(Conv2D(64, kernel_size=(1, 1), strides=(1, 1), padding='same', name='block3_conv1'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(ZeroPadding2D(padding=(1, 1), name='block3zeropad1'))\n",
    "model.add(Conv2D(64, kernel_size=(3, 3), strides=(1, 1), padding='valid', name='block3_conv2'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(Conv2D(128, kernel_size=(1, 1), strides=(1, 1), padding='same', name='block3_conv3'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same', name='maxpool4'))\n",
    "\n",
    "model.add(Conv2D(128, kernel_size=(3, 3), strides=(1, 1), padding='same', name='block4_conv1'))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(Flatten())\n",
    "\n",
    "model.add(Dense(1024))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(Dropout(0.2))\n",
    "\n",
    "model.add(Dense(512))\n",
    "model.add(LeakyReLU(alpha=1e-2))\n",
    "model.add(Dropout(0.2))\n",
    "\n",
    "model.add(Dense(128, activation='relu'))\n",
    "model.add(Dense(6, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /usr/local/lib/python3.5/dist-packages/keras/backend/tensorflow_backend.py:2885: calling reduce_sum (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "keep_dims is deprecated, use keepdims instead\n",
      "WARNING:tensorflow:From /usr/local/lib/python3.5/dist-packages/keras/backend/tensorflow_backend.py:1349: calling reduce_mean (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "keep_dims is deprecated, use keepdims instead\n"
     ]
    }
   ],
   "source": [
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adam',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "zeropad1 (ZeroPadding2D)     (None, 114, 87, 1)        0         \n",
      "_________________________________________________________________\n",
      "conv1 (Conv2D)               (None, 57, 44, 32)        2624      \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_1 (LeakyReLU)    (None, 57, 44, 32)        0         \n",
      "_________________________________________________________________\n",
      "maxpool1 (MaxPooling2D)      (None, 29, 22, 32)        0         \n",
      "_________________________________________________________________\n",
      "block1zeropad1 (ZeroPadding2 (None, 31, 24, 32)        0         \n",
      "_________________________________________________________________\n",
      "block1_conv1 (Conv2D)        (None, 31, 24, 64)        18496     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_2 (LeakyReLU)    (None, 31, 24, 64)        0         \n",
      "_________________________________________________________________\n",
      "block1zeropad2 (ZeroPadding2 (None, 33, 26, 64)        0         \n",
      "_________________________________________________________________\n",
      "block1_conv2 (Conv2D)        (None, 33, 26, 64)        36928     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_3 (LeakyReLU)    (None, 33, 26, 64)        0         \n",
      "_________________________________________________________________\n",
      "block1zeropad3 (ZeroPadding2 (None, 35, 28, 64)        0         \n",
      "_________________________________________________________________\n",
      "block1_conv3 (Conv2D)        (None, 35, 28, 64)        36928     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_4 (LeakyReLU)    (None, 35, 28, 64)        0         \n",
      "_________________________________________________________________\n",
      "block1zeropad4 (ZeroPadding2 (None, 37, 30, 64)        0         \n",
      "_________________________________________________________________\n",
      "block1_conv4 (Conv2D)        (None, 37, 30, 64)        36928     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_5 (LeakyReLU)    (None, 37, 30, 64)        0         \n",
      "_________________________________________________________________\n",
      "maxpool2 (MaxPooling2D)      (None, 19, 15, 64)        0         \n",
      "_________________________________________________________________\n",
      "block2_conv1 (Conv2D)        (None, 19, 15, 64)        4160      \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_6 (LeakyReLU)    (None, 19, 15, 64)        0         \n",
      "_________________________________________________________________\n",
      "block2zeropad2 (ZeroPadding2 (None, 21, 17, 64)        0         \n",
      "_________________________________________________________________\n",
      "block2_conv2 (Conv2D)        (None, 19, 15, 64)        36928     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_7 (LeakyReLU)    (None, 19, 15, 64)        0         \n",
      "_________________________________________________________________\n",
      "block2_conv3 (Conv2D)        (None, 19, 15, 128)       8320      \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_8 (LeakyReLU)    (None, 19, 15, 128)       0         \n",
      "_________________________________________________________________\n",
      "maxpool3 (MaxPooling2D)      (None, 10, 8, 128)        0         \n",
      "_________________________________________________________________\n",
      "block3_conv1 (Conv2D)        (None, 10, 8, 64)         8256      \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_9 (LeakyReLU)    (None, 10, 8, 64)         0         \n",
      "_________________________________________________________________\n",
      "block3zeropad1 (ZeroPadding2 (None, 12, 10, 64)        0         \n",
      "_________________________________________________________________\n",
      "block3_conv2 (Conv2D)        (None, 10, 8, 64)         36928     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_10 (LeakyReLU)   (None, 10, 8, 64)         0         \n",
      "_________________________________________________________________\n",
      "block3_conv3 (Conv2D)        (None, 10, 8, 128)        8320      \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_11 (LeakyReLU)   (None, 10, 8, 128)        0         \n",
      "_________________________________________________________________\n",
      "maxpool4 (MaxPooling2D)      (None, 5, 4, 128)         0         \n",
      "_________________________________________________________________\n",
      "block4_conv1 (Conv2D)        (None, 5, 4, 128)         147584    \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_12 (LeakyReLU)   (None, 5, 4, 128)         0         \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 2560)              0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 1024)              2622464   \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_13 (LeakyReLU)   (None, 1024)              0         \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 1024)              0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 512)               524800    \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_14 (LeakyReLU)   (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 128)               65664     \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 6)                 774       \n",
      "=================================================================\n",
      "Total params: 3,596,102\n",
      "Trainable params: 3,596,102\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "# summarize model and apply checkpoints\n",
    "print(model.summary())\n",
    "filepath = 'subst_LReLU/subst_LReLU_106x79-{epoch:02d}-{loss:.4f}.h5'\n",
    "checkpoint = ModelCheckpoint(filepath, monitor='loss', verbose=1, save_best_only=True, mode='min')\n",
    "callbacks_list = [checkpoint]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_datagen = ImageDataGenerator(\n",
    "    rescale=1. / 255,\n",
    "    shear_range=0.20,\n",
    "    zoom_range=0.15,\n",
    "    fill_mode='nearest')\n",
    "test_datagen = ImageDataGenerator(rescale=1. / 255)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 47232 images belonging to 6 classes.\n",
      "Found 11808 images belonging to 6 classes.\n"
     ]
    }
   ],
   "source": [
    "train_generator = train_datagen.flow_from_directory(\n",
    "    train_data_dir,\n",
    "    target_size=(width, height),\n",
    "    batch_size=batch_size,\n",
    "    color_mode='grayscale',\n",
    "    class_mode='categorical',\n",
    "    classes=['1', '2', '3', '4', '5', '6'],\n",
    "    shuffle=True,\n",
    "    seed=2018)\n",
    "\n",
    "test_generator = test_datagen.flow_from_directory(\n",
    "    test_data_dir,\n",
    "    target_size=(width, height),\n",
    "    batch_size=batch_size,\n",
    "    color_mode='grayscale',\n",
    "    class_mode='categorical',\n",
    "    classes=['1', '2', '3', '4', '5', '6'],\n",
    "    shuffle=True,\n",
    "    seed=2018)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7919 - acc: 0.1656Epoch 00001: loss improved from inf to 1.79191, saving model to subst_LReLU/subst_LReLU_106x79-01-1.7919.h5\n",
      "46/46 [==============================] - 61s 1s/step - loss: 1.7919 - acc: 0.1656 - val_loss: 1.7918 - val_acc: 0.1673\n",
      "Epoch 2/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1659Epoch 00002: loss improved from 1.79191 to 1.79181, saving model to subst_LReLU/subst_LReLU_106x79-02-1.7918.h5\n",
      "46/46 [==============================] - 54s 1s/step - loss: 1.7918 - acc: 0.1660 - val_loss: 1.7918 - val_acc: 0.1678\n",
      "Epoch 3/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7919 - acc: 0.1659Epoch 00003: loss did not improve\n",
      "46/46 [==============================] - 55s 1s/step - loss: 1.7919 - acc: 0.1659 - val_loss: 1.7918 - val_acc: 0.1669\n",
      "Epoch 4/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7919 - acc: 0.1637Epoch 00004: loss did not improve\n",
      "46/46 [==============================] - 53s 1s/step - loss: 1.7919 - acc: 0.1643 - val_loss: 1.7918 - val_acc: 0.1673\n",
      "Epoch 5/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1670Epoch 00005: loss improved from 1.79181 to 1.79180, saving model to subst_LReLU/subst_LReLU_106x79-05-1.7918.h5\n",
      "46/46 [==============================] - 55s 1s/step - loss: 1.7918 - acc: 0.1670 - val_loss: 1.7918 - val_acc: 0.1665\n",
      "Epoch 6/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1674Epoch 00006: loss improved from 1.79180 to 1.79179, saving model to subst_LReLU/subst_LReLU_106x79-06-1.7918.h5\n",
      "46/46 [==============================] - 54s 1s/step - loss: 1.7918 - acc: 0.1675 - val_loss: 1.7918 - val_acc: 0.1659\n",
      "Epoch 7/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1613Epoch 00007: loss did not improve\n",
      "46/46 [==============================] - 56s 1s/step - loss: 1.7918 - acc: 0.1617 - val_loss: 1.7918 - val_acc: 0.1657\n",
      "Epoch 8/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1638Epoch 00008: loss did not improve\n",
      "46/46 [==============================] - 57s 1s/step - loss: 1.7918 - acc: 0.1643 - val_loss: 1.7918 - val_acc: 0.1657\n",
      "Epoch 9/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1637Epoch 00009: loss improved from 1.79179 to 1.79178, saving model to subst_LReLU/subst_LReLU_106x79-09-1.7918.h5\n",
      "46/46 [==============================] - 56s 1s/step - loss: 1.7918 - acc: 0.1635 - val_loss: 1.7918 - val_acc: 0.1659\n",
      "Epoch 10/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1642Epoch 00010: loss did not improve\n",
      "46/46 [==============================] - 56s 1s/step - loss: 1.7918 - acc: 0.1643 - val_loss: 1.7918 - val_acc: 0.1678\n",
      "Epoch 11/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1659Epoch 00011: loss did not improve\n",
      "46/46 [==============================] - 57s 1s/step - loss: 1.7918 - acc: 0.1655 - val_loss: 1.7918 - val_acc: 0.1657\n",
      "Epoch 12/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1681Epoch 00012: loss did not improve\n",
      "46/46 [==============================] - 56s 1s/step - loss: 1.7918 - acc: 0.1681 - val_loss: 1.7918 - val_acc: 0.1665\n",
      "Epoch 13/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1673Epoch 00013: loss did not improve\n",
      "46/46 [==============================] - 57s 1s/step - loss: 1.7918 - acc: 0.1671 - val_loss: 1.7918 - val_acc: 0.1665\n",
      "Epoch 14/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1638Epoch 00014: loss did not improve\n",
      "46/46 [==============================] - 56s 1s/step - loss: 1.7918 - acc: 0.1636 - val_loss: 1.7918 - val_acc: 0.1673\n",
      "Epoch 15/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1647Epoch 00015: loss did not improve\n",
      "46/46 [==============================] - 56s 1s/step - loss: 1.7918 - acc: 0.1649 - val_loss: 1.7918 - val_acc: 0.1657\n",
      "Epoch 16/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1668Epoch 00016: loss did not improve\n",
      "46/46 [==============================] - 57s 1s/step - loss: 1.7918 - acc: 0.1668 - val_loss: 1.7918 - val_acc: 0.1665\n",
      "Epoch 17/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1655Epoch 00017: loss improved from 1.79178 to 1.79178, saving model to subst_LReLU/subst_LReLU_106x79-17-1.7918.h5\n",
      "46/46 [==============================] - 55s 1s/step - loss: 1.7918 - acc: 0.1657 - val_loss: 1.7918 - val_acc: 0.1665\n",
      "Epoch 18/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1646Epoch 00018: loss improved from 1.79178 to 1.79178, saving model to subst_LReLU/subst_LReLU_106x79-18-1.7918.h5\n",
      "46/46 [==============================] - 57s 1s/step - loss: 1.7918 - acc: 0.1645 - val_loss: 1.7918 - val_acc: 0.1669\n",
      "Epoch 19/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1655Epoch 00019: loss did not improve\n",
      "46/46 [==============================] - 55s 1s/step - loss: 1.7918 - acc: 0.1652 - val_loss: 1.7918 - val_acc: 0.1678\n",
      "Epoch 20/20\n",
      "45/46 [============================>.] - ETA: 1s - loss: 1.7918 - acc: 0.1645Epoch 00020: loss did not improve\n",
      "46/46 [==============================] - 57s 1s/step - loss: 1.7918 - acc: 0.1645 - val_loss: 1.7918 - val_acc: 0.1657\n"
     ]
    }
   ],
   "source": [
    "history = model.fit_generator(\n",
    "    train_generator,\n",
    "    steps_per_epoch=nb_train_samples // batch_size,\n",
    "    epochs=epochs,\n",
    "    verbose=1,\n",
    "    callbacks=callbacks_list,\n",
    "    shuffle=True,\n",
    "    validation_data=test_generator,\n",
    "    validation_steps=nb_test_samples // batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 16.67%\n",
      "Test loss: 1.7917614451591886\n",
      "Test accuracy 0.16666666666666666\n"
     ]
    }
   ],
   "source": [
    "score = model.evaluate_generator(test_generator)\n",
    "print(\"Accuracy: %.2f%%\" % (score[1]*100))\n",
    "print('Test loss:', score[0])\n",
    "print('Test accuracy', score[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Saved model weights\n"
     ]
    }
   ],
   "source": [
    "model_json = model.to_json()\n",
    "with open('subst_LReLU_106x79_model.json', 'w') as json_file:\n",
    "    json_file.write(model_json)\n",
    "    \n",
    "model.save_weights('subst_LReLU_106x79_weights.h5')\n",
    "print('Saved model weights')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe9115d0588>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#summarize for accuracy\n",
    "plt.plot(history.history['acc'])\n",
    "plt.plot(history.history['val_acc'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe9115d07b8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#summarize 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', 'test'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "plot_model(model, to_file='subst_LReLU/subst_LReLU_106x79_plot.png', show_shapes=True, show_layer_names=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}