model training-split-InceptionV3.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.applications.mobilenet_v2 import MobileNetV2 \n",
    "from keras.applications.mobilenet import MobileNet\n",
    "from keras.applications.densenet import DenseNet169\n",
    "from keras.applications.resnet50 import ResNet50\n",
    "from keras.applications.nasnet import NASNetMobile, NASNetLarge\n",
    "from keras.applications.xception import Xception\n",
    "from keras.applications.inception_v3 import InceptionV3\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.preprocessing.image import img_to_array\n",
    "from sklearn.preprocessing import LabelBinarizer\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import classification_report\n",
    "from keras.models import Sequential, Model, load_model\n",
    "from keras.layers import Input, Dense, Dropout\n",
    "from keras.optimizers import Adam, SGD, Adagrad, Adadelta\n",
    "from keras import backend as K\n",
    "from imutils import paths\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import argparse\n",
    "import imutils\n",
    "import random\n",
    "import pickle\n",
    "import cv2\n",
    "import os\n",
    "import math\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# DATASET_PATH = r\"G:\\landmarks_of_melbourne_sml\"\n",
    "DATASET_PATH = r\"E:\\dataset15\"\n",
    "TESTSET_PATH = r\"E:\\testset15-s\"\n",
    "\n",
    "# initialize the number of epochs to train for, initial learning rate, batch size, and image dimensions\n",
    "EPOCHS = 15\n",
    "INIT_LR = 0.001\n",
    "BS = 8\n",
    "LABEL_NUM = 15\n",
    "READING_BS = 1000\n",
    "\n",
    "# image dimension may need to change according to the model spec\n",
    "IMAGE_DIMS =  (224, 224, 3)\n",
    "# IMAGE_DIMS =  (299, 299, 3)\n",
    "\n",
    "# initialize the data and labels\n",
    "data = []\n",
    "labels = []\n",
    "\n",
    "data_test =[]\n",
    "labels_test=[]\n",
    "\n",
    "# grab the image paths and randomly shuffle them\n",
    "imagePaths = sorted(list(paths.list_images(DATASET_PATH)))\n",
    "random.Random(881278).shuffle(imagePaths)\n",
    "\n",
    "testPaths = sorted(list(paths.list_images(TESTSET_PATH)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "test data matrix: 498.62MB\n"
     ]
    }
   ],
   "source": [
    "# loop over the test images\n",
    "for testPath in testPaths:\n",
    "    # load the image, pre-process it, and store it in the data_test list\n",
    "    assert os.path.isfile(testPath)\n",
    "    image = cv2.imread(testPath)\n",
    "    image = cv2.resize(image, (IMAGE_DIMS[1], IMAGE_DIMS[0]))\n",
    "    image = img_to_array(image)\n",
    "    data_test.append(image)\n",
    "\n",
    "    # extract the class label from the image path and update the labels_test list\n",
    "    label = testPath.split(os.path.sep)[-2]\n",
    "    labels_test.append(label)\n",
    "    \n",
    "# scale the raw pixel intensities to the range [0, 1]\n",
    "data_test = np.array(data_test, dtype=\"float\") / 255.0\n",
    "labels_test = np.array(labels_test)\n",
    "print(\"test data matrix: {:.2f}MB\".format(data_test.nbytes / (1024 * 1000.0)))\n",
    "\n",
    "# binarize the labels\n",
    "lb = LabelBinarizer()\n",
    "labels_test = lb.fit_transform(labels_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\PC-user\\AppData\\Local\\conda\\conda\\envs\\tensorflow-gpu\\lib\\site-packages\\tensorflow\\python\\framework\\op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n"
     ]
    }
   ],
   "source": [
    "# create a custom input layer that accepts images of shape IMAGE_DIMS\n",
    "input_tensor = Input(shape=IMAGE_DIMS)\n",
    "\n",
    "# grab the Keras built-in model for transfer learning\n",
    "base_model = InceptionV3(\n",
    "        include_top=False,\n",
    "        weights='imagenet',\n",
    "        input_tensor=input_tensor,\n",
    "        input_shape=IMAGE_DIMS,\n",
    "        pooling='avg')\n",
    "\n",
    "# original model is trained on different dataset with different number of classes\n",
    "# so we need to append our own output layer\n",
    "output_tensor = Dense(LABEL_NUM, activation='softmax')(base_model.output)\n",
    "\n",
    "model = Model(inputs=input_tensor, outputs=output_tensor)\n",
    "\n",
    "model.compile(Adam(lr=1e-4), loss=\"categorical_crossentropy\", metrics=['accuracy', 'top_k_categorical_accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training data matrix: 1176.00MB\n",
      "WARNING:tensorflow:From C:\\Users\\PC-user\\AppData\\Local\\conda\\conda\\envs\\tensorflow-gpu\\lib\\site-packages\\tensorflow\\python\\ops\\math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.cast instead.\n",
      "Epoch 1/15\n",
      "125/125 [==============================] - 30s 239ms/step - loss: 1.1275 - acc: 0.6850 - top_k_categorical_accuracy: 0.8910 - val_loss: 0.6383 - val_acc: 0.8113 - val_top_k_categorical_accuracy: 0.9741\n",
      "Epoch 2/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.2781 - acc: 0.9370 - top_k_categorical_accuracy: 0.9930 - val_loss: 0.3536 - val_acc: 0.8868 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 3/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.1730 - acc: 0.9580 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.3722 - val_acc: 0.8797 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 4/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.1016 - acc: 0.9730 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3337 - val_acc: 0.8986 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 5/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.1252 - acc: 0.9630 - top_k_categorical_accuracy: 0.9980 - val_loss: 0.3148 - val_acc: 0.9080 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 6/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.1190 - acc: 0.9710 - top_k_categorical_accuracy: 0.9970 - val_loss: 0.3352 - val_acc: 0.9009 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 7/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0684 - acc: 0.9840 - top_k_categorical_accuracy: 0.9980 - val_loss: 0.3315 - val_acc: 0.9080 - val_top_k_categorical_accuracy: 0.9953\n",
      "Epoch 8/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0607 - acc: 0.9860 - top_k_categorical_accuracy: 0.9970 - val_loss: 0.4401 - val_acc: 0.8915 - val_top_k_categorical_accuracy: 0.9741\n",
      "Epoch 9/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0777 - acc: 0.9780 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.5005 - val_acc: 0.8726 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 10/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0842 - acc: 0.9720 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.7288 - val_acc: 0.8042 - val_top_k_categorical_accuracy: 0.9693\n",
      "Epoch 11/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0635 - acc: 0.9820 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.6122 - val_acc: 0.8420 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 12/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0507 - acc: 0.9850 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.3938 - val_acc: 0.8892 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 13/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0392 - acc: 0.9900 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3817 - val_acc: 0.8892 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 14/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0484 - acc: 0.9870 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3042 - val_acc: 0.9198 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 15/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0415 - acc: 0.9910 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.5989 - val_acc: 0.8561 - val_top_k_categorical_accuracy: 0.9858\n",
      "training data matrix: 1176.00MB\n",
      "Epoch 1/15\n",
      "125/125 [==============================] - 15s 121ms/step - loss: 0.3924 - acc: 0.8950 - top_k_categorical_accuracy: 0.9730 - val_loss: 0.3297 - val_acc: 0.9009 - val_top_k_categorical_accuracy: 0.9953\n",
      "Epoch 2/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.1709 - acc: 0.9510 - top_k_categorical_accuracy: 0.9930 - val_loss: 0.4670 - val_acc: 0.8679 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 3/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.1164 - acc: 0.9650 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.3362 - val_acc: 0.9151 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 4/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0664 - acc: 0.9780 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2590 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9953\n",
      "Epoch 5/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.1018 - acc: 0.9680 - top_k_categorical_accuracy: 0.9970 - val_loss: 0.2101 - val_acc: 0.9410 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 6/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0441 - acc: 0.9910 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.1775 - val_acc: 0.9481 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 7/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0399 - acc: 0.9920 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.1991 - val_acc: 0.9363 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 8/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0596 - acc: 0.9810 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.1959 - val_acc: 0.9387 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 9/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0402 - acc: 0.9850 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2399 - val_acc: 0.9316 - val_top_k_categorical_accuracy: 0.9976\n",
      "Epoch 10/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0307 - acc: 0.9930 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2938 - val_acc: 0.9222 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 11/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0288 - acc: 0.9880 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2026 - val_acc: 0.9458 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 12/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0300 - acc: 0.9900 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.1857 - val_acc: 0.9481 - val_top_k_categorical_accuracy: 0.9953\n",
      "Epoch 13/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0261 - acc: 0.9930 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.1875 - val_acc: 0.9316 - val_top_k_categorical_accuracy: 0.9976\n",
      "Epoch 14/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0164 - acc: 0.9960 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.1871 - val_acc: 0.9410 - val_top_k_categorical_accuracy: 0.9976\n",
      "Epoch 15/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0204 - acc: 0.9940 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2103 - val_acc: 0.9410 - val_top_k_categorical_accuracy: 0.9976\n",
      "training data matrix: 1176.00MB\n",
      "Epoch 1/15\n",
      "125/125 [==============================] - 15s 121ms/step - loss: 0.2490 - acc: 0.9250 - top_k_categorical_accuracy: 0.9890 - val_loss: 0.5126 - val_acc: 0.8750 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 2/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.1451 - acc: 0.9550 - top_k_categorical_accuracy: 0.9960 - val_loss: 0.3106 - val_acc: 0.8986 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 3/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0800 - acc: 0.9780 - top_k_categorical_accuracy: 0.9980 - val_loss: 0.2944 - val_acc: 0.9080 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 4/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0481 - acc: 0.9890 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2984 - val_acc: 0.9175 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 5/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0298 - acc: 0.9940 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.2396 - val_acc: 0.9387 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 6/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0357 - acc: 0.9870 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2908 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 7/15\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0506 - acc: 0.9820 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2689 - val_acc: 0.9316 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 8/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0441 - acc: 0.9870 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2645 - val_acc: 0.9434 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 9/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0278 - acc: 0.9920 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2269 - val_acc: 0.9505 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 10/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0564 - acc: 0.9850 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.4258 - val_acc: 0.9151 - val_top_k_categorical_accuracy: 0.9717\n",
      "Epoch 11/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0517 - acc: 0.9880 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.4284 - val_acc: 0.8986 - val_top_k_categorical_accuracy: 0.9788\n",
      "Epoch 12/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0454 - acc: 0.9890 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2777 - val_acc: 0.9269 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 13/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0250 - acc: 0.9960 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2468 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 14/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0304 - acc: 0.9900 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.4125 - val_acc: 0.9080 - val_top_k_categorical_accuracy: 0.9741\n",
      "Epoch 15/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0413 - acc: 0.9880 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2666 - val_acc: 0.9292 - val_top_k_categorical_accuracy: 0.9835\n",
      "training data matrix: 1176.00MB\n",
      "Epoch 1/15\n",
      "125/125 [==============================] - 15s 121ms/step - loss: 0.2583 - acc: 0.9250 - top_k_categorical_accuracy: 0.9910 - val_loss: 0.3555 - val_acc: 0.9057 - val_top_k_categorical_accuracy: 0.9835\n",
      "Epoch 2/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0986 - acc: 0.9710 - top_k_categorical_accuracy: 0.9980 - val_loss: 0.2313 - val_acc: 0.9269 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 3/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0677 - acc: 0.9800 - top_k_categorical_accuracy: 0.9970 - val_loss: 0.2426 - val_acc: 0.9363 - val_top_k_categorical_accuracy: 0.9835\n",
      "Epoch 4/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0434 - acc: 0.9880 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.2806 - val_acc: 0.9198 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 5/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0571 - acc: 0.9840 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.3283 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 6/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0387 - acc: 0.9860 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2786 - val_acc: 0.9387 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 7/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0340 - acc: 0.9870 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3098 - val_acc: 0.9245 - val_top_k_categorical_accuracy: 0.9811\n",
      "Epoch 8/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0423 - acc: 0.9850 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3841 - val_acc: 0.9033 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 9/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0402 - acc: 0.9900 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3245 - val_acc: 0.9127 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 10/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0170 - acc: 0.9940 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2528 - val_acc: 0.9316 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 11/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0308 - acc: 0.9910 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2745 - val_acc: 0.9316 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 12/15\n",
      "125/125 [==============================] - 15s 120ms/step - loss: 0.0181 - acc: 0.9970 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2479 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 13/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0170 - acc: 0.9960 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2331 - val_acc: 0.9458 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 14/15\n",
      "125/125 [==============================] - 15s 117ms/step - loss: 0.0257 - acc: 0.9950 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3230 - val_acc: 0.9080 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 15/15\n",
      "125/125 [==============================] - 15s 117ms/step - loss: 0.0276 - acc: 0.9900 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.1918 - val_acc: 0.9552 - val_top_k_categorical_accuracy: 0.9906\n",
      "training data matrix: 1176.00MB\n",
      "Epoch 1/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.1967 - acc: 0.9430 - top_k_categorical_accuracy: 0.9910 - val_loss: 0.3101 - val_acc: 0.9222 - val_top_k_categorical_accuracy: 0.9953\n",
      "Epoch 2/15\n",
      "125/125 [==============================] - 15s 117ms/step - loss: 0.1147 - acc: 0.9710 - top_k_categorical_accuracy: 0.9980 - val_loss: 0.4054 - val_acc: 0.9104 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 3/15\n",
      "125/125 [==============================] - 15s 119ms/step - loss: 0.0811 - acc: 0.9790 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.2323 - val_acc: 0.9387 - val_top_k_categorical_accuracy: 0.9953\n",
      "Epoch 4/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0563 - acc: 0.9880 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.3928 - val_acc: 0.9222 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 5/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0244 - acc: 0.9930 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3065 - val_acc: 0.9316 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 6/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0131 - acc: 0.9980 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3007 - val_acc: 0.9198 - val_top_k_categorical_accuracy: 0.9953\n",
      "Epoch 7/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0145 - acc: 0.9950 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.4164 - val_acc: 0.9080 - val_top_k_categorical_accuracy: 0.9976\n",
      "Epoch 8/15\n",
      "125/125 [==============================] - 15s 117ms/step - loss: 0.0440 - acc: 0.9900 - top_k_categorical_accuracy: 0.9990 - val_loss: 0.1884 - val_acc: 0.9481 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 9/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0192 - acc: 0.9960 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2303 - val_acc: 0.9410 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 10/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0216 - acc: 0.9960 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2385 - val_acc: 0.9434 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 11/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0281 - acc: 0.9880 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2556 - val_acc: 0.9575 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 12/15\n",
      "125/125 [==============================] - 15s 117ms/step - loss: 0.0159 - acc: 0.9970 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2658 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9835\n",
      "Epoch 13/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0180 - acc: 0.9950 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.4160 - val_acc: 0.9245 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 14/15\n",
      "125/125 [==============================] - 15s 118ms/step - loss: 0.0263 - acc: 0.9920 - top_k_categorical_accuracy: 0.9980 - val_loss: 0.3859 - val_acc: 0.9104 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 15/15\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "125/125 [==============================] - 15s 117ms/step - loss: 0.0141 - acc: 0.9950 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.4590 - val_acc: 0.9269 - val_top_k_categorical_accuracy: 0.9882\n",
      "training data matrix: 880.82MB\n",
      "Epoch 1/15\n",
      "94/94 [==============================] - 13s 143ms/step - loss: 0.2928 - acc: 0.9261 - top_k_categorical_accuracy: 0.9840 - val_loss: 1.4707 - val_acc: 0.7571 - val_top_k_categorical_accuracy: 0.9764\n",
      "Epoch 2/15\n",
      "94/94 [==============================] - 12s 126ms/step - loss: 0.1142 - acc: 0.9713 - top_k_categorical_accuracy: 0.9973 - val_loss: 0.5710 - val_acc: 0.8774 - val_top_k_categorical_accuracy: 0.9811\n",
      "Epoch 3/15\n",
      "94/94 [==============================] - 12s 126ms/step - loss: 0.0518 - acc: 0.9840 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3043 - val_acc: 0.9316 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 4/15\n",
      "94/94 [==============================] - 12s 132ms/step - loss: 0.0446 - acc: 0.9880 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2408 - val_acc: 0.9481 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 5/15\n",
      "94/94 [==============================] - 12s 129ms/step - loss: 0.0227 - acc: 0.9920 - top_k_categorical_accuracy: 0.9987 - val_loss: 0.3034 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 6/15\n",
      "94/94 [==============================] - 12s 130ms/step - loss: 0.0387 - acc: 0.9867 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2719 - val_acc: 0.9434 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 7/15\n",
      "94/94 [==============================] - 12s 129ms/step - loss: 0.0314 - acc: 0.9934 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3299 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 8/15\n",
      "94/94 [==============================] - 12s 130ms/step - loss: 0.0163 - acc: 0.9973 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2995 - val_acc: 0.9410 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 9/15\n",
      "94/94 [==============================] - 12s 128ms/step - loss: 0.0258 - acc: 0.9934 - top_k_categorical_accuracy: 0.9987 - val_loss: 0.3117 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 10/15\n",
      "94/94 [==============================] - 12s 128ms/step - loss: 0.0227 - acc: 0.9934 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3636 - val_acc: 0.9151 - val_top_k_categorical_accuracy: 0.9906\n",
      "Epoch 11/15\n",
      "94/94 [==============================] - 12s 128ms/step - loss: 0.0157 - acc: 0.9960 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3127 - val_acc: 0.9222 - val_top_k_categorical_accuracy: 0.9882\n",
      "Epoch 12/15\n",
      "94/94 [==============================] - 12s 129ms/step - loss: 0.0264 - acc: 0.9926 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2606 - val_acc: 0.9410 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 13/15\n",
      "94/94 [==============================] - 12s 129ms/step - loss: 0.0111 - acc: 0.9973 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2748 - val_acc: 0.9340 - val_top_k_categorical_accuracy: 0.9929\n",
      "Epoch 14/15\n",
      "94/94 [==============================] - 12s 131ms/step - loss: 0.0180 - acc: 0.9947 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.3376 - val_acc: 0.9316 - val_top_k_categorical_accuracy: 0.9858\n",
      "Epoch 15/15\n",
      "94/94 [==============================] - 12s 130ms/step - loss: 0.0146 - acc: 0.9973 - top_k_categorical_accuracy: 1.0000 - val_loss: 0.2413 - val_acc: 0.9387 - val_top_k_categorical_accuracy: 0.9906\n"
     ]
    }
   ],
   "source": [
    "chunk_paths = [imagePaths[i:i + READING_BS] for i in range(0, len(imagePaths), READING_BS)]\n",
    "for chunk_path in chunk_paths:\n",
    "    # loop over the each chuck\n",
    "    data = []\n",
    "    labels = []\n",
    "    for imagePath in chunk_path:\n",
    "        # load the image, pre-process it, and store it in the data list\n",
    "        assert os.path.isfile(imagePath)\n",
    "        image = cv2.imread(imagePath)\n",
    "        image = cv2.resize(image, (IMAGE_DIMS[1], IMAGE_DIMS[0]))\n",
    "        image = img_to_array(image)\n",
    "        data.append(image)\n",
    "\n",
    "        # extract the class label from the image path and update the labels list\n",
    "        label = imagePath.split(os.path.sep)[-2]\n",
    "        labels.append(label)\n",
    "    \n",
    "    # construct the image generator for data augmentation\n",
    "    # aug = ImageDataGenerator(rotation_range=25, width_shift_range=0.1, height_shift_range=0.1, \n",
    "    #                          shear_range=0.2, zoom_range=0.2, horizontal_flip=True)\n",
    "    aug = ImageDataGenerator(width_shift_range=0.1, height_shift_range=0.1)\n",
    "    \n",
    "    # scale the raw pixel intensities to the range [0, 1]\n",
    "    data = np.array(data, dtype=\"float\") / 255.0\n",
    "    labels = np.array(labels)\n",
    "    print(\"training data matrix: {:.2f}MB\".format(data.nbytes / (1024 * 1000.0)))\n",
    "\n",
    "    # binarize the labels\n",
    "    lb = LabelBinarizer()\n",
    "    labels = lb.fit_transform(labels)\n",
    "    \n",
    "    h = model.fit_generator(\n",
    "        aug.flow(data, labels, batch_size=BS),\n",
    "        validation_data=(data_test, labels_test),\n",
    "        steps_per_epoch=math.ceil(len(data)/BS),\n",
    "        epochs=EPOCHS, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save the model to disk\n",
    "model.save(r\"E:\\models\\224\\InceptionV3\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save the label binarizer to disk\n",
    "f = open(r\"E:\\models\\lb-dataset15-i\", \"wb\")\n",
    "f.write(pickle.dumps(lb))\n",
    "f.close()"
   ]
  }
 ],
 "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": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}