main.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.metrics import confusion_matrix, accuracy_score\n",
    "\n",
    "import tensorflow as tf\n",
    "import tensorflow.keras.layers as tfl\n",
    "from tensorflow.python.keras.preprocessing.image import ImageDataGenerator\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.python.keras.callbacks import ModelCheckpoint, EarlyStopping\n",
    "\n",
    "physical_devices = tf.config.list_physical_devices(\"GPU\")\n",
    "tf.config.experimental.set_memory_growth(physical_devices[0], True)\n",
    "\n",
    "BATCH_SIZE = 32\n",
    "IMG_SIZE = (256, 256) # (you choose)\n",
    "MODEL_N = 0 # (it is used just to save the models in different files.h5)\n",
    "SEED = 42\n",
    "\n",
    "# Folders containing sub-folders named after the classes.\n",
    "# Images are stored in the sub-folder corresponding to their true class.\n",
    "training = \"training/\"\n",
    "test = \"test/\"\n",
    "category_names = os.listdir(training)\n",
    "nb_categories = len(category_names)\n",
    "validation_split = 0.2 # (you decide)\n",
    "\n",
    "# set the weights for each class in the loss function, according to the representation\n",
    "# of each class in the training folder.\n",
    "class_weight = {0: 1, 1: 1} # (you decide)\n",
    "\n",
    "# define the model to train\n",
    "def your_model(img_size=(256, 256), categories=2, k=0):\n",
    "    '''\n",
    "    - \"categories\"\n",
    "        (integer) defining the number of categories for the softmax layer\n",
    "    - \"k\"\n",
    "        (integer) defining the pre-trained model:\n",
    "            k == 0:     Xception\n",
    "            k == 1:     ResNet50V2\n",
    "            k == 2:     ResNet101V2\n",
    "            k == 3:     ResNet152V2\n",
    "            k == 4:     InceptionV3\n",
    "            k == 5:     InceptionResNetV2\n",
    "            k == 6:     EfficientNetB4\n",
    "            k == 7:     EfficientNetB5\n",
    "            k == 8:     EfficientNetB6\n",
    "            k == 9:     EfficientNetB7\n",
    "    '''\n",
    "    input_shape = img_size + (3,)\n",
    "    if k == 0:\n",
    "        base_model = tf.keras.applications.Xception(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 1:\n",
    "        base_model = tf.keras.applications.ResNet50V2(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 2:\n",
    "        base_model = tf.keras.applications.ResNet101V2(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 3:\n",
    "        base_model = tf.keras.applications.ResNet152V2(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 4:\n",
    "        base_model = tf.keras.applications.InceptionV3(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 5:\n",
    "        base_model = tf.keras.applications.InceptionResNetV2(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 6:\n",
    "        base_model = tf.keras.applications.EfficientNetB4(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 7:\n",
    "        base_model = tf.keras.applications.EfficientNetB5(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 8:\n",
    "        base_model = tf.keras.applications.EfficientNetB6(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    elif k == 9:\n",
    "        base_model = tf.keras.applications.EfficientNetB7(input_shape=input_shape,\n",
    "                                                    include_top=False, \n",
    "                                                    weights='imagenet') # From imageNet\n",
    "    \n",
    "    # freeze all layers.\n",
    "    base_model.trainable = False\n",
    "    \n",
    "    # define the input tensor.\n",
    "    inputs = tf.keras.Input(shape=input_shape)\n",
    "    \n",
    "    # preprocess the input images according to the pre-trained model used. \n",
    "    # \"preprocess_input\" is a keras function called during the execution of the main.\n",
    "    x = preprocess_input(inputs)\n",
    "    \n",
    "    # pass the preprocessed inputs to the selected pretrained model.\n",
    "    x = base_model(x, training=False)\n",
    "    \n",
    "    # apply \"Dropout\" on the activation maps of the last convolutional layer, \n",
    "    # before passing them to the \"GAP\".\n",
    "    x = tfl.Dropout(rate=0.2, name='dropout_1')(x) # (I don't think you need this)\n",
    "    x = tfl.GlobalAveragePooling2D(name='gap')(x)\n",
    "    \n",
    "    # apply \"Dropout\" on the output vector of the \"GAP\", \n",
    "    # before passing it to the \"Softmax\" layer.\n",
    "    x = tfl.Dropout(rate=0.4, name='dropout_2')(x)\n",
    "    \n",
    "    # the parameters of the \"Softmax\" layer are also regularized using the L2-norm.\n",
    "    outputs = tfl.Dense(categories, activation='softmax', \n",
    "                        bias_regularizer=tf.keras.regularizers.l2(1e-4), \n",
    "                        activity_regularizer=tf.keras.regularizers.l2(1e-5), \n",
    "                        name='softmax')(x)\n",
    "    \n",
    "    # the model is created, passing as arguments, the input and the output of the\n",
    "    # computational graph.\n",
    "    model = tf.keras.Model(inputs, outputs)\n",
    "    return model\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 32589 images belonging to 14 classes.\n",
      "Found 21717 images belonging to 14 classes.\n",
      "Epoch 1/7\n",
      "1019/1019 [==============================] - 663s 646ms/step - loss: 2.8873 - accuracy: 0.8260 - val_loss: 0.4964 - val_accuracy: 0.8487\n",
      "Epoch 2/7\n",
      "1019/1019 [==============================] - 661s 648ms/step - loss: 1.4477 - accuracy: 0.8987 - val_loss: 0.4315 - val_accuracy: 0.8666\n",
      "Epoch 3/7\n",
      "1019/1019 [==============================] - 662s 649ms/step - loss: 1.1599 - accuracy: 0.9155 - val_loss: 0.3896 - val_accuracy: 0.8771\n",
      "Epoch 4/7\n",
      "1019/1019 [==============================] - 512s 502ms/step - loss: 1.0250 - accuracy: 0.9224 - val_loss: 0.3772 - val_accuracy: 0.8816\n",
      "Epoch 5/7\n",
      "1019/1019 [==============================] - 475s 466ms/step - loss: 0.9576 - accuracy: 0.9274 - val_loss: 0.4253 - val_accuracy: 0.8664\n",
      "Epoch 6/7\n",
      "1019/1019 [==============================] - 441s 433ms/step - loss: 0.8794 - accuracy: 0.9310 - val_loss: 0.4067 - val_accuracy: 0.8740\n",
      "Epoch 7/7\n",
      "1019/1019 [==============================] - 385s 378ms/step - loss: 0.8791 - accuracy: 0.9312 - val_loss: 0.4679 - val_accuracy: 0.8573\n",
      "Epoch 8/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.4778 - accuracy: 0.9648\n",
      "Epoch 00008: val_loss improved from inf to 0.43327, saving model to model60_fineTuned.h5\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\fuge\\anaconda3\\envs\\active2\\lib\\site-packages\\keras\\engine\\functional.py:1410: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.\n",
      "  layer_config = serialize_layer_fn(layer)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1019/1019 [==============================] - 386s 374ms/step - loss: 0.4778 - accuracy: 0.9648 - val_loss: 0.4333 - val_accuracy: 0.8926\n",
      "Epoch 9/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.2069 - accuracy: 0.9844\n",
      "Epoch 00009: val_loss improved from 0.43327 to 0.32775, saving model to model60_fineTuned.h5\n",
      "1019/1019 [==============================] - 384s 377ms/step - loss: 0.2069 - accuracy: 0.9844 - val_loss: 0.3277 - val_accuracy: 0.9207\n",
      "Epoch 10/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.1232 - accuracy: 0.9906\n",
      "Epoch 00010: val_loss improved from 0.32775 to 0.31428, saving model to model60_fineTuned.h5\n",
      "1019/1019 [==============================] - 393s 385ms/step - loss: 0.1232 - accuracy: 0.9906 - val_loss: 0.3143 - val_accuracy: 0.9324\n",
      "Epoch 11/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0902 - accuracy: 0.9934\n",
      "Epoch 00011: val_loss improved from 0.31428 to 0.24467, saving model to model60_fineTuned.h5\n",
      "1019/1019 [==============================] - 394s 386ms/step - loss: 0.0902 - accuracy: 0.9934 - val_loss: 0.2447 - val_accuracy: 0.9437\n",
      "Epoch 12/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0640 - accuracy: 0.9952\n",
      "Epoch 00012: val_loss improved from 0.24467 to 0.22440, saving model to model60_fineTuned.h5\n",
      "1019/1019 [==============================] - 384s 377ms/step - loss: 0.0640 - accuracy: 0.9952 - val_loss: 0.2244 - val_accuracy: 0.9509\n",
      "Epoch 13/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0540 - accuracy: 0.9953\n",
      "Epoch 00013: val_loss did not improve from 0.22440\n",
      "1019/1019 [==============================] - 383s 376ms/step - loss: 0.0540 - accuracy: 0.9953 - val_loss: 0.2400 - val_accuracy: 0.9391\n",
      "Epoch 14/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0453 - accuracy: 0.9961\n",
      "Epoch 00014: val_loss did not improve from 0.22440\n",
      "1019/1019 [==============================] - 376s 369ms/step - loss: 0.0453 - accuracy: 0.9961 - val_loss: 0.2483 - val_accuracy: 0.9397\n",
      "Epoch 15/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0333 - accuracy: 0.9971\n",
      "Epoch 00015: val_loss did not improve from 0.22440\n",
      "1019/1019 [==============================] - 375s 368ms/step - loss: 0.0333 - accuracy: 0.9971 - val_loss: 0.3373 - val_accuracy: 0.9230\n",
      "Epoch 16/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0341 - accuracy: 0.9972\n",
      "Epoch 00016: val_loss did not improve from 0.22440\n",
      "1019/1019 [==============================] - 378s 371ms/step - loss: 0.0341 - accuracy: 0.9972 - val_loss: 0.3478 - val_accuracy: 0.9254\n",
      "Epoch 17/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0434 - accuracy: 0.9970\n",
      "Epoch 00017: val_loss did not improve from 0.22440\n",
      "1019/1019 [==============================] - 375s 368ms/step - loss: 0.0434 - accuracy: 0.9970 - val_loss: 0.2361 - val_accuracy: 0.9455\n",
      "Epoch 18/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0324 - accuracy: 0.9973\n",
      "Epoch 00018: val_loss improved from 0.22440 to 0.20796, saving model to model60_fineTuned.h5\n",
      "1019/1019 [==============================] - 378s 371ms/step - loss: 0.0324 - accuracy: 0.9973 - val_loss: 0.2080 - val_accuracy: 0.9542\n",
      "Epoch 19/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0250 - accuracy: 0.9978\n",
      "Epoch 00019: val_loss did not improve from 0.20796\n",
      "1019/1019 [==============================] - 373s 366ms/step - loss: 0.0250 - accuracy: 0.9978 - val_loss: 0.3726 - val_accuracy: 0.9314\n",
      "Epoch 20/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0187 - accuracy: 0.9978\n",
      "Epoch 00020: val_loss did not improve from 0.20796\n",
      "1019/1019 [==============================] - 377s 370ms/step - loss: 0.0187 - accuracy: 0.9978 - val_loss: 0.2126 - val_accuracy: 0.9497\n",
      "Epoch 21/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0162 - accuracy: 0.9989\n",
      "Epoch 00021: val_loss did not improve from 0.20796\n",
      "1019/1019 [==============================] - 371s 364ms/step - loss: 0.0162 - accuracy: 0.9989 - val_loss: 0.3095 - val_accuracy: 0.9383\n",
      "Epoch 22/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0217 - accuracy: 0.9979\n",
      "Epoch 00022: val_loss improved from 0.20796 to 0.20094, saving model to model60_fineTuned.h5\n",
      "1019/1019 [==============================] - 376s 369ms/step - loss: 0.0217 - accuracy: 0.9979 - val_loss: 0.2009 - val_accuracy: 0.9548\n",
      "Epoch 23/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0164 - accuracy: 0.9987\n",
      "Epoch 00023: val_loss did not improve from 0.20094\n",
      "1019/1019 [==============================] - 374s 367ms/step - loss: 0.0164 - accuracy: 0.9987 - val_loss: 0.2071 - val_accuracy: 0.9525\n",
      "Epoch 24/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0090 - accuracy: 0.9991\n",
      "Epoch 00024: val_loss did not improve from 0.20094\n",
      "1019/1019 [==============================] - 373s 366ms/step - loss: 0.0090 - accuracy: 0.9991 - val_loss: 0.2277 - val_accuracy: 0.9556\n",
      "Epoch 25/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0137 - accuracy: 0.9988\n",
      "Epoch 00025: val_loss improved from 0.20094 to 0.19168, saving model to model60_fineTuned.h5\n",
      "1019/1019 [==============================] - 377s 369ms/step - loss: 0.0137 - accuracy: 0.9988 - val_loss: 0.1917 - val_accuracy: 0.9573\n",
      "Epoch 26/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0106 - accuracy: 0.9993\n",
      "Epoch 00026: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 378s 371ms/step - loss: 0.0106 - accuracy: 0.9993 - val_loss: 0.2317 - val_accuracy: 0.9534\n",
      "Epoch 27/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0059 - accuracy: 0.9992\n",
      "Epoch 00027: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 374s 367ms/step - loss: 0.0059 - accuracy: 0.9992 - val_loss: 0.4250 - val_accuracy: 0.9347\n",
      "Epoch 28/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0079 - accuracy: 0.9993\n",
      "Epoch 00028: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 378s 371ms/step - loss: 0.0079 - accuracy: 0.9993 - val_loss: 0.3801 - val_accuracy: 0.9298\n",
      "Epoch 29/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0145 - accuracy: 0.9987\n",
      "Epoch 00029: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 374s 367ms/step - loss: 0.0145 - accuracy: 0.9987 - val_loss: 0.3226 - val_accuracy: 0.9371\n",
      "Epoch 30/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0153 - accuracy: 0.9986\n",
      "Epoch 00030: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 380s 373ms/step - loss: 0.0153 - accuracy: 0.9986 - val_loss: 0.2712 - val_accuracy: 0.9406\n",
      "Epoch 31/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0088 - accuracy: 0.9990\n",
      "Epoch 00031: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 377s 369ms/step - loss: 0.0088 - accuracy: 0.9990 - val_loss: 0.2655 - val_accuracy: 0.9508\n",
      "Epoch 32/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0099 - accuracy: 0.9991\n",
      "Epoch 00032: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 376s 369ms/step - loss: 0.0099 - accuracy: 0.9991 - val_loss: 0.2334 - val_accuracy: 0.9548\n",
      "Epoch 33/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0130 - accuracy: 0.9989\n",
      "Epoch 00033: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 375s 368ms/step - loss: 0.0130 - accuracy: 0.9989 - val_loss: 0.3018 - val_accuracy: 0.9421\n",
      "Epoch 34/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0195 - accuracy: 0.9986\n",
      "Epoch 00034: val_loss did not improve from 0.19168\n",
      "1019/1019 [==============================] - 372s 365ms/step - loss: 0.0195 - accuracy: 0.9986 - val_loss: 0.2959 - val_accuracy: 0.9355\n",
      "Epoch 35/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0085 - accuracy: 0.9992\n",
      "Epoch 00035: val_loss did not improve from 0.19168\n",
      "Restoring model weights from the end of the best epoch.\n",
      "1019/1019 [==============================] - 375s 367ms/step - loss: 0.0085 - accuracy: 0.9992 - val_loss: 0.2654 - val_accuracy: 0.9460\n",
      "Epoch 00035: early stopping\n",
      "Found 32589 images belonging to 14 classes.\n",
      "Found 21717 images belonging to 14 classes.\n",
      "Epoch 1/7\n",
      "1019/1019 [==============================] - 354s 344ms/step - loss: 2.0333 - accuracy: 0.8627 - val_loss: 0.3870 - val_accuracy: 0.8822\n",
      "Epoch 2/7\n",
      "1019/1019 [==============================] - 356s 349ms/step - loss: 0.8399 - accuracy: 0.9392 - val_loss: 0.3507 - val_accuracy: 0.8946\n",
      "Epoch 3/7\n",
      "1019/1019 [==============================] - 355s 348ms/step - loss: 0.6960 - accuracy: 0.9478 - val_loss: 0.3480 - val_accuracy: 0.8975\n",
      "Epoch 4/7\n",
      "1019/1019 [==============================] - 350s 344ms/step - loss: 0.6035 - accuracy: 0.9523 - val_loss: 0.4199 - val_accuracy: 0.8818\n",
      "Epoch 5/7\n",
      "1019/1019 [==============================] - 352s 345ms/step - loss: 0.5666 - accuracy: 0.9550 - val_loss: 0.3171 - val_accuracy: 0.9064\n",
      "Epoch 6/7\n",
      "1019/1019 [==============================] - 352s 345ms/step - loss: 0.5422 - accuracy: 0.9558 - val_loss: 0.3413 - val_accuracy: 0.9036\n",
      "Epoch 7/7\n",
      "1019/1019 [==============================] - 350s 343ms/step - loss: 0.5275 - accuracy: 0.9558 - val_loss: 0.3615 - val_accuracy: 0.9017\n",
      "Epoch 8/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.4688 - accuracy: 0.9618\n",
      "Epoch 00008: val_loss improved from inf to 0.37363, saving model to model61_fineTuned.h5\n",
      "1019/1019 [==============================] - 355s 346ms/step - loss: 0.4688 - accuracy: 0.9618 - val_loss: 0.3736 - val_accuracy: 0.8769\n",
      "Epoch 9/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.1993 - accuracy: 0.9826\n",
      "Epoch 00009: val_loss improved from 0.37363 to 0.31688, saving model to model61_fineTuned.h5\n",
      "1019/1019 [==============================] - 355s 348ms/step - loss: 0.1993 - accuracy: 0.9826 - val_loss: 0.3169 - val_accuracy: 0.9292\n",
      "Epoch 10/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.1322 - accuracy: 0.9890\n",
      "Epoch 00010: val_loss improved from 0.31688 to 0.21391, saving model to model61_fineTuned.h5\n",
      "1019/1019 [==============================] - 353s 346ms/step - loss: 0.1322 - accuracy: 0.9890 - val_loss: 0.2139 - val_accuracy: 0.9431\n",
      "Epoch 11/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0953 - accuracy: 0.9915\n",
      "Epoch 00011: val_loss did not improve from 0.21391\n",
      "1019/1019 [==============================] - 355s 349ms/step - loss: 0.0953 - accuracy: 0.9915 - val_loss: 0.2375 - val_accuracy: 0.9341\n",
      "Epoch 12/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0832 - accuracy: 0.9936\n",
      "Epoch 00012: val_loss improved from 0.21391 to 0.13798, saving model to model61_fineTuned.h5\n",
      "1019/1019 [==============================] - 354s 347ms/step - loss: 0.0832 - accuracy: 0.9936 - val_loss: 0.1380 - val_accuracy: 0.9650\n",
      "Epoch 13/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0684 - accuracy: 0.9941\n",
      "Epoch 00013: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 363s 356ms/step - loss: 0.0684 - accuracy: 0.9941 - val_loss: 0.1971 - val_accuracy: 0.9476\n",
      "Epoch 14/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0647 - accuracy: 0.9937\n",
      "Epoch 00014: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 365s 358ms/step - loss: 0.0647 - accuracy: 0.9937 - val_loss: 0.6863 - val_accuracy: 0.8931\n",
      "Epoch 15/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0692 - accuracy: 0.9944\n",
      "Epoch 00015: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 363s 356ms/step - loss: 0.0692 - accuracy: 0.9944 - val_loss: 0.2810 - val_accuracy: 0.9382\n",
      "Epoch 16/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0377 - accuracy: 0.9965\n",
      "Epoch 00016: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 361s 354ms/step - loss: 0.0377 - accuracy: 0.9965 - val_loss: 0.2497 - val_accuracy: 0.9473\n",
      "Epoch 17/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0403 - accuracy: 0.9964\n",
      "Epoch 00017: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 358s 352ms/step - loss: 0.0403 - accuracy: 0.9964 - val_loss: 0.2243 - val_accuracy: 0.9458\n",
      "Epoch 18/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0333 - accuracy: 0.9975\n",
      "Epoch 00018: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 437s 429ms/step - loss: 0.0333 - accuracy: 0.9975 - val_loss: 0.2648 - val_accuracy: 0.9411\n",
      "Epoch 19/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0202 - accuracy: 0.9981\n",
      "Epoch 00019: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 710s 697ms/step - loss: 0.0202 - accuracy: 0.9981 - val_loss: 0.2306 - val_accuracy: 0.9555\n",
      "Epoch 20/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0234 - accuracy: 0.9980\n",
      "Epoch 00020: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 656s 643ms/step - loss: 0.0234 - accuracy: 0.9980 - val_loss: 0.2188 - val_accuracy: 0.9562\n",
      "Epoch 21/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0327 - accuracy: 0.9971\n",
      "Epoch 00021: val_loss did not improve from 0.13798\n",
      "1019/1019 [==============================] - 599s 588ms/step - loss: 0.0327 - accuracy: 0.9971 - val_loss: 0.2767 - val_accuracy: 0.9519\n",
      "Epoch 22/60\n",
      "1019/1019 [==============================] - ETA: 0s - loss: 0.0228 - accuracy: 0.9976\n",
      "Epoch 00022: val_loss did not improve from 0.13798\n",
      "Restoring model weights from the end of the best epoch.\n",
      "1019/1019 [==============================] - 438s 429ms/step - loss: 0.0228 - accuracy: 0.9976 - val_loss: 0.2305 - val_accuracy: 0.9549\n",
      "Epoch 00022: early stopping\n"
     ]
    }
   ],
   "source": [
    "# For Loop ·······················································································\n",
    "accs = []\n",
    "val_accs = []\n",
    "losses = []\n",
    "val_losses = []\n",
    "\n",
    "# this for loop compute a fitted model for each of the 10 pre-trained models in use.\n",
    "used_models = [0, 1]\n",
    "for kkk in used_models:\n",
    "    tf.keras.backend.clear_session()\n",
    "    \n",
    "    # prepare the training set.\n",
    "    train_datagen = ImageDataGenerator(rotation_range=40, \n",
    "                                        width_shift_range=0.2, height_shift_range=0.2, \n",
    "                                        brightness_range=[0.7, 1.3], \n",
    "                                        zoom_range=0.2, \n",
    "                                        fill_mode='constant', cval=0, \n",
    "                                        horizontal_flip=True, vertical_flip=True, \n",
    "                                        validation_split=validation_split)\n",
    "    \n",
    "    train_dataset = train_datagen.flow_from_directory(training, \n",
    "                                        target_size=IMG_SIZE, batch_size=BATCH_SIZE, \n",
    "                                        class_mode='categorical', \n",
    "                                        subset='training', \n",
    "                                        shuffle=True, seed=SEED+kkk*3)\n",
    "    \n",
    "    # prepare the validation set.\n",
    "    valid_datagen = ImageDataGenerator(validation_split=validation_split)\n",
    "    \n",
    "    validation_dataset = valid_datagen.flow_from_directory(training, \n",
    "                                        target_size=IMG_SIZE, batch_size=BATCH_SIZE, \n",
    "                                        class_mode='categorical', \n",
    "                                        subset='validation', \n",
    "                                        shuffle=True, seed=SEED+kkk*3)\n",
    "    \n",
    "    # choose the preprocessing function according to the selected pre-trained model.\n",
    "    if kkk == 0:\n",
    "        preprocess_input = tf.keras.applications.xception.preprocess_input\n",
    "    elif kkk in [1, 2, 3]:\n",
    "        preprocess_input = tf.keras.applications.resnet_v2.preprocess_input\n",
    "    elif kkk == 4:\n",
    "        preprocess_input = tf.keras.applications.inception_v3.preprocess_input\n",
    "    elif kkk == 5:\n",
    "        preprocess_input = tf.keras.applications.inception_resnet_v2.preprocess_input\n",
    "    elif kkk in [6, 7, 8, 9]:\n",
    "        preprocess_input = tf.keras.applications.efficientnet.preprocess_input\n",
    "\n",
    "    # load the model using the function defined in the previous cell of this notebook.\n",
    "    model2 = your_model(IMG_SIZE, nb_categories, kkk)\n",
    "    \n",
    "    # ····························································································\n",
    "    # Transfer Learning ··········································································\n",
    "    # compile the model.\n",
    "    base_learning_rate = 0.001\n",
    "    model2.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=base_learning_rate),\n",
    "                loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "    \n",
    "    # model fit.\n",
    "    initial_epochs = 7\n",
    "    history = model2.fit(train_dataset, validation_data=validation_dataset, epochs=initial_epochs, \n",
    "                         class_weight=class_weight, verbose=1)\n",
    "    \n",
    "    # store the history of the model fitted using transfer learning.\n",
    "    acc = [0.] + history.history['accuracy']\n",
    "    val_acc = [0.] + history.history['val_accuracy']\n",
    "    loss = history.history['loss']\n",
    "    val_loss = history.history['val_loss']\n",
    "    \n",
    "    # ····························································································\n",
    "    # Fine Tuning ················································································\n",
    "    # choose which layers to unfreeze, according the the selected pre-trained model.\n",
    "    if kkk == 0:\n",
    "        base_model = model2.layers[3]\n",
    "        fine_tune_at = 96 # (un-freeze from \"block11_sepconv1_act\" and below)\n",
    "    elif kkk == 1:\n",
    "        base_model = model2.layers[3]\n",
    "        fine_tune_at = 154 # (un-freeze from \"conv5_block1_preact_bn\" and below)\n",
    "    elif kkk == 2:\n",
    "        base_model = model2.layers[3]\n",
    "        fine_tune_at = 341 # (un-freeze from \"conv5_block1_preact_bn\" and below)\n",
    "    elif kkk == 3:\n",
    "        base_model = model2.layers[3]\n",
    "        fine_tune_at = 528 # (un-freeze from \"conv5_block1_preact_bn\" and below)\n",
    "    elif kkk == 4:\n",
    "        base_model = model2.layers[3]\n",
    "        fine_tune_at = 249 # (un-freeze from \"conv2d_385\" and below)\n",
    "    elif kkk == 5:\n",
    "        base_model = model2.layers[3]\n",
    "        fine_tune_at = 630 # (un-freeze from \"block8_1_mixed\" and below)\n",
    "    elif kkk == 6:\n",
    "        base_model = model2.layers[1]\n",
    "        fine_tune_at = 443 # (un-freeze from \"block7a_expand_conv\" and below)\n",
    "    elif kkk == 7:\n",
    "        base_model = model2.layers[1]\n",
    "        fine_tune_at = 530 # (un-freeze from \"block7a_expand_conv\" and below)\n",
    "    elif kkk == 8:\n",
    "        base_model = model2.layers[1]\n",
    "        fine_tune_at = 620 # (un-freeze from \"block7a_expand_conv\" and below)\n",
    "    elif kkk == 9:\n",
    "        base_model = model2.layers[1]\n",
    "        fine_tune_at = 752 # (un-freeze from \"block7a_expand_conv\" and below)\n",
    "    \n",
    "    # un-freeze everything, then re-freeze all the layers up to the chosen layer.\n",
    "    base_model.trainable = True\n",
    "    for layer in base_model.layers[:fine_tune_at]:\n",
    "        layer.trainable = False\n",
    "    \n",
    "    # compile the model and set checkpoint and earlystopping criteria.\n",
    "    model2.compile(optimizer=Adam(learning_rate=0.1*base_learning_rate, amsgrad=True), \n",
    "                loss='categorical_crossentropy', \n",
    "                metrics=['accuracy'])\n",
    "    checkpoint = ModelCheckpoint(f'model{MODEL_N+kkk}_fineTuned.h5', monitor='val_loss', \n",
    "                    verbose=1, save_best_only=True, save_weights_only=False, mode='auto')\n",
    "    earlystop = EarlyStopping(monitor=\"val_loss\", min_delta=0.005, patience=10,\n",
    "                    verbose=1, restore_best_weights=True)\n",
    "    \n",
    "    # model fit\n",
    "    fine_tune_epochs = 53\n",
    "    total_epochs =  initial_epochs + fine_tune_epochs\n",
    "    history_fine = model2.fit(train_dataset, validation_data=validation_dataset, \n",
    "                            epochs=total_epochs, initial_epoch=history.epoch[-1]+1, class_weight=class_weight, \n",
    "                            verbose=1, callbacks=[checkpoint, earlystop])\n",
    "    \n",
    "    # add the history of the fitted fine-tuned model to the previous one.\n",
    "    acc += history_fine.history['accuracy']\n",
    "    val_acc += history_fine.history['val_accuracy']\n",
    "    loss += history_fine.history['loss']\n",
    "    val_loss += history_fine.history['val_loss']\n",
    "    \n",
    "    # append the history of all the fitted models in the corresponding lists.\n",
    "    accs.append(acc)\n",
    "    val_accs.append(val_acc)\n",
    "    losses.append(loss)\n",
    "    val_losses.append(val_loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(8, 8))\n",
    "\n",
    "plt.subplot(2, 1, 1)\n",
    "plt.ylim([0.8,1.0])\n",
    "plt.xlim((1, total_epochs+1))\n",
    "plt.plot([initial_epochs,initial_epochs],\n",
    "        plt.ylim(), label='Start Fine Tuning')\n",
    "for acc, val_acc in zip(accs, val_accs):\n",
    "    plt.plot(acc, label='Training Accuracy')\n",
    "    plt.plot(val_acc, label='Validation Accuracy')\n",
    "plt.legend(loc='lower right')\n",
    "plt.title('Training and Validation Accuracy')\n",
    "\n",
    "plt.subplot(2, 1, 2)\n",
    "plt.ylim([0,1.0])\n",
    "plt.xlim((0, total_epochs))\n",
    "plt.plot([initial_epochs-1,initial_epochs-1],\n",
    "        plt.ylim(), label='Start Fine Tuning')\n",
    "for loss, val_loss in zip(losses, val_losses):\n",
    "    plt.plot(loss, label='Training Loss')\n",
    "    plt.plot(val_loss, label='Validation Loss')\n",
    "plt.legend(loc='upper right')\n",
    "plt.title('Training and Validation Loss')\n",
    "\n",
    "plt.xlabel('epoch')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 54306 images belonging to 14 classes.\n",
      "0.9878650609509079\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>10</th>\n",
       "      <th>11</th>\n",
       "      <th>12</th>\n",
       "      <th>13</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>3062</td>\n",
       "      <td>0</td>\n",
       "      <td>76</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>24</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>1502</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1903</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3852</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4062</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5506</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2655</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>12</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>9</td>\n",
       "      <td>10</td>\n",
       "      <td>2409</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>30</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2140</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>371</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5090</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1834</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1565</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>42</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>20</td>\n",
       "      <td>3</td>\n",
       "      <td>6</td>\n",
       "      <td>23</td>\n",
       "      <td>48</td>\n",
       "      <td>249</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>43</td>\n",
       "      <td>13</td>\n",
       "      <td>17696</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      0     1     2     3     4     5     6     7     8    9     10    11  \\\n",
       "0   3062     0    76     0     1     1    24     1     0    0     0     2   \n",
       "1      0  1502     0     0     0     0     0     0     0    0     0     0   \n",
       "2      2     0  1903     0     0     0     0     0     0    0     1     0   \n",
       "3      0     0     0  3852     0     0     0     0     0    0     0     0   \n",
       "4      0     0     0     0  4062     0     0     0     0    0     0     0   \n",
       "5      0     0     0     0     0  5506     0     1     0    0     0     0   \n",
       "6      1     1     0     0     0     0  2655     0     0    0     0     0   \n",
       "7     12     0     1     2     0     9    10  2409     2    0     0     0   \n",
       "8      1     0     0     0     1     0     0     0  2140    0     0     0   \n",
       "9      0     0     0     0     0     0     0     0     0  371     0     0   \n",
       "10     0     0     0     0     0     0     0     0     0    0  5090     0   \n",
       "11     0     0     0     0     0     0     0     1     0    0     0  1834   \n",
       "12     0     0     0     0     0     0     0     0     0    0     0     0   \n",
       "13    42     0     5    20     3     6    23    48   249    0    12    43   \n",
       "\n",
       "      12     13  \n",
       "0      1      3  \n",
       "1      0      0  \n",
       "2      0      0  \n",
       "3      0      0  \n",
       "4      1      0  \n",
       "5      0      0  \n",
       "6      0      0  \n",
       "7      0     30  \n",
       "8      0     10  \n",
       "9      0      0  \n",
       "10     0      0  \n",
       "11     0      0  \n",
       "12  1565      0  \n",
       "13    13  17696  "
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_num = MODEL_N\n",
    "\n",
    "tf.keras.backend.clear_session()\n",
    "\n",
    "test_datagen = ImageDataGenerator()\n",
    "test_dataset = valid_datagen.flow_from_directory(test, \n",
    "                                        target_size=IMG_SIZE, batch_size=BATCH_SIZE, \n",
    "                                        class_mode='categorical', \n",
    "                                        shuffle=False)\n",
    "y_true = test_dataset.classes\n",
    "\n",
    "model_loaded0 = tf.keras.models.load_model(f'model{model_num}_fineTuned.h5')\n",
    "model_loaded1 = tf.keras.models.load_model(f'model{model_num + 1}_fineTuned.h5')\n",
    "# model_loaded2 = tf.keras.models.load_model(f'model{model_num + 2}_fineTuned.h5')\n",
    "# model_loaded3 = tf.keras.models.load_model(f'model{model_num + 3}_fineTuned.h5')\n",
    "# model_loaded4 = tf.keras.models.load_model(f'model{model_num + 4}_fineTuned.h5')\n",
    "# model_loaded5 = tf.keras.models.load_model(f'model{model_num + 5}_fineTuned.h5')\n",
    "\n",
    "pred0 = model_loaded0.predict(test_dataset)\n",
    "pred1 = model_loaded1.predict(test_dataset)\n",
    "# pred2 = model_loaded2.predict(test_dataset)\n",
    "# pred3 = model_loaded3.predict(test_dataset)\n",
    "# pred4 = model_loaded4.predict(test_dataset)\n",
    "# pred5 = model_loaded5.predict(test_dataset)\n",
    "\n",
    "# pred = pred0 + pred1 + pred2 + pred3 + pred4 + pred5\n",
    "pred = pred0 + pred1\n",
    "y_pred = np.argmax(pred, axis=1)\n",
    "\n",
    "print(accuracy_score(y_true, y_pred))\n",
    "pd.DataFrame(confusion_matrix(y_true, y_pred))"
   ]
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "bda94d36c09323af121674f6ce1d4e4aadf75de425d2bfe41f4de98d4061af08"
  },
  "kernelspec": {
   "display_name": "Python 3.8.12 64-bit ('active2': conda)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.12"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}