train.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## train.ipynb: Define and Train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "# Import the necessary libraries\n",
    "import numpy as np\n",
    "import scipy\n",
    "import os\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Flatten\n",
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras.optimizers import SGD\n",
    "from keras.callbacks import ReduceLROnPlateau, ModelCheckpoint\n",
    "from keras import backend as K\n",
    "K.set_image_dim_ordering('th')\n",
    "from keras.utils import np_utils\n",
    "#from sklearn.cross_validation import StratifiedKFold"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Global Variables\n",
    "# The number of principal components to be retained in the PCA algorithm, \n",
    "# the number of retained features  n\n",
    "numPCAcomponents = 30\n",
    "# Patches windows size\n",
    "windowSize = 5\n",
    "# The proportion of Test sets\n",
    "testRatio = 0.50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load Preprocessed data from file\n",
    "X_train = np.load(\"./predata/XtrainWindowSize\" \n",
    "                  + str(windowSize) + \"PCA\" + str(numPCAcomponents) + \n",
    "                  \"testRatio\" + str(testRatio)  + \".npy\")\n",
    "y_train = np.load(\"./predata/ytrainWindowSize\" \n",
    "                  + str(windowSize) + \"PCA\" + str(numPCAcomponents) + \n",
    "                  \"testRatio\" + str(testRatio) + \".npy\")\n",
    "X_test = np.load(\"./predata/XtestWindowSize\" \n",
    "                  + str(windowSize) + \"PCA\" + str(numPCAcomponents) + \n",
    "                 \"testRatio\" + str(testRatio)  + \".npy\")\n",
    "y_test = np.load(\"./predata/ytestWindowSize\" \n",
    "                  + str(windowSize) + \"PCA\" + str(numPCAcomponents) + \n",
    "                 \"testRatio\" + str(testRatio) + \".npy\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(30, 5, 5)\n"
     ]
    }
   ],
   "source": [
    "# Reshape data into (numberofsumples, channels, height, width)\n",
    "X_train = np.reshape(X_train, (X_train.shape[0], X_train.shape[3], \n",
    "                               X_train.shape[1], X_train.shape[2]))\n",
    "X_test = np.reshape(X_test, (X_test.shape[0],X_test.shape[3], \n",
    "                             X_test.shape[1], X_test.shape[2]))\n",
    "\n",
    "# convert class labels to on-hot encoding\n",
    "y_train = np_utils.to_categorical(y_train)\n",
    "y_test = np_utils.to_categorical(y_test)\n",
    "\n",
    "# Define the input shape \n",
    "input_shape= X_train[0].shape\n",
    "print(input_shape)\n",
    "\n",
    "# number of filters\n",
    "C1 = 3*numPCAcomponents"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define the model structure\n",
    "model = Sequential()\n",
    "\n",
    "model.add(Conv2D(C1, (3, 3), activation='relu', input_shape=input_shape))\n",
    "model.add(Conv2D(3*C1, (3, 3), activation='relu'))\n",
    "model.add(Dropout(0.25))\n",
    "\n",
    "model.add(Flatten())\n",
    "model.add(Dense(6*numPCAcomponents, activation='relu'))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(Dense(16, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define optimization and train method\n",
    "reduce_lr = ReduceLROnPlateau(monitor='val_acc', factor=0.9, patience=25, \n",
    "                              min_lr=0.000001, verbose=1)\n",
    "checkpointer = ModelCheckpoint(filepath=\"checkpoint.hdf5\", verbose=1, \n",
    "                              save_best_only=False)\n",
    "sgd = SGD(lr=0.001, decay=1e-6, momentum=0.9, nesterov=True)\n",
    "model.compile(loss='categorical_crossentropy', optimizer=sgd, \n",
    "                              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:Variable *= will be deprecated. Use variable.assign_mul if you want assignment to the variable value or 'x = x * y' if you want a new python Tensor object.\n",
      "Train on 20110 samples, validate on 5183 samples\n",
      "Epoch 1/100\n",
      "20110/20110 [==============================] - 5s 233us/step - loss: 1.2813 - acc: 0.6164 - val_loss: 0.6084 - val_acc: 0.8057\n",
      "\n",
      "Epoch 00001: saving model to checkpoint.hdf5\n",
      "Epoch 2/100\n",
      "20110/20110 [==============================] - 4s 177us/step - loss: 0.3752 - acc: 0.8783 - val_loss: 0.3269 - val_acc: 0.8956\n",
      "\n",
      "Epoch 00002: saving model to checkpoint.hdf5\n",
      "Epoch 3/100\n",
      "20110/20110 [==============================] - 4s 175us/step - loss: 0.2231 - acc: 0.9304 - val_loss: 0.2492 - val_acc: 0.9168\n",
      "\n",
      "Epoch 00003: saving model to checkpoint.hdf5\n",
      "Epoch 4/100\n",
      "20110/20110 [==============================] - 4s 175us/step - loss: 0.1534 - acc: 0.9529 - val_loss: 0.1856 - val_acc: 0.9429\n",
      "\n",
      "Epoch 00004: saving model to checkpoint.hdf5\n",
      "Epoch 5/100\n",
      "20110/20110 [==============================] - 4s 174us/step - loss: 0.1112 - acc: 0.9662 - val_loss: 0.1563 - val_acc: 0.9518\n",
      "\n",
      "Epoch 00005: saving model to checkpoint.hdf5\n",
      "Epoch 6/100\n",
      "20110/20110 [==============================] - 4s 176us/step - loss: 0.0856 - acc: 0.9748 - val_loss: 0.1393 - val_acc: 0.9510\n",
      "\n",
      "Epoch 00006: saving model to checkpoint.hdf5\n",
      "Epoch 7/100\n",
      "20110/20110 [==============================] - 4s 177us/step - loss: 0.0659 - acc: 0.9807 - val_loss: 0.1164 - val_acc: 0.9628\n",
      "\n",
      "Epoch 00007: saving model to checkpoint.hdf5\n",
      "Epoch 8/100\n",
      "20110/20110 [==============================] - 4s 174us/step - loss: 0.0513 - acc: 0.9866 - val_loss: 0.1127 - val_acc: 0.9618\n",
      "\n",
      "Epoch 00008: saving model to checkpoint.hdf5\n",
      "Epoch 9/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0418 - acc: 0.9889 - val_loss: 0.1063 - val_acc: 0.9637\n",
      "\n",
      "Epoch 00009: saving model to checkpoint.hdf5\n",
      "Epoch 10/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0372 - acc: 0.9902 - val_loss: 0.0968 - val_acc: 0.9689\n",
      "\n",
      "Epoch 00010: saving model to checkpoint.hdf5\n",
      "Epoch 11/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0324 - acc: 0.9912 - val_loss: 0.0882 - val_acc: 0.9714\n",
      "\n",
      "Epoch 00011: saving model to checkpoint.hdf5\n",
      "Epoch 12/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0278 - acc: 0.9933 - val_loss: 0.0910 - val_acc: 0.9703\n",
      "\n",
      "Epoch 00012: saving model to checkpoint.hdf5\n",
      "Epoch 13/100\n",
      "20110/20110 [==============================] - 4s 179us/step - loss: 0.0232 - acc: 0.9946 - val_loss: 0.0852 - val_acc: 0.9730\n",
      "\n",
      "Epoch 00013: saving model to checkpoint.hdf5\n",
      "Epoch 14/100\n",
      "20110/20110 [==============================] - 4s 177us/step - loss: 0.0216 - acc: 0.9950 - val_loss: 0.0834 - val_acc: 0.9728\n",
      "\n",
      "Epoch 00014: saving model to checkpoint.hdf5\n",
      "Epoch 15/100\n",
      "20110/20110 [==============================] - 3s 173us/step - loss: 0.0183 - acc: 0.9956 - val_loss: 0.0842 - val_acc: 0.9726\n",
      "\n",
      "Epoch 00015: saving model to checkpoint.hdf5\n",
      "Epoch 16/100\n",
      "20110/20110 [==============================] - 3s 173us/step - loss: 0.0164 - acc: 0.9966 - val_loss: 0.0833 - val_acc: 0.9751\n",
      "\n",
      "Epoch 00016: saving model to checkpoint.hdf5\n",
      "Epoch 17/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0140 - acc: 0.9973 - val_loss: 0.0827 - val_acc: 0.9732\n",
      "\n",
      "Epoch 00017: saving model to checkpoint.hdf5\n",
      "Epoch 18/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0125 - acc: 0.9975 - val_loss: 0.0805 - val_acc: 0.9753\n",
      "\n",
      "Epoch 00018: saving model to checkpoint.hdf5\n",
      "Epoch 19/100\n",
      "20110/20110 [==============================] - 3s 173us/step - loss: 0.0122 - acc: 0.9975 - val_loss: 0.0789 - val_acc: 0.9745\n",
      "\n",
      "Epoch 00019: saving model to checkpoint.hdf5\n",
      "Epoch 20/100\n",
      "20110/20110 [==============================] - 3s 173us/step - loss: 0.0121 - acc: 0.9976 - val_loss: 0.0768 - val_acc: 0.9753\n",
      "\n",
      "Epoch 00020: saving model to checkpoint.hdf5\n",
      "Epoch 21/100\n",
      "20110/20110 [==============================] - 4s 182us/step - loss: 0.0111 - acc: 0.9976 - val_loss: 0.0776 - val_acc: 0.9757\n",
      "\n",
      "Epoch 00021: saving model to checkpoint.hdf5\n",
      "Epoch 22/100\n",
      "20110/20110 [==============================] - 4s 176us/step - loss: 0.0104 - acc: 0.9975 - val_loss: 0.0781 - val_acc: 0.9749\n",
      "\n",
      "Epoch 00022: saving model to checkpoint.hdf5\n",
      "Epoch 23/100\n",
      "20110/20110 [==============================] - 3s 174us/step - loss: 0.0086 - acc: 0.9985 - val_loss: 0.0771 - val_acc: 0.9759\n",
      "\n",
      "Epoch 00023: saving model to checkpoint.hdf5\n",
      "Epoch 24/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0082 - acc: 0.9986 - val_loss: 0.0761 - val_acc: 0.9755\n",
      "\n",
      "Epoch 00024: saving model to checkpoint.hdf5\n",
      "Epoch 25/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0080 - acc: 0.9986 - val_loss: 0.0785 - val_acc: 0.9761\n",
      "\n",
      "Epoch 00025: saving model to checkpoint.hdf5\n",
      "Epoch 26/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0082 - acc: 0.9983 - val_loss: 0.0771 - val_acc: 0.9776\n",
      "\n",
      "Epoch 00026: saving model to checkpoint.hdf5\n",
      "Epoch 27/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0071 - acc: 0.9988 - val_loss: 0.0770 - val_acc: 0.9778\n",
      "\n",
      "Epoch 00027: saving model to checkpoint.hdf5\n",
      "Epoch 28/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0069 - acc: 0.9987 - val_loss: 0.0773 - val_acc: 0.9782\n",
      "\n",
      "Epoch 00028: saving model to checkpoint.hdf5\n",
      "Epoch 29/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0059 - acc: 0.9993 - val_loss: 0.0783 - val_acc: 0.9770\n",
      "\n",
      "Epoch 00029: saving model to checkpoint.hdf5\n",
      "Epoch 30/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0057 - acc: 0.9991 - val_loss: 0.0783 - val_acc: 0.9757\n",
      "\n",
      "Epoch 00030: saving model to checkpoint.hdf5\n",
      "Epoch 31/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0052 - acc: 0.9993 - val_loss: 0.0789 - val_acc: 0.9765\n",
      "\n",
      "Epoch 00031: saving model to checkpoint.hdf5\n",
      "Epoch 32/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0051 - acc: 0.9994 - val_loss: 0.0775 - val_acc: 0.9774\n",
      "\n",
      "Epoch 00032: saving model to checkpoint.hdf5\n",
      "Epoch 33/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0054 - acc: 0.9989 - val_loss: 0.0767 - val_acc: 0.9778\n",
      "\n",
      "Epoch 00033: saving model to checkpoint.hdf5\n",
      "Epoch 34/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0049 - acc: 0.9993 - val_loss: 0.0773 - val_acc: 0.9776\n",
      "\n",
      "Epoch 00034: saving model to checkpoint.hdf5\n",
      "Epoch 35/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0052 - acc: 0.9991 - val_loss: 0.0771 - val_acc: 0.9786\n",
      "\n",
      "Epoch 00035: saving model to checkpoint.hdf5\n",
      "Epoch 36/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0045 - acc: 0.9992 - val_loss: 0.0751 - val_acc: 0.9782\n",
      "\n",
      "Epoch 00036: saving model to checkpoint.hdf5\n",
      "Epoch 37/100\n",
      "20110/20110 [==============================] - 3s 167us/step - loss: 0.0046 - acc: 0.9993 - val_loss: 0.0752 - val_acc: 0.9776\n",
      "\n",
      "Epoch 00037: saving model to checkpoint.hdf5\n",
      "Epoch 38/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0038 - acc: 0.9993 - val_loss: 0.0763 - val_acc: 0.9782\n",
      "\n",
      "Epoch 00038: saving model to checkpoint.hdf5\n",
      "Epoch 39/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0041 - acc: 0.9993 - val_loss: 0.0766 - val_acc: 0.9780\n",
      "\n",
      "Epoch 00039: saving model to checkpoint.hdf5\n",
      "Epoch 40/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0044 - acc: 0.9989 - val_loss: 0.0763 - val_acc: 0.9794\n",
      "\n",
      "Epoch 00040: saving model to checkpoint.hdf5\n",
      "Epoch 41/100\n",
      "20110/20110 [==============================] - 3s 164us/step - loss: 0.0040 - acc: 0.9995 - val_loss: 0.0755 - val_acc: 0.9774\n",
      "\n",
      "Epoch 00041: saving model to checkpoint.hdf5\n",
      "Epoch 42/100\n",
      "20110/20110 [==============================] - 3s 164us/step - loss: 0.0038 - acc: 0.9992 - val_loss: 0.0741 - val_acc: 0.9782\n",
      "\n",
      "Epoch 00042: saving model to checkpoint.hdf5\n",
      "Epoch 43/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0032 - acc: 0.9997 - val_loss: 0.0762 - val_acc: 0.9772\n",
      "\n",
      "Epoch 00043: saving model to checkpoint.hdf5\n",
      "Epoch 44/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "20110/20110 [==============================] - 4s 189us/step - loss: 0.0034 - acc: 0.9995 - val_loss: 0.0752 - val_acc: 0.9784\n",
      "\n",
      "Epoch 00044: saving model to checkpoint.hdf5\n",
      "Epoch 45/100\n",
      "20110/20110 [==============================] - 4s 190us/step - loss: 0.0032 - acc: 0.9995 - val_loss: 0.0770 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00045: saving model to checkpoint.hdf5\n",
      "Epoch 46/100\n",
      "20110/20110 [==============================] - 4s 191us/step - loss: 0.0029 - acc: 0.9998 - val_loss: 0.0768 - val_acc: 0.9776\n",
      "\n",
      "Epoch 00046: saving model to checkpoint.hdf5\n",
      "Epoch 47/100\n",
      "20110/20110 [==============================] - 4s 189us/step - loss: 0.0034 - acc: 0.9995 - val_loss: 0.0767 - val_acc: 0.9774\n",
      "\n",
      "Epoch 00047: saving model to checkpoint.hdf5\n",
      "Epoch 48/100\n",
      "20110/20110 [==============================] - 4s 189us/step - loss: 0.0038 - acc: 0.9994 - val_loss: 0.0786 - val_acc: 0.9778\n",
      "\n",
      "Epoch 00048: saving model to checkpoint.hdf5\n",
      "Epoch 49/100\n",
      "20110/20110 [==============================] - 4s 188us/step - loss: 0.0034 - acc: 0.9995 - val_loss: 0.0788 - val_acc: 0.9780\n",
      "\n",
      "Epoch 00049: saving model to checkpoint.hdf5\n",
      "Epoch 50/100\n",
      "20110/20110 [==============================] - 4s 188us/step - loss: 0.0028 - acc: 0.9995 - val_loss: 0.0785 - val_acc: 0.9784\n",
      "\n",
      "Epoch 00050: saving model to checkpoint.hdf5\n",
      "Epoch 51/100\n",
      "20110/20110 [==============================] - 4s 180us/step - loss: 0.0028 - acc: 0.9995 - val_loss: 0.0763 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00051: saving model to checkpoint.hdf5\n",
      "Epoch 52/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0028 - acc: 0.9995 - val_loss: 0.0765 - val_acc: 0.9782\n",
      "\n",
      "Epoch 00052: saving model to checkpoint.hdf5\n",
      "Epoch 53/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0022 - acc: 0.9998 - val_loss: 0.0781 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00053: saving model to checkpoint.hdf5\n",
      "Epoch 54/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0023 - acc: 0.9998 - val_loss: 0.0771 - val_acc: 0.9790\n",
      "\n",
      "Epoch 00054: saving model to checkpoint.hdf5\n",
      "Epoch 55/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0024 - acc: 0.9996 - val_loss: 0.0776 - val_acc: 0.9784\n",
      "\n",
      "Epoch 00055: saving model to checkpoint.hdf5\n",
      "Epoch 56/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0024 - acc: 0.9997 - val_loss: 0.0774 - val_acc: 0.9790\n",
      "\n",
      "Epoch 00056: saving model to checkpoint.hdf5\n",
      "Epoch 57/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0022 - acc: 0.9999 - val_loss: 0.0786 - val_acc: 0.9784\n",
      "\n",
      "Epoch 00057: saving model to checkpoint.hdf5\n",
      "Epoch 58/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0023 - acc: 0.9998 - val_loss: 0.0784 - val_acc: 0.9786\n",
      "\n",
      "Epoch 00058: saving model to checkpoint.hdf5\n",
      "Epoch 59/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0025 - acc: 0.9993 - val_loss: 0.0788 - val_acc: 0.9784\n",
      "\n",
      "Epoch 00059: saving model to checkpoint.hdf5\n",
      "Epoch 60/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0025 - acc: 0.9995 - val_loss: 0.0788 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00060: saving model to checkpoint.hdf5\n",
      "Epoch 61/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0020 - acc: 0.9997 - val_loss: 0.0788 - val_acc: 0.9792\n",
      "\n",
      "Epoch 00061: saving model to checkpoint.hdf5\n",
      "Epoch 62/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0022 - acc: 0.9998 - val_loss: 0.0781 - val_acc: 0.9786\n",
      "\n",
      "Epoch 00062: saving model to checkpoint.hdf5\n",
      "Epoch 63/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0021 - acc: 0.9997 - val_loss: 0.0767 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00063: saving model to checkpoint.hdf5\n",
      "Epoch 64/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0021 - acc: 0.9998 - val_loss: 0.0770 - val_acc: 0.9797\n",
      "\n",
      "Epoch 00064: saving model to checkpoint.hdf5\n",
      "Epoch 65/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0019 - acc: 0.9997 - val_loss: 0.0769 - val_acc: 0.9792\n",
      "\n",
      "Epoch 00065: saving model to checkpoint.hdf5\n",
      "Epoch 66/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0017 - acc: 0.9998 - val_loss: 0.0768 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00066: saving model to checkpoint.hdf5\n",
      "Epoch 67/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0019 - acc: 0.9997 - val_loss: 0.0789 - val_acc: 0.9792\n",
      "\n",
      "Epoch 00067: saving model to checkpoint.hdf5\n",
      "Epoch 68/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0022 - acc: 0.9997 - val_loss: 0.0795 - val_acc: 0.9794\n",
      "\n",
      "Epoch 00068: saving model to checkpoint.hdf5\n",
      "Epoch 69/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0021 - acc: 0.9997 - val_loss: 0.0802 - val_acc: 0.9786\n",
      "\n",
      "Epoch 00069: saving model to checkpoint.hdf5\n",
      "Epoch 70/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0020 - acc: 0.9996 - val_loss: 0.0765 - val_acc: 0.9797\n",
      "\n",
      "Epoch 00070: saving model to checkpoint.hdf5\n",
      "Epoch 71/100\n",
      "20110/20110 [==============================] - 3s 167us/step - loss: 0.0021 - acc: 0.9997 - val_loss: 0.0763 - val_acc: 0.9803\n",
      "\n",
      "Epoch 00071: saving model to checkpoint.hdf5\n",
      "Epoch 72/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0015 - acc: 0.9999 - val_loss: 0.0755 - val_acc: 0.9795\n",
      "\n",
      "Epoch 00072: saving model to checkpoint.hdf5\n",
      "Epoch 73/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0017 - acc: 0.9997 - val_loss: 0.0755 - val_acc: 0.9790\n",
      "\n",
      "Epoch 00073: saving model to checkpoint.hdf5\n",
      "Epoch 74/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0017 - acc: 0.9997 - val_loss: 0.0766 - val_acc: 0.9795\n",
      "\n",
      "Epoch 00074: saving model to checkpoint.hdf5\n",
      "Epoch 75/100\n",
      "20110/20110 [==============================] - 3s 166us/step - loss: 0.0019 - acc: 0.9998 - val_loss: 0.0762 - val_acc: 0.9794\n",
      "\n",
      "Epoch 00075: saving model to checkpoint.hdf5\n",
      "Epoch 76/100\n",
      "20110/20110 [==============================] - 3s 160us/step - loss: 0.0015 - acc: 0.9999 - val_loss: 0.0779 - val_acc: 0.9792\n",
      "\n",
      "Epoch 00076: saving model to checkpoint.hdf5\n",
      "Epoch 77/100\n",
      "20110/20110 [==============================] - 3s 164us/step - loss: 0.0014 - acc: 0.9999 - val_loss: 0.0804 - val_acc: 0.9786\n",
      "\n",
      "Epoch 00077: saving model to checkpoint.hdf5\n",
      "Epoch 78/100\n",
      "20110/20110 [==============================] - 4s 188us/step - loss: 0.0014 - acc: 0.9999 - val_loss: 0.0794 - val_acc: 0.9794\n",
      "\n",
      "Epoch 00078: saving model to checkpoint.hdf5\n",
      "Epoch 79/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0016 - acc: 0.9998 - val_loss: 0.0817 - val_acc: 0.9784\n",
      "\n",
      "Epoch 00079: saving model to checkpoint.hdf5\n",
      "Epoch 80/100\n",
      "20110/20110 [==============================] - 4s 207us/step - loss: 0.0016 - acc: 0.9998 - val_loss: 0.0794 - val_acc: 0.9782\n",
      "\n",
      "Epoch 00080: saving model to checkpoint.hdf5\n",
      "Epoch 81/100\n",
      "20110/20110 [==============================] - 3s 169us/step - loss: 0.0013 - acc: 1.0000 - val_loss: 0.0791 - val_acc: 0.9797\n",
      "\n",
      "Epoch 00081: saving model to checkpoint.hdf5\n",
      "Epoch 82/100\n",
      "20110/20110 [==============================] - 3s 168us/step - loss: 0.0020 - acc: 0.9995 - val_loss: 0.0795 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00082: saving model to checkpoint.hdf5\n",
      "Epoch 83/100\n",
      "20110/20110 [==============================] - 4s 175us/step - loss: 0.0014 - acc: 0.9998 - val_loss: 0.0781 - val_acc: 0.9795\n",
      "\n",
      "Epoch 00083: saving model to checkpoint.hdf5\n",
      "Epoch 84/100\n",
      "20110/20110 [==============================] - 3s 166us/step - loss: 0.0017 - acc: 0.9996 - val_loss: 0.0773 - val_acc: 0.9797\n",
      "\n",
      "Epoch 00084: saving model to checkpoint.hdf5\n",
      "Epoch 85/100\n",
      "20110/20110 [==============================] - 3s 173us/step - loss: 0.0012 - acc: 0.9999 - val_loss: 0.0777 - val_acc: 0.9795\n",
      "\n",
      "Epoch 00085: saving model to checkpoint.hdf5\n",
      "Epoch 86/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0015 - acc: 0.9998 - val_loss: 0.0783 - val_acc: 0.9790\n",
      "\n",
      "Epoch 00086: saving model to checkpoint.hdf5\n",
      "Epoch 87/100\n",
      "20110/20110 [==============================] - 3s 165us/step - loss: 0.0016 - acc: 0.9996 - val_loss: 0.0803 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00087: saving model to checkpoint.hdf5\n",
      "Epoch 88/100\n",
      "20110/20110 [==============================] - 4s 174us/step - loss: 0.0015 - acc: 0.9998 - val_loss: 0.0802 - val_acc: 0.9790\n",
      "\n",
      "Epoch 00088: saving model to checkpoint.hdf5\n",
      "Epoch 89/100\n",
      "20110/20110 [==============================] - 4s 177us/step - loss: 0.0011 - acc: 0.9999 - val_loss: 0.0796 - val_acc: 0.9794\n",
      "\n",
      "Epoch 00089: saving model to checkpoint.hdf5\n",
      "Epoch 90/100\n",
      "20110/20110 [==============================] - 4s 175us/step - loss: 0.0014 - acc: 0.9998 - val_loss: 0.0782 - val_acc: 0.9792\n",
      "\n",
      "Epoch 00090: saving model to checkpoint.hdf5\n",
      "Epoch 91/100\n",
      "20110/20110 [==============================] - 3s 174us/step - loss: 0.0014 - acc: 0.9998 - val_loss: 0.0797 - val_acc: 0.9786\n",
      "\n",
      "Epoch 00091: saving model to checkpoint.hdf5\n",
      "Epoch 92/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0014 - acc: 0.9999 - val_loss: 0.0797 - val_acc: 0.9790\n",
      "\n",
      "Epoch 00092: saving model to checkpoint.hdf5\n",
      "Epoch 93/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0013 - acc: 0.9998 - val_loss: 0.0813 - val_acc: 0.9792\n",
      "\n",
      "Epoch 00093: saving model to checkpoint.hdf5\n",
      "Epoch 94/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0012 - acc: 0.9999 - val_loss: 0.0790 - val_acc: 0.9799\n",
      "\n",
      "Epoch 00094: saving model to checkpoint.hdf5\n",
      "Epoch 95/100\n",
      "20110/20110 [==============================] - 3s 173us/step - loss: 0.0011 - acc: 0.9998 - val_loss: 0.0781 - val_acc: 0.9801\n",
      "\n",
      "Epoch 00095: saving model to checkpoint.hdf5\n",
      "Epoch 96/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0012 - acc: 0.9998 - val_loss: 0.0787 - val_acc: 0.9801\n",
      "\n",
      "Epoch 00096: saving model to checkpoint.hdf5\n",
      "Epoch 97/100\n",
      "20110/20110 [==============================] - 3s 170us/step - loss: 0.0014 - acc: 0.9998 - val_loss: 0.0787 - val_acc: 0.9792\n",
      "\n",
      "Epoch 00097: ReduceLROnPlateau reducing learning rate to 0.0009000000427477062.\n",
      "\n",
      "Epoch 00097: saving model to checkpoint.hdf5\n",
      "Epoch 98/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0012 - acc: 0.9998 - val_loss: 0.0801 - val_acc: 0.9795\n",
      "\n",
      "Epoch 00098: saving model to checkpoint.hdf5\n",
      "Epoch 99/100\n",
      "20110/20110 [==============================] - 3s 172us/step - loss: 0.0010 - acc: 1.0000 - val_loss: 0.0798 - val_acc: 0.9794\n",
      "\n",
      "Epoch 00099: saving model to checkpoint.hdf5\n",
      "Epoch 100/100\n",
      "20110/20110 [==============================] - 3s 171us/step - loss: 0.0010 - acc: 0.9999 - val_loss: 0.0811 - val_acc: 0.9799\n",
      "\n",
      "Epoch 00100: saving model to checkpoint.hdf5\n"
     ]
    }
   ],
   "source": [
    "# Start to train model \n",
    "history = model.fit(X_train, y_train, \n",
    "                    batch_size=32, \n",
    "                    epochs=100, \n",
    "                    verbose=1, \n",
    "                    validation_data=(X_test, y_test),\n",
    "                    callbacks=[reduce_lr, checkpointer],\n",
    "                    shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save the model with h5py\n",
    "import h5py\n",
    "from keras.models import load_model\n",
    "model.save('./model/HSI_model_epochs100.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'val_acc', 'loss', 'acc', 'lr'])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# using plot_model module to save the model figure\n",
    "from keras.utils import plot_model\n",
    "plot_model(model, to_file='./model/model.png', show_shapes=True)\n",
    "print(history.history.keys())\n",
    "\n",
    "# show the model figure\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "model_img = plt.imread('./model/model.png')\n",
    "plt.imshow(model_img, shape=(10, 10))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# summarize history 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.grid(True)\n",
    "plt.legend(['train', 'test'], loc='upper left') \n",
    "plt.savefig(\"./result/model_accuracy_100.svg\")\n",
    "plt.show()\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.grid(True)\n",
    "plt.legend(['train', 'test'], loc='upper left') \n",
    "plt.savefig(\"./result/model_loss_100.svg\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.6 with tf",
   "language": "python",
   "name": "tf"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}