neural_network_with_conv_and_lstm.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "from create_train_and_validation import create_train_and_validation_set\n",
    "from keras.layers import Dense, LSTM, Input, Conv2D, Dense, LSTM,MaxPooling2D , Flatten, TimeDistributed,Activation\n",
    "from keras.models import Sequential\n",
    "from keras.optimizers import Adam\n",
    "from keras.preprocessing.image import ImageDataGenerator, DirectoryIterator\n",
    "from metrics import f1_score\n",
    "from sklearn.metrics import classification_report\n",
    "from sklearn.preprocessing import LabelBinarizer\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "import keras\n",
    "import math\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import sys\n",
    "import tensorflow as tf\n",
    "\n",
    "from image_sequence_generator import ImageSequenceGenerator\n",
    "from utils import labels_to_binary"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Layer structure\n",
    "\n",
    "<img src=\"images/layer_structure.png\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Important: \n",
    "* modelB stands for model that uses binary feature labels\n",
    "* modelC stands for model that uses categorical feature labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "time_distributed_1 (TimeDist (None, 15, 61, 61, 96)    14208     \n",
      "_________________________________________________________________\n",
      "time_distributed_2 (TimeDist (None, 15, 30, 30, 96)    0         \n",
      "_________________________________________________________________\n",
      "time_distributed_3 (TimeDist (None, 15, 14, 14, 384)   332160    \n",
      "_________________________________________________________________\n",
      "time_distributed_4 (TimeDist (None, 15, 7, 7, 384)     0         \n",
      "_________________________________________________________________\n",
      "time_distributed_5 (TimeDist (None, 15, 7, 7, 512)     1769984   \n",
      "_________________________________________________________________\n",
      "time_distributed_6 (TimeDist (None, 15, 7, 7, 512)     2359808   \n",
      "_________________________________________________________________\n",
      "time_distributed_7 (TimeDist (None, 15, 7, 7, 384)     1769856   \n",
      "_________________________________________________________________\n",
      "time_distributed_8 (TimeDist (None, 15, 3, 3, 384)     0         \n",
      "_________________________________________________________________\n",
      "time_distributed_9 (TimeDist (None, 15, 3456)          0         \n",
      "_________________________________________________________________\n",
      "time_distributed_10 (TimeDis (None, 15, 4096)          14159872  \n",
      "_________________________________________________________________\n",
      "lstm_1 (LSTM)                (None, 256)               4457472   \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 3)                 771       \n",
      "_________________________________________________________________\n",
      "activation_1 (Activation)    (None, 3)                 0         \n",
      "=================================================================\n",
      "Total params: 24,864,131\n",
      "Trainable params: 24,864,131\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "Model: \"sequential_2\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "time_distributed_11 (TimeDis (None, 15, 61, 61, 96)    14208     \n",
      "_________________________________________________________________\n",
      "time_distributed_12 (TimeDis (None, 15, 30, 30, 96)    0         \n",
      "_________________________________________________________________\n",
      "time_distributed_13 (TimeDis (None, 15, 14, 14, 384)   332160    \n",
      "_________________________________________________________________\n",
      "time_distributed_14 (TimeDis (None, 15, 7, 7, 384)     0         \n",
      "_________________________________________________________________\n",
      "time_distributed_15 (TimeDis (None, 15, 7, 7, 512)     1769984   \n",
      "_________________________________________________________________\n",
      "time_distributed_16 (TimeDis (None, 15, 7, 7, 512)     2359808   \n",
      "_________________________________________________________________\n",
      "time_distributed_17 (TimeDis (None, 15, 7, 7, 384)     1769856   \n",
      "_________________________________________________________________\n",
      "time_distributed_18 (TimeDis (None, 15, 3, 3, 384)     0         \n",
      "_________________________________________________________________\n",
      "time_distributed_19 (TimeDis (None, 15, 3456)          0         \n",
      "_________________________________________________________________\n",
      "time_distributed_20 (TimeDis (None, 15, 4096)          14159872  \n",
      "_________________________________________________________________\n",
      "lstm_2 (LSTM)                (None, 256)               4457472   \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 8)                 2056      \n",
      "_________________________________________________________________\n",
      "activation_2 (Activation)    (None, 8)                 0         \n",
      "=================================================================\n",
      "Total params: 24,865,416\n",
      "Trainable params: 24,865,416\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "im_size = 128\n",
    "time_steps = 16 # length of image sequence\n",
    "channels = 3\n",
    "\n",
    "#modelB for binary response\n",
    "modelB = Sequential()\n",
    "modelB.add(TimeDistributed(Conv2D(filters=96, kernel_size=7,  strides=2, padding='valid'), input_shape=(time_steps,im_size,im_size,channels))) # first input shape is the len of seq\n",
    "modelB.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2))))\n",
    "modelB.add(TimeDistributed(Conv2D(filters=384, kernel_size=3,  strides=2, padding='valid')))\n",
    "modelB.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2))))\n",
    "modelB.add(TimeDistributed(Conv2D(filters=512, kernel_size=3, padding='same')))\n",
    "modelB.add(TimeDistributed(Conv2D(filters=512, kernel_size=3, padding='same')))\n",
    "modelB.add(TimeDistributed(Conv2D(filters=384, kernel_size=3, padding='same')))\n",
    "modelB.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2))))\n",
    "modelB.add(TimeDistributed(Flatten())) #The Flatten layer is only needed because LSTM shape should have one dimension per input.\n",
    "modelB.add(TimeDistributed(Dense(4096)))\n",
    "modelB.add(LSTM(256, return_sequences=False))\n",
    "modelB.add((Dense(3)))\n",
    "modelB.add((Activation('softmax')))\n",
    "print(modelB.summary())\n",
    "\n",
    "#modelC for categorical response\n",
    "modelC = Sequential()\n",
    "modelC.add(TimeDistributed(Conv2D(filters=96, kernel_size=7,  strides=2, padding='valid'), input_shape=(time_steps,im_size,im_size,channels))) # first input shape is the len of seq\n",
    "modelC.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2))))\n",
    "modelC.add(TimeDistributed(Conv2D(filters=384, kernel_size=3,  strides=2, padding='valid')))\n",
    "modelC.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2))))\n",
    "modelC.add(TimeDistributed(Conv2D(filters=512, kernel_size=3, padding='same')))\n",
    "modelC.add(TimeDistributed(Conv2D(filters=512, kernel_size=3, padding='same')))\n",
    "modelC.add(TimeDistributed(Conv2D(filters=384, kernel_size=3, padding='same')))\n",
    "modelC.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2))))\n",
    "modelC.add(TimeDistributed(Flatten())) #The Flatten layer is only needed because LSTM shape should have one dimension per input.\n",
    "modelC.add(TimeDistributed(Dense(4096)))\n",
    "modelC.add(LSTM(256, return_sequences=False))\n",
    "modelC.add((Dense(8)))\n",
    "modelC.add((Activation('softmax')))\n",
    "print(modelC.summary())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Binary labels\n",
    "\n",
    "Comment: following warning can be ignored, it occurs during converting labels into binary \n",
    "\n",
    "Warning: .../miniconda3/envs/stacc/lib/python3.6/site-packages/sklearn/preprocessing/label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
    "  .format(sorted(unknown, key=str)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "initial labels: \n",
      " ['BOO', 'BLO', 'BOR', 'BLR', 'OLR', 'OLO', 'OOR', 'OOO']\n",
      "labels in binary: \n",
      " [[1 0 0]\n",
      " [1 1 0]\n",
      " [1 0 1]\n",
      " [1 1 1]\n",
      " [0 1 1]\n",
      " [0 1 0]\n",
      " [0 0 1]\n",
      " [0 0 0]]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n"
     ]
    }
   ],
   "source": [
    "ini_labels = ['BOO', 'BLO', 'BOR', 'BLR', 'OLR', 'OLO', 'OOR', 'OOO']\n",
    "print('initial labels: \\n', ini_labels)\n",
    "labels_binary = labels_to_binary(ini_labels)\n",
    "print('labels in binary: \\n', labels_binary)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Split all non-test folders into train and validation\n",
    "Train data is 70% and validation is 30%.\n",
    "Output is \n",
    "* list of paths to folders belonging to training set (train_folder_list), validation set (valid_folder_list) and test set (test_folder_list)\n",
    "* number of image sequences in train, validation and test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "total number of non-test folders is  533  for difficulty  All\n",
      "total number of non-test image sequences is  40350\n",
      "total number of test image sequences is  13189\n",
      "\n",
      "final train set contains  28180  image sequences ( 70 % )\n",
      "final validation set contains  12170  image sequences ( 30 % )\n",
      "final test set contains  13189  image sequences\n"
     ]
    }
   ],
   "source": [
    "path=\"./rear_signal_dataset\"\n",
    "difficulty=\"All\"\n",
    "\n",
    "train_folder_list, valid_folder_list, test_folder_list, count_train_seq, count_valid_seq, count_test_seq = create_train_and_validation_set(path, difficulty)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### True labels of test data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "39195\n",
      "First test data binary labels:\n",
      " [[1 0 0]\n",
      " [0 1 0]\n",
      " [0 0 0]\n",
      " ...\n",
      " [0 0 0]\n",
      " [0 1 0]\n",
      " [0 0 0]]\n",
      "First test data categorical:\n",
      " [[0 0 0 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]\n",
      " ...\n",
      " [0 0 0 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]]\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "\n",
    "lb = LabelBinarizer()\n",
    "labels = set(['BOO', 'BLO', 'BOR', 'BLR', 'OLR', 'OLO', 'OOR', 'OOO'])\n",
    "lb.fit(list(labels))\n",
    "\n",
    "raw_label_list = []\n",
    "for folder in test_folder_list:\n",
    "    path, dirs, files = next(os.walk(folder))\n",
    "    if(len(dirs)>0):\n",
    "        _ , _, files2 = next(os.walk(path+\"/\"+dirs[0]))\n",
    "        val=len(files2)-time_steps\n",
    "        raw_label_list.extend(folder.split(\"_\")[-2]*val)\n",
    "print(len(raw_label_list))\n",
    "Y_test_binary = labels_to_binary(raw_label_list)\n",
    "Y_test_categ = lb.transform(raw_label_list)\n",
    "print('First test data binary labels:\\n', Y_test_binary)\n",
    "print('First test data categorical:\\n', Y_test_categ)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "https://www.pyimagesearch.com/2018/12/24/how-to-use-keras-fit-and-fit_generator-a-hands-on-tutorial/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Info\n",
      "difficulty: All \n",
      "label type: binary \n",
      "image size: 128 \n",
      "epochs: 5 \n",
      "batch size: 20, \n",
      "learning_rate: 0.01\n",
      "NUM_TRAIN_SEQ: 28180\n",
      "NUM_VALID_SEQ: 12170\n",
      "NUM_TEST_SEQ: 13189\n"
     ]
    }
   ],
   "source": [
    "label_type = \"binary\"\n",
    "# initialize the number of epochs to train for and batch size\n",
    "NUM_EPOCHS = 5 #10\n",
    "BS = 20 #32\n",
    "learning_rate = 0.01\n",
    "#standard batch size is 32 või 64, but kernal dies with BS larger than 8\n",
    "\n",
    "print('Info\\ndifficulty: {0} \\nlabel type: {1} \\nimage size: {2} \\nepochs: {3} \\nbatch size: {4}, \\nlearning_rate: {5}'.format(difficulty,label_type,im_size, NUM_EPOCHS, BS,learning_rate))\n",
    "\n",
    "# initialize the total number of training and testing sequences\n",
    "# initialize the share of sample\n",
    "ss = 1\n",
    "NUM_TRAIN_SEQ = round(count_train_seq*ss,0)\n",
    "NUM_VALID_SEQ = round(count_valid_seq*ss,0)\n",
    "NUM_TEST_SEQ = round(count_test_seq*ss,0)\n",
    "\n",
    "print('NUM_TRAIN_SEQ:', NUM_TRAIN_SEQ)\n",
    "print('NUM_VALID_SEQ:', NUM_VALID_SEQ)\n",
    "print('NUM_TEST_SEQ:', NUM_TEST_SEQ)\n",
    "\n",
    "# TODO: not in use\n",
    "# construct the training image generator for data augmentation\n",
    "# image data augmentation object will randomly rotate, flip, shear, etc. our training images.\n",
    "aug = ImageDataGenerator(rotation_range=20, zoom_range=0.15,\n",
    "    width_shift_range=0.2, height_shift_range=0.2, shear_range=0.15,\n",
    "    horizontal_flip=True, fill_mode=\"nearest\")\n",
    "\n",
    "img_seq_gen_train = ImageSequenceGenerator()\n",
    "img_seq_gen_valid = ImageSequenceGenerator()\n",
    "\n",
    "trainGenB = img_seq_gen_train.png_image_generator(path, bs=BS, folder_list=train_folder_list, difficulty=difficulty, label_type = \"binary\", aug=None)\n",
    "validGenB = img_seq_gen_valid.png_image_generator(path, bs=BS, folder_list=valid_folder_list, difficulty=difficulty, label_type = \"binary\", aug=None)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Important: if target is categorical use categorical_crossentropy, if target is binary use binary_crossentropy\n",
    "TODO:\n",
    "* salvestada mudel pärast treenimist, siis ei pea iga kord kui notebook uuesti mudelit nullist treenima hakata (näiteks https://machinelearningmastery.com/save-load-keras-deep-learning-models/)\n",
    "* teha joonised mudeli lossi ja accuracy muutustest treenimise käigus (näiteks https://machinelearningmastery.com/display-deep-learning-model-training-history-in-keras/)\n",
    "\n",
    "Comment: following warning can be ignored, it occurs during converting labels into binary \n",
    "\n",
    "Warning: .../miniconda3/envs/stacc/lib/python3.6/site-packages/sklearn/preprocessing/label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
    "  .format(sorted(unknown, key=str)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Binary model: steps per epoch: 1405, validation steps: 611\n",
      "[INFO] training w/ generator...\n",
      "Epoch 1/5\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1405/1405 [==============================] - 987s 703ms/step - loss: 0.5315 - categorical_accuracy: 0.7456 - accuracy: 0.7441 - val_loss: 0.5161 - val_categorical_accuracy: 0.6702 - val_accuracy: 0.7194\n",
      "Epoch 2/5\n",
      "1405/1405 [==============================] - 976s 694ms/step - loss: 0.5293 - categorical_accuracy: 0.7203 - accuracy: 0.7429 - val_loss: 0.5763 - val_categorical_accuracy: 0.7069 - val_accuracy: 0.7158\n",
      "Epoch 3/5\n",
      "1405/1405 [==============================] - 1033s 735ms/step - loss: 0.5263 - categorical_accuracy: 0.7289 - accuracy: 0.7458 - val_loss: 0.5374 - val_categorical_accuracy: 0.5736 - val_accuracy: 0.6948\n",
      "Epoch 4/5\n",
      "1405/1405 [==============================] - 1024s 729ms/step - loss: 0.5281 - categorical_accuracy: 0.7241 - accuracy: 0.7426 - val_loss: 0.5578 - val_categorical_accuracy: 0.7409 - val_accuracy: 0.7189\n",
      "Epoch 5/5\n",
      "1405/1405 [==============================] - 1036s 738ms/step - loss: 0.5249 - categorical_accuracy: 0.7258 - accuracy: 0.7448 - val_loss: 0.5225 - val_categorical_accuracy: 0.7289 - val_accuracy: 0.7286\n"
     ]
    }
   ],
   "source": [
    "opt = Adam(learning_rate=learning_rate)\n",
    "modelB.compile(loss='binary_crossentropy', optimizer=opt, metrics=['categorical_accuracy','accuracy'])\n",
    "\n",
    "print('Binary model: steps per epoch: {0}, validation steps: {1}'.format(NUM_TRAIN_SEQ // BS, NUM_VALID_SEQ // BS))    \n",
    "    \n",
    "# train the network\n",
    "print(\"[INFO] training w/ generator...\")\n",
    "historyB = modelB.fit(\n",
    "    x=trainGenB,\n",
    "    steps_per_epoch=NUM_TRAIN_SEQ // BS,\n",
    "    validation_data=validGenB,\n",
    "    validation_steps=NUM_VALID_SEQ // BS,\n",
    "    epochs=NUM_EPOCHS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Categorical model: steps per epoch: 1405, validation steps: 611\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "1405/1405 [==============================] - 1018s 725ms/step - loss: 1.5634 - accuracy: 0.3831 - val_loss: 2.0694 - val_accuracy: 0.3264\n",
      "Epoch 2/5\n",
      "1405/1405 [==============================] - 1150s 818ms/step - loss: 1.5108 - accuracy: 0.3905 - val_loss: 1.7336 - val_accuracy: 0.3444\n",
      "Epoch 3/5\n",
      "1405/1405 [==============================] - 1033s 735ms/step - loss: 1.4860 - accuracy: 0.3975 - val_loss: 1.6246 - val_accuracy: 0.3658\n",
      "Epoch 4/5\n",
      "1405/1405 [==============================] - 1068s 760ms/step - loss: 1.5055 - accuracy: 0.3965 - val_loss: 1.3755 - val_accuracy: 0.3231\n",
      "Epoch 5/5\n",
      "1405/1405 [==============================] - 1118s 795ms/step - loss: 1.4850 - accuracy: 0.4049 - val_loss: 1.6781 - val_accuracy: 0.3386\n"
     ]
    }
   ],
   "source": [
    "# categorical model \n",
    "img_seq_gen_train = ImageSequenceGenerator()\n",
    "img_seq_gen_valid = ImageSequenceGenerator()\n",
    "\n",
    "trainGenC = img_seq_gen_train.png_image_generator(path, bs=BS, folder_list=train_folder_list, difficulty=difficulty, label_type = \"categorical\", aug=None)\n",
    "validGenC = img_seq_gen_valid.png_image_generator(path, bs=BS, folder_list=valid_folder_list, difficulty=difficulty, label_type = \"categorical\", aug=None)\n",
    "\n",
    "modelC.compile(loss=\"categorical_crossentropy\", optimizer=opt,metrics=[\"accuracy\"])\n",
    "print('Categorical model: steps per epoch: {0}, validation steps: {1}'.format(NUM_TRAIN_SEQ // BS, NUM_VALID_SEQ // BS))    \n",
    "\n",
    "historyC = modelC.fit(\n",
    "    x=trainGenC,\n",
    "    steps_per_epoch=NUM_TRAIN_SEQ // BS,\n",
    "    validation_data=validGenC,\n",
    "    validation_steps=NUM_VALID_SEQ // BS,\n",
    "    epochs=NUM_EPOCHS)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Saving the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Saving binary model under name:  modelb_All_binary_epochs5_bs20_imsize128\n",
      "\n",
      "Saving categorical model under name:  modelc_All_binary_epochs5_bs20_imsize128\n"
     ]
    }
   ],
   "source": [
    "model_file_name_b = 'modelb_' + difficulty + '_' + label_type + '_epochs' + str(NUM_EPOCHS) + '_bs' + str(BS) + '_imsize' + str(im_size)\n",
    "print('\\nSaving binary model under name: ', model_file_name_b)\n",
    "modelB.save('saved_models/'+model_file_name_b)\n",
    "\n",
    "model_file_name_c = 'modelc_' + difficulty + '_' + label_type + '_epochs' + str(NUM_EPOCHS) + '_bs' + str(BS) + '_imsize' + str(im_size)\n",
    "print('\\nSaving categorical model under name: ', model_file_name_c)\n",
    "modelC.save('saved_models/'+model_file_name_c)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loading the saved model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\keras\\engine\\saving.py:384: UserWarning: Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n",
      "  warnings.warn('Error in loading the saved optimizer '\n"
     ]
    }
   ],
   "source": [
    "model = keras.models.load_model('saved_models/'+\"modelb_All_binary_epochs5_bs20_imsize128\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training accuracy and loss visualisation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# modelB\n",
    "# summarize history for accuracy\n",
    "plt.plot(historyB.history['accuracy'])\n",
    "plt.plot(historyB.history['val_accuracy'])\n",
    "#plt.title('model binary accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "plt.savefig('images/accuracy_' + model_file_name_b + '.png',dpi=200)\n",
    "plt.show()\n",
    "\n",
    "# summarize history for loss\n",
    "plt.plot(historyB.history['loss'])\n",
    "plt.plot(historyB.history['val_loss'])\n",
    "#plt.title('model binary loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "plt.savefig('images/loss_' + model_file_name_b + '.png',dpi=200)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# modelC\n",
    "# summarize history for accuracy\n",
    "plt.plot(historyC.history['accuracy'])\n",
    "plt.plot(historyC.history['val_accuracy'])\n",
    "#plt.title('model binary accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "plt.savefig('images/accuracy_' + model_file_name_c + '.png',dpi=200)\n",
    "plt.show()\n",
    "\n",
    "# summarize history for loss\n",
    "plt.plot(historyC.history['loss'])\n",
    "plt.plot(historyC.history['val_loss'])\n",
    "#plt.title('model binary loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "plt.savefig('images/loss_' + model_file_name_c + '.png',dpi=200)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Prediction\n",
    "* Source: https://www.pyimagesearch.com/2018/12/24/how-to-use-keras-fit-and-fit_generator-a-hands-on-tutorial/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "predicted probability:\n",
      " [[0.37053224 0.5042987  0.12516905]\n",
      " [0.52209926 0.38811985 0.08978093]\n",
      " [0.43873665 0.42071447 0.14054878]\n",
      " [0.4315433  0.22450283 0.34395388]]\n",
      "predictions:\n",
      " [[0 1 0]\n",
      " [1 0 0]\n",
      " [0 0 0]\n",
      " [0 0 0]]\n",
      "true labels:\n",
      " [[1 0 0]\n",
      " [0 1 0]\n",
      " [0 0 0]\n",
      " [1 0 0]]\n",
      "[INFO] evaluating network...\n",
      "39195\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           B       0.27      0.48      0.35      3120\n",
      "           L       0.29      0.01      0.03      2541\n",
      "           R       0.13      0.02      0.03      1488\n",
      "\n",
      "   micro avg       0.27      0.22      0.24      7149\n",
      "   macro avg       0.23      0.17      0.14      7149\n",
      "weighted avg       0.25      0.22      0.17      7149\n",
      " samples avg       0.12      0.12      0.12      7149\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# re-initialize our testing data generator, this time for evaluating\n",
    "from image_sequence_generator_old import ImageSequenceGeneratorOld\n",
    "\n",
    "img_seq_gen_test = ImageSequenceGeneratorOld()\n",
    "testGen = img_seq_gen_test.png_image_generator(path, bs=BS, folder_list=test_folder_list, difficulty=difficulty, label_type = label_type, aug=None)\n",
    "\n",
    "# label with the corresponding predicted probabilities for each class e.g. [0.4, 0.7, 0.2]\n",
    "predict_prob = model.predict(x=testGen, steps=(NUM_TEST_SEQ // BS) + 1)\n",
    "print('predicted probability:\\n',predict_prob[0:4])\n",
    "# convert probabilities to binary, with given threshold  e.g. [0,1,0]\n",
    "threshold = 0.5\n",
    "Y_predict_binary = np.where(predict_prob > threshold, 1, 0)\n",
    "print('predictions:\\n', Y_predict_binary[0:4])\n",
    "print('true labels:\\n', Y_test_binary[0:4])\n",
    "\n",
    "# show a nicely formatted classification report\n",
    "print(\"[INFO] evaluating network...\")\n",
    "print(len(Y_test_binary))\n",
    "print(classification_report(Y_test_binary[0:len(Y_predict_binary)], Y_predict_binary, target_names=['B','L','R']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "131\n",
      "131\n",
      "131\n",
      "131\n",
      "133\n",
      "[INFO] validation w/ generator...\n",
      "Epoch 1/20\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "26/26 [==============================] - 24s 915ms/step - loss: 0.5751 - accuracy: 0.6917 - val_loss: 0.6135 - val_accuracy: 0.7500\n",
      "Epoch 2/20\n",
      "26/26 [==============================] - 20s 780ms/step - loss: 0.5724 - accuracy: 0.6942 - val_loss: 0.5604 - val_accuracy: 0.7250\n",
      "Epoch 3/20\n",
      "26/26 [==============================] - 20s 752ms/step - loss: 0.5699 - accuracy: 0.7071 - val_loss: 0.4776 - val_accuracy: 0.7222\n",
      "Epoch 4/20\n",
      "26/26 [==============================] - 18s 695ms/step - loss: 0.5907 - accuracy: 0.6917 - val_loss: 0.5902 - val_accuracy: 0.7306\n",
      "Epoch 5/20\n",
      "26/26 [==============================] - 18s 702ms/step - loss: 0.5858 - accuracy: 0.7006 - val_loss: 0.5308 - val_accuracy: 0.7889\n",
      "Epoch 6/20\n",
      "26/26 [==============================] - 17s 665ms/step - loss: 0.5724 - accuracy: 0.7179 - val_loss: 0.5233 - val_accuracy: 0.7639\n",
      "Epoch 7/20\n",
      "26/26 [==============================] - 18s 690ms/step - loss: 0.6365 - accuracy: 0.6846 - val_loss: 0.7096 - val_accuracy: 0.7194\n",
      "Epoch 8/20\n",
      "26/26 [==============================] - 17s 668ms/step - loss: 0.6015 - accuracy: 0.6897 - val_loss: 0.5624 - val_accuracy: 0.7583\n",
      "Epoch 9/20\n",
      "26/26 [==============================] - 17s 662ms/step - loss: 0.5755 - accuracy: 0.7006 - val_loss: 0.5741 - val_accuracy: 0.6889\n",
      "Epoch 10/20\n",
      "26/26 [==============================] - 15s 576ms/step - loss: 0.5644 - accuracy: 0.7122 - val_loss: 0.4467 - val_accuracy: 0.7111\n",
      "Epoch 11/20\n",
      "26/26 [==============================] - 17s 657ms/step - loss: 0.5555 - accuracy: 0.7096 - val_loss: 0.5431 - val_accuracy: 0.7250\n",
      "Epoch 12/20\n",
      "26/26 [==============================] - 17s 666ms/step - loss: 0.5705 - accuracy: 0.7051 - val_loss: 0.4684 - val_accuracy: 0.7750\n",
      "Epoch 13/20\n",
      "26/26 [==============================] - 17s 648ms/step - loss: 0.5614 - accuracy: 0.7083 - val_loss: 0.5890 - val_accuracy: 0.7472\n",
      "Epoch 14/20\n",
      "26/26 [==============================] - 17s 640ms/step - loss: 0.5792 - accuracy: 0.6949 - val_loss: 0.5139 - val_accuracy: 0.8139\n",
      "Epoch 15/20\n",
      "26/26 [==============================] - 16s 603ms/step - loss: 0.5845 - accuracy: 0.6910 - val_loss: 0.5419 - val_accuracy: 0.7472\n",
      "Epoch 16/20\n",
      "26/26 [==============================] - 17s 653ms/step - loss: 0.5696 - accuracy: 0.7160 - val_loss: 0.5261 - val_accuracy: 0.7389\n",
      "Epoch 17/20\n",
      "26/26 [==============================] - 16s 607ms/step - loss: 0.5758 - accuracy: 0.7173 - val_loss: 0.5268 - val_accuracy: 0.7417\n",
      "Epoch 18/20\n",
      "26/26 [==============================] - 15s 593ms/step - loss: 0.5886 - accuracy: 0.6974 - val_loss: 0.7407 - val_accuracy: 0.7861\n",
      "Epoch 19/20\n",
      "26/26 [==============================] - 15s 569ms/step - loss: 0.5790 - accuracy: 0.6942 - val_loss: 0.5773 - val_accuracy: 0.7472\n",
      "Epoch 20/20\n",
      "26/26 [==============================] - 17s 639ms/step - loss: 0.5771 - accuracy: 0.7128 - val_loss: 0.5422 - val_accuracy: 0.7417\n",
      "[INFO] validation w/ generator...\n",
      "Epoch 1/20\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "26/26 [==============================] - 18s 706ms/step - loss: 0.5490 - accuracy: 0.7308 - val_loss: 0.4918 - val_accuracy: 0.7750\n",
      "Epoch 2/20\n",
      "26/26 [==============================] - 19s 724ms/step - loss: 0.5846 - accuracy: 0.6942 - val_loss: 0.5264 - val_accuracy: 0.7694\n",
      "Epoch 3/20\n",
      "26/26 [==============================] - 18s 690ms/step - loss: 0.5708 - accuracy: 0.7135 - val_loss: 0.5345 - val_accuracy: 0.7556\n",
      "Epoch 4/20\n",
      "26/26 [==============================] - 17s 645ms/step - loss: 0.5776 - accuracy: 0.7167 - val_loss: 0.5300 - val_accuracy: 0.7444\n",
      "Epoch 5/20\n",
      "26/26 [==============================] - 17s 641ms/step - loss: 0.5649 - accuracy: 0.7186 - val_loss: 0.4890 - val_accuracy: 0.7528\n",
      "Epoch 6/20\n",
      "26/26 [==============================] - 18s 678ms/step - loss: 0.5779 - accuracy: 0.6981 - val_loss: 0.5067 - val_accuracy: 0.7778\n",
      "Epoch 7/20\n",
      "26/26 [==============================] - 14s 550ms/step - loss: 0.5805 - accuracy: 0.6968 - val_loss: 0.5170 - val_accuracy: 0.7722\n",
      "Epoch 8/20\n",
      "26/26 [==============================] - 15s 570ms/step - loss: 0.5719 - accuracy: 0.7096 - val_loss: 0.5141 - val_accuracy: 0.7500\n",
      "Epoch 9/20\n",
      "26/26 [==============================] - 15s 559ms/step - loss: 0.5926 - accuracy: 0.6936 - val_loss: 0.4717 - val_accuracy: 0.7667\n",
      "Epoch 10/20\n",
      "26/26 [==============================] - 14s 552ms/step - loss: 0.5572 - accuracy: 0.7263 - val_loss: 0.5628 - val_accuracy: 0.6972\n",
      "Epoch 11/20\n",
      "26/26 [==============================] - 15s 566ms/step - loss: 0.5621 - accuracy: 0.7263 - val_loss: 0.5945 - val_accuracy: 0.7611\n",
      "Epoch 12/20\n",
      "26/26 [==============================] - 16s 632ms/step - loss: 0.5592 - accuracy: 0.7256 - val_loss: 0.5308 - val_accuracy: 0.7444\n",
      "Epoch 13/20\n",
      "26/26 [==============================] - 17s 652ms/step - loss: 0.5628 - accuracy: 0.7244 - val_loss: 0.5078 - val_accuracy: 0.7806\n",
      "Epoch 14/20\n",
      "26/26 [==============================] - 14s 545ms/step - loss: 0.5708 - accuracy: 0.7269 - val_loss: 0.5572 - val_accuracy: 0.7056\n",
      "Epoch 15/20\n",
      "26/26 [==============================] - 14s 549ms/step - loss: 0.5784 - accuracy: 0.7205 - val_loss: 0.6342 - val_accuracy: 0.6917\n",
      "Epoch 16/20\n",
      "26/26 [==============================] - 15s 568ms/step - loss: 0.5751 - accuracy: 0.7019 - val_loss: 0.5496 - val_accuracy: 0.7056\n",
      "Epoch 17/20\n",
      "26/26 [==============================] - 15s 584ms/step - loss: 0.5754 - accuracy: 0.7077 - val_loss: 0.5372 - val_accuracy: 0.7694\n",
      "Epoch 18/20\n",
      "26/26 [==============================] - 15s 563ms/step - loss: 0.5764 - accuracy: 0.7006 - val_loss: 0.5448 - val_accuracy: 0.7611\n",
      "Epoch 19/20\n",
      "26/26 [==============================] - 15s 564ms/step - loss: 0.5632 - accuracy: 0.7269 - val_loss: 0.5142 - val_accuracy: 0.6972\n",
      "Epoch 20/20\n",
      "26/26 [==============================] - 15s 561ms/step - loss: 0.5822 - accuracy: 0.7077 - val_loss: 0.5489 - val_accuracy: 0.7639\n",
      "[INFO] validation w/ generator...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "26/26 [==============================] - 20s 757ms/step - loss: 0.5783 - accuracy: 0.7071 - val_loss: 0.4884 - val_accuracy: 0.7139\n",
      "Epoch 2/20\n",
      "26/26 [==============================] - 16s 616ms/step - loss: 0.6009 - accuracy: 0.6865 - val_loss: 0.5605 - val_accuracy: 0.7694\n",
      "Epoch 3/20\n",
      "26/26 [==============================] - 17s 639ms/step - loss: 0.5811 - accuracy: 0.7006 - val_loss: 0.5005 - val_accuracy: 0.7250\n",
      "Epoch 4/20\n",
      "26/26 [==============================] - 17s 644ms/step - loss: 0.5669 - accuracy: 0.7141 - val_loss: 0.6257 - val_accuracy: 0.7111\n",
      "Epoch 5/20\n",
      "26/26 [==============================] - 17s 673ms/step - loss: 0.5741 - accuracy: 0.7115 - val_loss: 0.5476 - val_accuracy: 0.7000\n",
      "Epoch 6/20\n",
      "26/26 [==============================] - 15s 585ms/step - loss: 0.5887 - accuracy: 0.6962 - val_loss: 0.5486 - val_accuracy: 0.7000\n",
      "Epoch 7/20\n",
      "26/26 [==============================] - 17s 670ms/step - loss: 0.6035 - accuracy: 0.6859 - val_loss: 0.6146 - val_accuracy: 0.7444\n",
      "Epoch 8/20\n",
      "26/26 [==============================] - 17s 645ms/step - loss: 0.5909 - accuracy: 0.6891 - val_loss: 0.5815 - val_accuracy: 0.6778\n",
      "Epoch 9/20\n",
      "26/26 [==============================] - 16s 611ms/step - loss: 0.5704 - accuracy: 0.7045 - val_loss: 0.5699 - val_accuracy: 0.7278\n",
      "Epoch 10/20\n",
      "26/26 [==============================] - 17s 651ms/step - loss: 0.5614 - accuracy: 0.7167 - val_loss: 0.5824 - val_accuracy: 0.7083\n",
      "Epoch 11/20\n",
      "26/26 [==============================] - 16s 597ms/step - loss: 0.5577 - accuracy: 0.7224 - val_loss: 0.5523 - val_accuracy: 0.7639\n",
      "Epoch 12/20\n",
      "26/26 [==============================] - 17s 635ms/step - loss: 0.5644 - accuracy: 0.7199 - val_loss: 0.5797 - val_accuracy: 0.7139\n",
      "Epoch 13/20\n",
      "26/26 [==============================] - 17s 661ms/step - loss: 0.5610 - accuracy: 0.7218 - val_loss: 0.4802 - val_accuracy: 0.7361\n",
      "Epoch 14/20\n",
      "26/26 [==============================] - 17s 659ms/step - loss: 0.5582 - accuracy: 0.7173 - val_loss: 0.5000 - val_accuracy: 0.7417\n",
      "Epoch 15/20\n",
      "26/26 [==============================] - 15s 574ms/step - loss: 0.5663 - accuracy: 0.7077 - val_loss: 0.5629 - val_accuracy: 0.7278\n",
      "Epoch 16/20\n",
      "26/26 [==============================] - 17s 653ms/step - loss: 0.5728 - accuracy: 0.7013 - val_loss: 0.5050 - val_accuracy: 0.7056\n",
      "Epoch 17/20\n",
      "26/26 [==============================] - 18s 677ms/step - loss: 0.5689 - accuracy: 0.7205 - val_loss: 0.5556 - val_accuracy: 0.7417\n",
      "Epoch 18/20\n",
      "26/26 [==============================] - 15s 592ms/step - loss: 0.5642 - accuracy: 0.7218 - val_loss: 0.5464 - val_accuracy: 0.6667\n",
      "Epoch 19/20\n",
      "26/26 [==============================] - 17s 673ms/step - loss: 0.5795 - accuracy: 0.7045 - val_loss: 0.5894 - val_accuracy: 0.7611\n",
      "Epoch 20/20\n",
      "26/26 [==============================] - 16s 604ms/step - loss: 0.5689 - accuracy: 0.7192 - val_loss: 0.5687 - val_accuracy: 0.6972\n",
      "[INFO] validation w/ generator...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "26/26 [==============================] - 20s 781ms/step - loss: 0.5714 - accuracy: 0.7103 - val_loss: 0.5314 - val_accuracy: 0.7083\n",
      "Epoch 2/20\n",
      "26/26 [==============================] - 19s 721ms/step - loss: 0.5705 - accuracy: 0.7064 - val_loss: 0.5298 - val_accuracy: 0.7167\n",
      "Epoch 3/20\n",
      "26/26 [==============================] - 17s 660ms/step - loss: 0.5724 - accuracy: 0.7045 - val_loss: 0.5497 - val_accuracy: 0.7556\n",
      "Epoch 4/20\n",
      "26/26 [==============================] - 18s 690ms/step - loss: 0.5765 - accuracy: 0.7077 - val_loss: 0.5778 - val_accuracy: 0.7500\n",
      "Epoch 5/20\n",
      "26/26 [==============================] - 16s 632ms/step - loss: 0.5690 - accuracy: 0.7122 - val_loss: 0.6123 - val_accuracy: 0.6889\n",
      "Epoch 6/20\n",
      "26/26 [==============================] - 17s 657ms/step - loss: 0.5514 - accuracy: 0.7295 - val_loss: 0.4821 - val_accuracy: 0.7000\n",
      "Epoch 7/20\n",
      "26/26 [==============================] - 17s 652ms/step - loss: 0.5558 - accuracy: 0.7109 - val_loss: 0.5189 - val_accuracy: 0.7194\n",
      "Epoch 8/20\n",
      "26/26 [==============================] - 18s 692ms/step - loss: 0.5673 - accuracy: 0.6987 - val_loss: 0.4808 - val_accuracy: 0.7417\n",
      "Epoch 9/20\n",
      "26/26 [==============================] - 17s 635ms/step - loss: 0.5592 - accuracy: 0.7109 - val_loss: 0.5349 - val_accuracy: 0.7278\n",
      "Epoch 10/20\n",
      "26/26 [==============================] - 17s 667ms/step - loss: 0.5540 - accuracy: 0.7167 - val_loss: 0.5939 - val_accuracy: 0.7139\n",
      "Epoch 11/20\n",
      "26/26 [==============================] - 18s 708ms/step - loss: 0.5845 - accuracy: 0.7019 - val_loss: 0.4427 - val_accuracy: 0.7361\n",
      "Epoch 12/20\n",
      "26/26 [==============================] - 19s 739ms/step - loss: 0.5566 - accuracy: 0.7256 - val_loss: 0.5632 - val_accuracy: 0.6861\n",
      "Epoch 13/20\n",
      "26/26 [==============================] - 17s 638ms/step - loss: 0.5983 - accuracy: 0.6936 - val_loss: 0.6035 - val_accuracy: 0.7694\n",
      "Epoch 14/20\n",
      "26/26 [==============================] - 18s 678ms/step - loss: 0.5729 - accuracy: 0.7090 - val_loss: 0.5699 - val_accuracy: 0.7139\n",
      "Epoch 15/20\n",
      "26/26 [==============================] - 16s 631ms/step - loss: 0.5677 - accuracy: 0.7064 - val_loss: 0.6180 - val_accuracy: 0.7333\n",
      "Epoch 16/20\n",
      "26/26 [==============================] - 16s 612ms/step - loss: 0.5547 - accuracy: 0.7276 - val_loss: 0.5374 - val_accuracy: 0.7056\n",
      "Epoch 17/20\n",
      "26/26 [==============================] - 17s 671ms/step - loss: 0.5404 - accuracy: 0.7436 - val_loss: 0.4746 - val_accuracy: 0.6889\n",
      "Epoch 18/20\n",
      "26/26 [==============================] - 17s 669ms/step - loss: 0.5839 - accuracy: 0.6929 - val_loss: 0.4956 - val_accuracy: 0.7444\n",
      "Epoch 19/20\n",
      "26/26 [==============================] - 17s 659ms/step - loss: 0.5637 - accuracy: 0.7077 - val_loss: 0.4650 - val_accuracy: 0.7444\n",
      "Epoch 20/20\n",
      "26/26 [==============================] - 17s 662ms/step - loss: 0.5667 - accuracy: 0.7147 - val_loss: 0.5742 - val_accuracy: 0.7583\n",
      "[INFO] validation w/ generator...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\sklearn\\preprocessing\\label.py:951: UserWarning: unknown class(es) ['O'] will be ignored\n",
      "  .format(sorted(unknown, key=str)))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "26/26 [==============================] - 19s 726ms/step - loss: 0.5403 - accuracy: 0.7327 - val_loss: 0.6945 - val_accuracy: 0.5889\n",
      "Epoch 2/20\n",
      "26/26 [==============================] - 18s 677ms/step - loss: 0.5480 - accuracy: 0.7327 - val_loss: 0.7031 - val_accuracy: 0.6056\n",
      "Epoch 3/20\n",
      "26/26 [==============================] - 15s 593ms/step - loss: 0.5412 - accuracy: 0.7474 - val_loss: 0.6238 - val_accuracy: 0.6167\n",
      "Epoch 4/20\n",
      "26/26 [==============================] - 16s 613ms/step - loss: 0.5348 - accuracy: 0.7417 - val_loss: 0.7397 - val_accuracy: 0.5917\n",
      "Epoch 5/20\n",
      "26/26 [==============================] - 16s 604ms/step - loss: 0.5355 - accuracy: 0.7423 - val_loss: 0.6270 - val_accuracy: 0.6444\n",
      "Epoch 6/20\n",
      "26/26 [==============================] - 16s 633ms/step - loss: 0.5436 - accuracy: 0.7340 - val_loss: 0.6589 - val_accuracy: 0.6000\n",
      "Epoch 7/20\n",
      "26/26 [==============================] - 16s 605ms/step - loss: 0.5385 - accuracy: 0.7429 - val_loss: 0.6758 - val_accuracy: 0.6361\n",
      "Epoch 8/20\n",
      "26/26 [==============================] - 16s 604ms/step - loss: 0.5264 - accuracy: 0.7564 - val_loss: 0.5759 - val_accuracy: 0.6056\n",
      "Epoch 9/20\n",
      "26/26 [==============================] - 16s 604ms/step - loss: 0.5438 - accuracy: 0.7410 - val_loss: 0.7238 - val_accuracy: 0.5889\n",
      "Epoch 10/20\n",
      "26/26 [==============================] - 18s 678ms/step - loss: 0.5471 - accuracy: 0.7346 - val_loss: 0.7292 - val_accuracy: 0.6111\n",
      "Epoch 11/20\n",
      "26/26 [==============================] - 15s 586ms/step - loss: 0.5329 - accuracy: 0.7468 - val_loss: 0.7496 - val_accuracy: 0.5667\n",
      "Epoch 12/20\n",
      "26/26 [==============================] - 16s 625ms/step - loss: 0.5318 - accuracy: 0.7410 - val_loss: 0.6679 - val_accuracy: 0.6583\n",
      "Epoch 13/20\n",
      "26/26 [==============================] - 16s 626ms/step - loss: 0.5423 - accuracy: 0.7391 - val_loss: 0.6853 - val_accuracy: 0.6000\n",
      "Epoch 14/20\n",
      "26/26 [==============================] - 20s 777ms/step - loss: 0.5448 - accuracy: 0.7353 - val_loss: 0.6133 - val_accuracy: 0.5861\n",
      "Epoch 15/20\n",
      "26/26 [==============================] - 30s 1s/step - loss: 0.5511 - accuracy: 0.7250 - val_loss: 0.7112 - val_accuracy: 0.5917\n",
      "Epoch 16/20\n",
      "26/26 [==============================] - 16s 597ms/step - loss: 0.5380 - accuracy: 0.7353 - val_loss: 0.6694 - val_accuracy: 0.6111\n",
      "Epoch 17/20\n",
      "26/26 [==============================] - 14s 557ms/step - loss: 0.5457 - accuracy: 0.7346 - val_loss: 0.7018 - val_accuracy: 0.6222\n",
      "Epoch 18/20\n",
      "26/26 [==============================] - 16s 599ms/step - loss: 0.5265 - accuracy: 0.7532 - val_loss: 0.6543 - val_accuracy: 0.6000\n",
      "Epoch 19/20\n",
      "26/26 [==============================] - 15s 587ms/step - loss: 0.5391 - accuracy: 0.7295 - val_loss: 0.6938 - val_accuracy: 0.5861\n",
      "Epoch 20/20\n",
      "26/26 [==============================] - 17s 640ms/step - loss: 0.5298 - accuracy: 0.7391 - val_loss: 0.6778 - val_accuracy: 0.6222\n"
     ]
    }
   ],
   "source": [
    "##### 5-FOLD CROSS-VALIDATION\n",
    "path=\"./rear_signal_dataset\"\n",
    "all_folder_list = train_folder_list + valid_folder_list + test_folder_list\n",
    "n_folder=len(all_folder_list)\n",
    "\n",
    "val1 = all_folder_list[0:math.floor(n_folder/5)]\n",
    "val2 = all_folder_list[math.floor(n_folder/5):(2*math.floor(n_folder/5))]\n",
    "val3 = all_folder_list[2*(math.floor(n_folder/5)):(3*math.floor(n_folder/5))]\n",
    "val4 = all_folder_list[3*(math.floor(n_folder/5)):(4*math.floor(n_folder/5))]\n",
    "val5 = all_folder_list[4*(math.floor(n_folder/5)):(n_folder)]\n",
    "\n",
    "print(len(val1))\n",
    "print(len(val2))\n",
    "print(len(val3))\n",
    "print(len(val4))\n",
    "print(len(val5))\n",
    "\n",
    "NUM_EPOCHS = 20\n",
    "\n",
    "validation_sets = np.array([val1, val2, val3,val4,val5])\n",
    "val_accuracies = []\n",
    "for validation_set in validation_sets:\n",
    "    train_set = [ele for ele in all_folder_list if ele not in valid_set]\n",
    "    # example #4 from:\n",
    "    # https://www.geeksforgeeks.org/remove-multiple-elements-from-a-list-in-python/ \n",
    "    \n",
    "    NUM_TRAIN_VALID = len(train_set)\n",
    "    NUM_VALID_VALID = len(valid_set)\n",
    "    \n",
    "    img_seq_gen_train_VAL = ImageSequenceGenerator()\n",
    "    img_seq_gen_valid_VAL = ImageSequenceGenerator()\n",
    "    \n",
    "    trainGenB_VAL = img_seq_gen_train_VAL.png_image_generator(path, bs=BS, difficulty=difficulty, folder_list=train_set, label_type = \"binary\", aug=None)\n",
    "    validGenB_VAL = img_seq_gen_valid_VAL.png_image_generator(path, bs=BS, difficulty=difficulty, folder_list=validation_set, label_type = \"binary\", aug=None)\n",
    "    \n",
    "    modelB.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])\n",
    "    print(\"[INFO] validation w/ generator...\")\n",
    "    historyB = modelB.fit(\n",
    "        x=trainGenB_VAL,\n",
    "        steps_per_epoch=NUM_TRAIN_VALID // BS,\n",
    "        validation_data=validGenB_VAL,\n",
    "        validation_steps=NUM_VALID_VALID // BS,\n",
    "        epochs=NUM_EPOCHS)\n",
    "    \n",
    "    val_accuracies.append(historyB.history['val_accuracy'][-1])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "for val in [val_accuracies]:\n",
    "    # modelB\n",
    "    # summarize history for validation accuracy\n",
    "    plt.xticks(np.arange(len(val_accuracies)), np.arange(1, len(val_accuracies)+1))\n",
    "    plt.plot(val)\n",
    "    plt.ylabel('val_accuracy')\n",
    "    plt.xlabel('fold')\n",
    "    plt.legend(['validation accuracy'], loc='upper left')\n",
    "    plt.savefig('images/fold_validation_accuracy_' + str(fold) + '.png',dpi=200)\n",
    "    plt.show()"
   ]
  }
 ],
 "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}