sound_classification_CNN.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# WORKING CODE WITH 3 MODELS"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* MODEL 1:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(5997, 130, 13)\n",
      "(5997, 130, 13, 1)\n",
      "(1500, 130, 13, 1)\n",
      "(2499, 130, 13, 1)\n",
      "Model: \"sequential_52\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_143 (Conv2D)          (None, 130, 13, 32)       320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_114 (MaxPoolin (None, 65, 7, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_117 (Bat (None, 65, 7, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_144 (Conv2D)          (None, 65, 7, 32)         9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_115 (MaxPoolin (None, 33, 4, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_118 (Bat (None, 33, 4, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_145 (Conv2D)          (None, 33, 4, 32)         4128      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_116 (MaxPoolin (None, 17, 2, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_119 (Bat (None, 17, 2, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_146 (Conv2D)          (None, 17, 2, 32)         4128      \n",
      "_________________________________________________________________\n",
      "batch_normalization_120 (Bat (None, 17, 2, 32)         128       \n",
      "_________________________________________________________________\n",
      "flatten_20 (Flatten)         (None, 1088)              0         \n",
      "_________________________________________________________________\n",
      "dense_64 (Dense)             (None, 128)               139392    \n",
      "_________________________________________________________________\n",
      "dropout_37 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_65 (Dense)             (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "dropout_38 (Dropout)         (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_66 (Dense)             (None, 10)                650       \n",
      "=================================================================\n",
      "Total params: 166,634\n",
      "Trainable params: 166,378\n",
      "Non-trainable params: 256\n",
      "_________________________________________________________________\n",
      "Train on 5997 samples, validate on 1500 samples\n",
      "Epoch 1/30\n",
      "5997/5997 [==============================] - 21s 4ms/sample - loss: 2.1081 - accuracy: 0.3643 - val_loss: 1.7616 - val_accuracy: 0.4867\n",
      "Epoch 2/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 1.7169 - accuracy: 0.4954 - val_loss: 1.4875 - val_accuracy: 0.5613\n",
      "Epoch 3/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 1.5194 - accuracy: 0.5603 - val_loss: 1.3307 - val_accuracy: 0.6027\n",
      "Epoch 4/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 1.3302 - accuracy: 0.6228 - val_loss: 1.2825 - val_accuracy: 0.6407\n",
      "Epoch 5/30\n",
      "5997/5997 [==============================] - 15s 3ms/sample - loss: 1.2209 - accuracy: 0.6532 - val_loss: 1.2167 - val_accuracy: 0.6387\n",
      "Epoch 6/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 1.1378 - accuracy: 0.6865 - val_loss: 1.1439 - val_accuracy: 0.6827\n",
      "Epoch 7/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 1.0462 - accuracy: 0.7072 - val_loss: 1.2488 - val_accuracy: 0.6567\n",
      "Epoch 8/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.9889 - accuracy: 0.7279 - val_loss: 1.1051 - val_accuracy: 0.7053\n",
      "Epoch 9/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.9090 - accuracy: 0.7549 - val_loss: 1.0312 - val_accuracy: 0.7127\n",
      "Epoch 10/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.8695 - accuracy: 0.7692 - val_loss: 0.9359 - val_accuracy: 0.7507\n",
      "Epoch 11/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.8068 - accuracy: 0.7944 - val_loss: 0.9674 - val_accuracy: 0.7553\n",
      "Epoch 12/30\n",
      "5997/5997 [==============================] - 15s 3ms/sample - loss: 0.7760 - accuracy: 0.8007 - val_loss: 1.0809 - val_accuracy: 0.7107\n",
      "Epoch 13/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.7495 - accuracy: 0.8184 - val_loss: 1.1264 - val_accuracy: 0.6960\n",
      "Epoch 14/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.6974 - accuracy: 0.8306 - val_loss: 1.2280 - val_accuracy: 0.7013\n",
      "Epoch 15/30\n",
      "5997/5997 [==============================] - 15s 3ms/sample - loss: 0.6778 - accuracy: 0.8361 - val_loss: 1.1703 - val_accuracy: 0.7100\n",
      "Epoch 16/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.6569 - accuracy: 0.8403 - val_loss: 1.1382 - val_accuracy: 0.7120\n",
      "Epoch 17/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.6177 - accuracy: 0.8578 - val_loss: 1.0286 - val_accuracy: 0.7493\n",
      "Epoch 18/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.5744 - accuracy: 0.8769 - val_loss: 1.1999 - val_accuracy: 0.7233\n",
      "Epoch 19/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.5917 - accuracy: 0.8723 - val_loss: 1.3091 - val_accuracy: 0.6833\n",
      "Epoch 20/30\n",
      "5997/5997 [==============================] - 15s 3ms/sample - loss: 0.5877 - accuracy: 0.8754 - val_loss: 1.4691 - val_accuracy: 0.6553\n",
      "Epoch 21/30\n",
      "5997/5997 [==============================] - 15s 3ms/sample - loss: 0.5357 - accuracy: 0.8886 - val_loss: 1.1491 - val_accuracy: 0.7400\n",
      "Epoch 22/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.5426 - accuracy: 0.8918 - val_loss: 1.3523 - val_accuracy: 0.7033\n",
      "Epoch 23/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.5185 - accuracy: 0.8976 - val_loss: 1.1987 - val_accuracy: 0.7287\n",
      "Epoch 24/30\n",
      "5997/5997 [==============================] - 15s 3ms/sample - loss: 0.5038 - accuracy: 0.9046 - val_loss: 1.2771 - val_accuracy: 0.7160\n",
      "Epoch 25/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.5082 - accuracy: 0.9080 - val_loss: 1.3631 - val_accuracy: 0.7027\n",
      "Epoch 26/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.4881 - accuracy: 0.9161 - val_loss: 1.1378 - val_accuracy: 0.7553\n",
      "Epoch 27/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.4644 - accuracy: 0.9196 - val_loss: 1.1227 - val_accuracy: 0.7340\n",
      "Epoch 28/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.4682 - accuracy: 0.9205 - val_loss: 1.0709 - val_accuracy: 0.7673\n",
      "Epoch 29/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.4446 - accuracy: 0.9280 - val_loss: 1.1318 - val_accuracy: 0.7660\n",
      "Epoch 30/30\n",
      "5997/5997 [==============================] - 15s 2ms/sample - loss: 0.4573 - accuracy: 0.9275 - val_loss: 1.1209 - val_accuracy: 0.7647\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2499/2499 [==============================] - 1s 496us/sample - loss: 1.1271 - accuracy: 0.7643\n",
      "Accuracy on test set is 0.7643057107925415\n",
      "Target: 5, Predicted label: [5]\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "import tensorflow.keras as keras\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "DATASET_PATH = \"data.json\"\n",
    "\n",
    "def load_data(dataset_path):\n",
    "    with open(dataset_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "        \n",
    "        \n",
    "    # converting lists into numpy arrays\n",
    "    X = np.array(data[\"mfcc\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "    \n",
    "    return X, y\n",
    "\n",
    "def plot_history(history):\n",
    "    \n",
    "    fig, axis = plt.subplots(2)\n",
    "    \n",
    "    # create accuracy subplot\n",
    "    axis[0].plot(history.history[\"accuracy\"], label = \"train accuracy\" )\n",
    "    axis[0].plot(history.history[\"val_accuracy\"], label = \"test_accuracy\" )\n",
    "    axis[0].set_ylabel(\"Accuracy\")\n",
    "    axis[0].set_xlabel(\"Epochs\")\n",
    "    axis[0].legend(loc = \"lower right\")\n",
    "    axis[0].set_title(\"Acccuracy eval\")\n",
    "    \n",
    "    \n",
    "    # create error subplot\n",
    "    axis[1].plot(history.history[\"loss\"], label = \"train error\" )\n",
    "    axis[1].plot(history.history[\"val_loss\"], label = \"test error\" )\n",
    "    axis[1].set_ylabel(\"Error\")\n",
    "    axis[1].set_xlabel(\"Epochs\")\n",
    "    axis[1].legend(loc = \"upper right\")\n",
    "    axis[1].set_title(\"Error eval\")\n",
    "    \n",
    "    plt.show()\n",
    "\n",
    "def prepare_datasets(test_size, validation_size):\n",
    "    \n",
    "    # load dataset\n",
    "    X, y = load_data(DATASET_PATH)\n",
    "    \n",
    "    # create train/test split\n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = test_size)\n",
    "    \n",
    "    # create train/validataion split\n",
    "    X_train, X_validation, y_train, y_validation = train_test_split(X_train, y_train, test_size= validation_size)\n",
    "    \n",
    "    print(X_train.shape)\n",
    "    \n",
    "    # so in X_train we have 3d array --> (sample_size, mfcc_vector, mfcc_coefficient)\n",
    "    # so we have to introduce another dimension in it. and make it 4d array. \n",
    "    X_train = X_train[... , np.newaxis]\n",
    "    # ... 3dots mean keep the rest of the shape same.\n",
    "    X_validation = X_validation[..., np.newaxis]\n",
    "    X_test = X_test[..., np.newaxis]\n",
    "    \n",
    "    print(X_train.shape)\n",
    "    print(X_validation.shape)\n",
    "    print(X_test.shape)\n",
    "    \n",
    "    \n",
    "    return X_train, X_validation, X_test, y_train, y_validation, y_test\n",
    "    \n",
    "\n",
    "def build_model(input_shape):\n",
    "    \n",
    "    # create model\n",
    "    model = keras.Sequential()\n",
    "    \n",
    "    # 1st conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), padding='same', activation = \"relu\", input_shape = input_shape))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "#     model.add(keras.layers.MaxPooling2D(pool_size=(3, 3)))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "    \n",
    "    # 2nd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), padding='same', activation = \"relu\"))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "#     model.add(keras.layers.MaxPooling2D(pool_size=(3, 3)))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "    \n",
    "    # 3rd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (2, 2), padding='same', activation = \"relu\"))\n",
    "    model.add(keras.layers.MaxPooling2D((2, 2), strides=(2, 2), padding='same'))\n",
    "#     model.add(keras.layers.MaxPooling2D(pool_size=(2, 2)))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "    \n",
    "    # 4th conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (2, 2), padding='same', activation = \"relu\"))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # Flatten the output and feed it into dense layer\n",
    "    model.add(keras.layers.Flatten())\n",
    "    \n",
    "    # 1st dense layer\n",
    "    model.add(keras.layers.Dense(128, activation = \"relu\", kernel_regularizer=keras.regularizers.l2(0.001)))\n",
    "    model.add(keras.layers.Dropout(0.5))\n",
    "    \n",
    "    # 2nd dense layer\n",
    "    model.add(keras.layers.Dense(64, activation = \"relu\", kernel_regularizer=keras.regularizers.l2(0.001)))\n",
    "    model.add(keras.layers.Dropout(0.5))\n",
    "    \n",
    "    # output layer\n",
    "    model.add(keras.layers.Dense(10, activation = \"softmax\"))\n",
    "    \n",
    "    return model\n",
    "\n",
    "\n",
    "def predict(model, X, y):\n",
    "    \"\"\"Predict a single sample using the trained model\n",
    "\n",
    "    :param model: Trained classifier\n",
    "    :param X: Input data\n",
    "    :param y (int): Target\n",
    "    \"\"\"\n",
    "\n",
    "    # add a dimension to input data for sample - model.predict() expects a 4d array in this case\n",
    "    X = X[np.newaxis, ...] # array shape (1, 130, 13, 1)\n",
    "\n",
    "    # perform prediction\n",
    "    prediction = model.predict(X)\n",
    "\n",
    "    # get index with max value\n",
    "    predicted_index = np.argmax(prediction, axis=1)\n",
    "\n",
    "    print(\"Target: {}, Predicted label: {}\".format(y, predicted_index))\n",
    "\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    # Create train, validation and test set\n",
    "    X_train, X_validation, X_test, y_train, y_validation, y_test = prepare_datasets(0.25, 0.20)\n",
    "#     X_train, X_validation, X_test, y_train, y_validation, y_test = prepare_datasets(0.25, 0.2)\n",
    "\n",
    "    \n",
    "    input_shape = (X_train.shape[1], X_train.shape[2], 1)\n",
    "    \n",
    "    # build the CNN network architecture\n",
    "    model = build_model(input_shape)\n",
    "    \n",
    "    # compile the network\n",
    "    optimizer = keras.optimizers.Adam(learning_rate = 0.001)\n",
    "    model.compile(optimizer = optimizer,\n",
    "                  loss = \"sparse_categorical_crossentropy\",\n",
    "                  metrics = [\"accuracy\"]\n",
    "                 )\n",
    "    \n",
    "    model.summary()\n",
    "    \n",
    "    # train the CNN\n",
    "    history = model.fit(X_train,\n",
    "              y_train,\n",
    "              validation_data = (X_validation, y_validation),\n",
    "              batch_size = 32,\n",
    "              epochs = 30\n",
    "             )\n",
    "    \n",
    "    # plot accuracy/error for training and validation\n",
    "    plot_history(history)\n",
    "    \n",
    "    # evaluate the CNN on the test set\n",
    "    test_error, test_accuracy = model.evaluate(X_test, y_test, verbose = 1)\n",
    "    print(\"Accuracy on test set is {}\".format(test_accuracy))\n",
    "    \n",
    "    # pick a sample to predict from the test set\n",
    "    X_to_predict = X_test[50]\n",
    "    y_to_predict = y_test[50]\n",
    "\n",
    "    # predict sample\n",
    "    predict(model, X_to_predict, y_to_predict)\n",
    "    \n",
    "    \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ANOTHER EXAMPLE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* MODEL 2:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_12\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_38 (Conv2D)           (None, 128, 11, 32)       320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_26 (MaxPooling (None, 64, 6, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_26 (Batc (None, 64, 6, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_39 (Conv2D)           (None, 62, 4, 32)         9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_27 (MaxPooling (None, 31, 2, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_27 (Batc (None, 31, 2, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_40 (Conv2D)           (None, 30, 1, 32)         4128      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_28 (MaxPooling (None, 15, 1, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_28 (Batc (None, 15, 1, 32)         128       \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 480)               0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 64)                30784     \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 10)                650       \n",
      "=================================================================\n",
      "Total params: 45,514\n",
      "Trainable params: 45,322\n",
      "Non-trainable params: 192\n",
      "_________________________________________________________________\n",
      "Train on 5997 samples, validate on 1500 samples\n",
      "Epoch 1/30\n",
      "5997/5997 [==============================] - 20s 3ms/sample - loss: 2.4049 - accuracy: 0.2375 - val_loss: 1.8283 - val_accuracy: 0.3740\n",
      "Epoch 2/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 1.8124 - accuracy: 0.3780 - val_loss: 1.5092 - val_accuracy: 0.4660\n",
      "Epoch 3/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 1.5920 - accuracy: 0.4411 - val_loss: 1.3749 - val_accuracy: 0.5033\n",
      "Epoch 4/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 1.4851 - accuracy: 0.4732 - val_loss: 1.2701 - val_accuracy: 0.5260\n",
      "Epoch 5/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 1.3880 - accuracy: 0.5033 - val_loss: 1.2143 - val_accuracy: 0.5433\n",
      "Epoch 6/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 1.3185 - accuracy: 0.5303 - val_loss: 1.1669 - val_accuracy: 0.5567\n",
      "Epoch 7/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 1.2671 - accuracy: 0.5543 - val_loss: 1.1166 - val_accuracy: 0.5793\n",
      "Epoch 8/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 1.2007 - accuracy: 0.5773 - val_loss: 1.0785 - val_accuracy: 0.6007\n",
      "Epoch 9/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 1.1548 - accuracy: 0.5866 - val_loss: 1.0496 - val_accuracy: 0.6167\n",
      "Epoch 10/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 1.1200 - accuracy: 0.6061 - val_loss: 1.0283 - val_accuracy: 0.6240\n",
      "Epoch 11/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 1.0868 - accuracy: 0.6113 - val_loss: 1.0030 - val_accuracy: 0.6367\n",
      "Epoch 12/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 1.0483 - accuracy: 0.6241 - val_loss: 0.9803 - val_accuracy: 0.6380\n",
      "Epoch 13/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 1.0268 - accuracy: 0.6422 - val_loss: 0.9569 - val_accuracy: 0.6527\n",
      "Epoch 14/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 0.9970 - accuracy: 0.6528 - val_loss: 0.9430 - val_accuracy: 0.6540\n",
      "Epoch 15/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 0.9704 - accuracy: 0.6647 - val_loss: 0.9280 - val_accuracy: 0.6560\n",
      "Epoch 16/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 0.9554 - accuracy: 0.6672 - val_loss: 0.9039 - val_accuracy: 0.6673\n",
      "Epoch 17/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 0.9133 - accuracy: 0.6847 - val_loss: 0.8925 - val_accuracy: 0.6713\n",
      "Epoch 18/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 0.8911 - accuracy: 0.6905 - val_loss: 0.8757 - val_accuracy: 0.6827\n",
      "Epoch 19/30\n",
      "5997/5997 [==============================] - 12s 2ms/sample - loss: 0.8753 - accuracy: 0.6857 - val_loss: 0.8555 - val_accuracy: 0.6887\n",
      "Epoch 20/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 0.8395 - accuracy: 0.7109 - val_loss: 0.8421 - val_accuracy: 0.6953\n",
      "Epoch 21/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 0.8324 - accuracy: 0.7122 - val_loss: 0.8376 - val_accuracy: 0.7027\n",
      "Epoch 22/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 0.8228 - accuracy: 0.7142 - val_loss: 0.8510 - val_accuracy: 0.6927\n",
      "Epoch 23/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 0.8044 - accuracy: 0.7152 - val_loss: 0.8439 - val_accuracy: 0.7067\n",
      "Epoch 24/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 0.7801 - accuracy: 0.7275 - val_loss: 0.8110 - val_accuracy: 0.7160\n",
      "Epoch 25/30\n",
      "5997/5997 [==============================] - 12s 2ms/sample - loss: 0.7581 - accuracy: 0.7382 - val_loss: 0.8064 - val_accuracy: 0.7133\n",
      "Epoch 26/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 0.7396 - accuracy: 0.7479 - val_loss: 0.8238 - val_accuracy: 0.7100\n",
      "Epoch 27/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 0.7276 - accuracy: 0.7464 - val_loss: 0.7923 - val_accuracy: 0.7073\n",
      "Epoch 28/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 0.7324 - accuracy: 0.7409 - val_loss: 0.7823 - val_accuracy: 0.7180\n",
      "Epoch 29/30\n",
      "5997/5997 [==============================] - 11s 2ms/sample - loss: 0.7182 - accuracy: 0.7504 - val_loss: 0.7657 - val_accuracy: 0.7320\n",
      "Epoch 30/30\n",
      "5997/5997 [==============================] - 10s 2ms/sample - loss: 0.6923 - accuracy: 0.7555 - val_loss: 0.7609 - val_accuracy: 0.7333\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2499/2499 - 1s - loss: 0.8239 - accuracy: 0.7171\n",
      "\n",
      "Test accuracy: 0.71708685\n",
      "Target: 1, Predicted label: [1]\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "import tensorflow.keras as keras\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "DATA_PATH = \"data.json\"\n",
    "\n",
    "\n",
    "def load_data(data_path):\n",
    "    \"\"\"Loads training dataset from json file.\n",
    "\n",
    "        :param data_path (str): Path to json file containing data\n",
    "        :return X (ndarray): Inputs\n",
    "        :return y (ndarray): Targets\n",
    "    \"\"\"\n",
    "\n",
    "    with open(data_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "\n",
    "    X = np.array(data[\"mfcc\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "    return X, y\n",
    "\n",
    "\n",
    "def plot_history(history):\n",
    "    \"\"\"Plots accuracy/loss for training/validation set as a function of the epochs\n",
    "\n",
    "        :param history: Training history of model\n",
    "        :return:\n",
    "    \"\"\"\n",
    "\n",
    "    fig, axs = plt.subplots(2)\n",
    "\n",
    "    # create accuracy sublpot\n",
    "    axs[0].plot(history.history[\"accuracy\"], label=\"train accuracy\")\n",
    "    axs[0].plot(history.history[\"val_accuracy\"], label=\"test accuracy\")\n",
    "    axs[0].set_ylabel(\"Accuracy\")\n",
    "    axs[0].legend(loc=\"lower right\")\n",
    "    axs[0].set_title(\"Accuracy eval\")\n",
    "\n",
    "    # create error sublpot\n",
    "    axs[1].plot(history.history[\"loss\"], label=\"train error\")\n",
    "    axs[1].plot(history.history[\"val_loss\"], label=\"test error\")\n",
    "    axs[1].set_ylabel(\"Error\")\n",
    "    axs[1].set_xlabel(\"Epoch\")\n",
    "    axs[1].legend(loc=\"upper right\")\n",
    "    axs[1].set_title(\"Error eval\")\n",
    "\n",
    "    plt.show()\n",
    "\n",
    "\n",
    "def prepare_datasets(test_size, validation_size):\n",
    "    \"\"\"Loads data and splits it into train, validation and test sets.\n",
    "\n",
    "    :param test_size (float): Value in [0, 1] indicating percentage of data set to allocate to test split\n",
    "    :param validation_size (float): Value in [0, 1] indicating percentage of train set to allocate to validation split\n",
    "\n",
    "    :return X_train (ndarray): Input training set\n",
    "    :return X_validation (ndarray): Input validation set\n",
    "    :return X_test (ndarray): Input test set\n",
    "    :return y_train (ndarray): Target training set\n",
    "    :return y_validation (ndarray): Target validation set\n",
    "    :return y_test (ndarray): Target test set\n",
    "    \"\"\"\n",
    "\n",
    "    # load data\n",
    "    X, y = load_data(DATA_PATH)\n",
    "\n",
    "    # create train, validation and test split\n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)\n",
    "    X_train, X_validation, y_train, y_validation = train_test_split(X_train, y_train, test_size=validation_size)\n",
    "\n",
    "    # add an axis to input sets\n",
    "    X_train = X_train[..., np.newaxis]\n",
    "    X_validation = X_validation[..., np.newaxis]\n",
    "    X_test = X_test[..., np.newaxis]\n",
    "\n",
    "    return X_train, X_validation, X_test, y_train, y_validation, y_test\n",
    "\n",
    "\n",
    "def build_model(input_shape):\n",
    "    \"\"\"Generates CNN model\n",
    "\n",
    "    :param input_shape (tuple): Shape of input set\n",
    "    :return model: CNN model\n",
    "    \"\"\"\n",
    "\n",
    "    # build network topology\n",
    "    model = keras.Sequential()\n",
    "\n",
    "    # 1st conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=input_shape))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 2nd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 3rd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (2, 2), activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((2, 2), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # flatten output and feed it into dense layer\n",
    "    model.add(keras.layers.Flatten())\n",
    "    model.add(keras.layers.Dense(64, activation='relu'))\n",
    "    model.add(keras.layers.Dropout(0.3))\n",
    "\n",
    "    # output layer\n",
    "    model.add(keras.layers.Dense(10, activation='softmax'))\n",
    "\n",
    "    return model\n",
    "\n",
    "\n",
    "def predict(model, X, y):\n",
    "    \"\"\"Predict a single sample using the trained model\n",
    "\n",
    "    :param model: Trained classifier\n",
    "    :param X: Input data\n",
    "    :param y (int): Target\n",
    "    \"\"\"\n",
    "\n",
    "    # add a dimension to input data for sample - model.predict() expects a 4d array in this case\n",
    "    X = X[np.newaxis, ...] # array shape (1, 130, 13, 1)\n",
    "\n",
    "    # perform prediction\n",
    "    prediction = model.predict(X)\n",
    "\n",
    "    # get index with max value\n",
    "    predicted_index = np.argmax(prediction, axis=1)\n",
    "\n",
    "    print(\"Target: {}, Predicted label: {}\".format(y, predicted_index))\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    # get train, validation, test splits\n",
    "    X_train, X_validation, X_test, y_train, y_validation, y_test = prepare_datasets(0.25, 0.2)\n",
    "\n",
    "    # create network\n",
    "    input_shape = (X_train.shape[1], X_train.shape[2], 1)\n",
    "    model = build_model(input_shape)\n",
    "\n",
    "    # compile model\n",
    "    optimizer = keras.optimizers.Adam(learning_rate=0.0001)\n",
    "    model.compile(optimizer=optimizer,\n",
    "                  loss='sparse_categorical_crossentropy',\n",
    "                  metrics=['accuracy'])\n",
    "    \n",
    "\n",
    "    model.summary()\n",
    "\n",
    "    # train model\n",
    "    history = model.fit(X_train, y_train, validation_data=(X_validation, y_validation), batch_size=32, epochs=30)\n",
    "\n",
    "    # plot accuracy/error for training and validation\n",
    "    plot_history(history)\n",
    "\n",
    "    # evaluate model on test set\n",
    "    test_loss, test_acc = model.evaluate(X_test, y_test, verbose=2)\n",
    "    print('\\nTest accuracy:', test_acc)\n",
    "\n",
    "    # pick a sample to predict from the test set\n",
    "    X_to_predict = X_test[100]\n",
    "    y_to_predict = y_test[100]\n",
    "\n",
    "    # predict sample\n",
    "    predict(model, X_to_predict, y_to_predict)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# EXPERIMENT 3"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* MODEL 3:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_42\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_129 (Conv2D)          (None, 106, 11, 32)       320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_104 (MaxPoolin (None, 53, 6, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_104 (Bat (None, 53, 6, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_130 (Conv2D)          (None, 51, 4, 32)         9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_105 (MaxPoolin (None, 26, 2, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_105 (Bat (None, 26, 2, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_131 (Conv2D)          (None, 25, 1, 32)         4128      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_106 (MaxPoolin (None, 13, 1, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_106 (Bat (None, 13, 1, 32)         128       \n",
      "_________________________________________________________________\n",
      "flatten_17 (Flatten)         (None, 416)               0         \n",
      "_________________________________________________________________\n",
      "dense_37 (Dense)             (None, 64)                26688     \n",
      "_________________________________________________________________\n",
      "dropout_20 (Dropout)         (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_38 (Dense)             (None, 10)                650       \n",
      "=================================================================\n",
      "Total params: 41,418\n",
      "Trainable params: 41,226\n",
      "Non-trainable params: 192\n",
      "_________________________________________________________________\n",
      "Train on 4909 samples, validate on 1228 samples\n",
      "Epoch 1/30\n",
      "4909/4909 [==============================] - 16s 3ms/sample - loss: 2.2171 - accuracy: 0.2931 - val_loss: 1.4131 - val_accuracy: 0.5945\n",
      "Epoch 2/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 1.2258 - accuracy: 0.5842 - val_loss: 0.8603 - val_accuracy: 0.6946\n",
      "Epoch 3/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.9127 - accuracy: 0.6580 - val_loss: 0.7102 - val_accuracy: 0.6995\n",
      "Epoch 4/30\n",
      "4909/4909 [==============================] - 8s 2ms/sample - loss: 0.8162 - accuracy: 0.6712 - val_loss: 0.6661 - val_accuracy: 0.7125\n",
      "Epoch 5/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.7280 - accuracy: 0.6826 - val_loss: 0.6328 - val_accuracy: 0.7085\n",
      "Epoch 6/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.6874 - accuracy: 0.7063 - val_loss: 0.6086 - val_accuracy: 0.7182\n",
      "Epoch 7/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.6413 - accuracy: 0.7238 - val_loss: 0.5984 - val_accuracy: 0.7158\n",
      "Epoch 8/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.6237 - accuracy: 0.7189 - val_loss: 0.5862 - val_accuracy: 0.7182\n",
      "Epoch 9/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.6004 - accuracy: 0.7221 - val_loss: 0.5830 - val_accuracy: 0.7191\n",
      "Epoch 10/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.5743 - accuracy: 0.7397 - val_loss: 0.5829 - val_accuracy: 0.7256\n",
      "Epoch 11/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.5493 - accuracy: 0.7488 - val_loss: 0.5642 - val_accuracy: 0.7296\n",
      "Epoch 12/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.5368 - accuracy: 0.7517 - val_loss: 0.5673 - val_accuracy: 0.7296\n",
      "Epoch 13/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.5231 - accuracy: 0.7568 - val_loss: 0.5583 - val_accuracy: 0.7362\n",
      "Epoch 14/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.5112 - accuracy: 0.7670 - val_loss: 0.5514 - val_accuracy: 0.7378\n",
      "Epoch 15/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4988 - accuracy: 0.7737 - val_loss: 0.5556 - val_accuracy: 0.7378\n",
      "Epoch 16/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4819 - accuracy: 0.7761 - val_loss: 0.5466 - val_accuracy: 0.7337\n",
      "Epoch 17/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4650 - accuracy: 0.7916 - val_loss: 0.5506 - val_accuracy: 0.7443\n",
      "Epoch 18/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4607 - accuracy: 0.7875 - val_loss: 0.5638 - val_accuracy: 0.7435\n",
      "Epoch 19/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4467 - accuracy: 0.7973 - val_loss: 0.5415 - val_accuracy: 0.7394\n",
      "Epoch 20/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4417 - accuracy: 0.7973 - val_loss: 0.5507 - val_accuracy: 0.7386\n",
      "Epoch 21/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4339 - accuracy: 0.8014 - val_loss: 0.5608 - val_accuracy: 0.7410\n",
      "Epoch 22/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4318 - accuracy: 0.8057 - val_loss: 0.5550 - val_accuracy: 0.7435\n",
      "Epoch 23/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4252 - accuracy: 0.8095 - val_loss: 0.5452 - val_accuracy: 0.7378\n",
      "Epoch 24/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4086 - accuracy: 0.8207 - val_loss: 0.5492 - val_accuracy: 0.7451\n",
      "Epoch 25/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.4050 - accuracy: 0.8209 - val_loss: 0.5447 - val_accuracy: 0.7419\n",
      "Epoch 26/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.3977 - accuracy: 0.8242 - val_loss: 0.5520 - val_accuracy: 0.7500\n",
      "Epoch 27/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.3904 - accuracy: 0.8248 - val_loss: 0.5466 - val_accuracy: 0.7533\n",
      "Epoch 28/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.3832 - accuracy: 0.8328 - val_loss: 0.5690 - val_accuracy: 0.7329\n",
      "Epoch 29/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.3762 - accuracy: 0.8344 - val_loss: 0.5436 - val_accuracy: 0.7508\n",
      "Epoch 30/30\n",
      "4909/4909 [==============================] - 7s 1ms/sample - loss: 0.3687 - accuracy: 0.8381 - val_loss: 0.5448 - val_accuracy: 0.7467\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2046/2046 - 1s - loss: 0.5229 - accuracy: 0.7512\n",
      "\n",
      "Test accuracy: 0.7512219\n",
      "Target: 0, Predicted label: [0]\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "import tensorflow.keras as keras\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "DATA_PATH = \"sick_sound.json\"\n",
    "\n",
    "\n",
    "def load_data(data_path):\n",
    "    \"\"\"Loads training dataset from json file.\n",
    "\n",
    "        :param data_path (str): Path to json file containing data\n",
    "        :return X (ndarray): Inputs\n",
    "        :return y (ndarray): Targets\n",
    "    \"\"\"\n",
    "\n",
    "    with open(data_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "\n",
    "    X = np.array(data[\"mfcc\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "    return X, y\n",
    "\n",
    "\n",
    "def plot_history(history):\n",
    "    \"\"\"Plots accuracy/loss for training/validation set as a function of the epochs\n",
    "\n",
    "        :param history: Training history of model\n",
    "        :return:\n",
    "    \"\"\"\n",
    "\n",
    "    fig, axs = plt.subplots(2)\n",
    "\n",
    "    # create accuracy sublpot\n",
    "    axs[0].plot(history.history[\"accuracy\"], label=\"train accuracy\")\n",
    "    axs[0].plot(history.history[\"val_accuracy\"], label=\"test accuracy\")\n",
    "    axs[0].set_ylabel(\"Accuracy\")\n",
    "    axs[0].legend(loc=\"lower right\")\n",
    "    axs[0].set_title(\"Accuracy eval\")\n",
    "\n",
    "    # create error sublpot\n",
    "    axs[1].plot(history.history[\"loss\"], label=\"train error\")\n",
    "    axs[1].plot(history.history[\"val_loss\"], label=\"test error\")\n",
    "    axs[1].set_ylabel(\"Error\")\n",
    "    axs[1].set_xlabel(\"Epoch\")\n",
    "    axs[1].legend(loc=\"upper right\")\n",
    "    axs[1].set_title(\"Error eval\")\n",
    "\n",
    "    plt.show()\n",
    "\n",
    "\n",
    "def prepare_datasets(test_size, validation_size):\n",
    "    \"\"\"Loads data and splits it into train, validation and test sets.\n",
    "\n",
    "    :param test_size (float): Value in [0, 1] indicating percentage of data set to allocate to test split\n",
    "    :param validation_size (float): Value in [0, 1] indicating percentage of train set to allocate to validation split\n",
    "\n",
    "    :return X_train (ndarray): Input training set\n",
    "    :return X_validation (ndarray): Input validation set\n",
    "    :return X_test (ndarray): Input test set\n",
    "    :return y_train (ndarray): Target training set\n",
    "    :return y_validation (ndarray): Target validation set\n",
    "    :return y_test (ndarray): Target test set\n",
    "    \"\"\"\n",
    "\n",
    "    # load data\n",
    "    X, y = load_data(DATA_PATH)\n",
    "\n",
    "    # create train, validation and test split\n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)\n",
    "    X_train, X_validation, y_train, y_validation = train_test_split(X_train, y_train, test_size=validation_size)\n",
    "\n",
    "    # add an axis to input sets\n",
    "    X_train = X_train[..., np.newaxis]\n",
    "    X_validation = X_validation[..., np.newaxis]\n",
    "    X_test = X_test[..., np.newaxis]\n",
    "\n",
    "    return X_train, X_validation, X_test, y_train, y_validation, y_test\n",
    "\n",
    "\n",
    "def build_model(input_shape):\n",
    "    \"\"\"Generates CNN model\n",
    "\n",
    "    :param input_shape (tuple): Shape of input set\n",
    "    :return model: CNN model\n",
    "    \"\"\"\n",
    "\n",
    "    # build network topology\n",
    "    model = keras.Sequential()\n",
    "\n",
    "    # 1st conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=input_shape))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 2nd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 3rd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (2, 2), activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((2, 2), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # flatten output and feed it into dense layer\n",
    "    model.add(keras.layers.Flatten())\n",
    "    model.add(keras.layers.Dense(64, activation='relu'))\n",
    "    model.add(keras.layers.Dropout(0.3))\n",
    "\n",
    "    # output layer\n",
    "    model.add(keras.layers.Dense(2, activation='softmax'))\n",
    "\n",
    "    return model\n",
    "\n",
    "\n",
    "def predict(model, X, y):\n",
    "    \"\"\"Predict a single sample using the trained model\n",
    "\n",
    "    :param model: Trained classifier\n",
    "    :param X: Input data\n",
    "    :param y (int): Target\n",
    "    \"\"\"\n",
    "\n",
    "    # add a dimension to input data for sample - model.predict() expects a 4d array in this case\n",
    "    X = X[np.newaxis, ...] # array shape (1, 130, 13, 1)\n",
    "\n",
    "    # perform prediction\n",
    "    prediction = model.predict(X)\n",
    "\n",
    "    # get index with max value\n",
    "    predicted_index = np.argmax(prediction, axis=1)\n",
    "\n",
    "    print(\"Target: {}, Predicted label: {}\".format(y, predicted_index))\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    # get train, validation, test splits\n",
    "    X_train, X_validation, X_test, y_train, y_validation, y_test = prepare_datasets(0.25, 0.2)\n",
    "\n",
    "    # create network\n",
    "    input_shape = (X_train.shape[1], X_train.shape[2], 1)\n",
    "    model = build_model(input_shape)\n",
    "\n",
    "    # compile model\n",
    "    optimizer = keras.optimizers.Adam(learning_rate=0.0001)\n",
    "    model.compile(optimizer=optimizer,\n",
    "                  loss='sparse_categorical_crossentropy',\n",
    "                  metrics=['accuracy'])\n",
    "    \n",
    "\n",
    "    model.summary()\n",
    "\n",
    "    # train model\n",
    "    history = model.fit(X_train, y_train, validation_data=(X_validation, y_validation), batch_size=32, epochs=30)\n",
    "\n",
    "    # plot accuracy/error for training and validation\n",
    "    plot_history(history)\n",
    "\n",
    "    # evaluate model on test set\n",
    "    test_loss, test_acc = model.evaluate(X_test, y_test, verbose=2)\n",
    "    print('\\nTest accuracy:', test_acc)\n",
    "\n",
    "    # pick a sample to predict from the test set\n",
    "    X_to_predict = X_test[0]\n",
    "    y_to_predict = y_test[0]\n",
    "\n",
    "    # predict sample\n",
    "    predict(model, X_to_predict, y_to_predict)\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### PLOTING CONFUSION MATRIX"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ 126 1216]\n",
      " [  28  676]]\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "yhat_probs = model.predict(X_test, verbose=0)\n",
    "tt = np.argmax(yhat_probs, axis = 1)\n",
    "\n",
    "cm=confusion_matrix(y_test, tt)\n",
    "print(cm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Experiment : 4 \n",
    "* MODEL 4 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_64\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_191 (Conv2D)          (None, 108, 13, 64)       640       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_139 (MaxPoolin (None, 54, 7, 64)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_165 (Bat (None, 54, 7, 64)         256       \n",
      "_________________________________________________________________\n",
      "conv2d_192 (Conv2D)          (None, 54, 7, 64)         36928     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_140 (MaxPoolin (None, 27, 4, 64)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_166 (Bat (None, 27, 4, 64)         256       \n",
      "_________________________________________________________________\n",
      "conv2d_193 (Conv2D)          (None, 27, 4, 32)         8224      \n",
      "_________________________________________________________________\n",
      "batch_normalization_167 (Bat (None, 27, 4, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_194 (Conv2D)          (None, 27, 4, 32)         4128      \n",
      "_________________________________________________________________\n",
      "batch_normalization_168 (Bat (None, 27, 4, 32)         128       \n",
      "_________________________________________________________________\n",
      "flatten_32 (Flatten)         (None, 3456)              0         \n",
      "_________________________________________________________________\n",
      "dense_100 (Dense)            (None, 256)               884992    \n",
      "_________________________________________________________________\n",
      "dropout_61 (Dropout)         (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_101 (Dense)            (None, 64)                16448     \n",
      "_________________________________________________________________\n",
      "dropout_62 (Dropout)         (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_102 (Dense)            (None, 2)                 130       \n",
      "=================================================================\n",
      "Total params: 952,258\n",
      "Trainable params: 951,874\n",
      "Non-trainable params: 384\n",
      "_________________________________________________________________\n",
      "Train on 4909 samples, validate on 1228 samples\n",
      "Epoch 1/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 1.2984 - accuracy: 0.6145\n",
      "Epoch 00001: val_loss improved from inf to 1.16911, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 25s 5ms/sample - loss: 1.2978 - accuracy: 0.6136 - val_loss: 1.1691 - val_accuracy: 0.6531\n",
      "Epoch 2/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 1.0965 - accuracy: 0.6495\n",
      "Epoch 00002: val_loss improved from 1.16911 to 1.04541, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 25s 5ms/sample - loss: 1.0967 - accuracy: 0.6486 - val_loss: 1.0454 - val_accuracy: 0.6743\n",
      "Epoch 3/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.9988 - accuracy: 0.6721\n",
      "Epoch 00003: val_loss improved from 1.04541 to 0.95557, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.9984 - accuracy: 0.6714 - val_loss: 0.9556 - val_accuracy: 0.6832\n",
      "Epoch 4/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.9041 - accuracy: 0.6875\n",
      "Epoch 00004: val_loss improved from 0.95557 to 0.86850, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 25s 5ms/sample - loss: 0.9041 - accuracy: 0.6879 - val_loss: 0.8685 - val_accuracy: 0.7223\n",
      "Epoch 5/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.8312 - accuracy: 0.7116\n",
      "Epoch 00005: val_loss improved from 0.86850 to 0.79460, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.8308 - accuracy: 0.7120 - val_loss: 0.7946 - val_accuracy: 0.7117\n",
      "Epoch 6/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.7601 - accuracy: 0.7354\n",
      "Epoch 00006: val_loss did not improve from 0.79460\n",
      "4909/4909 [==============================] - 24s 5ms/sample - loss: 0.7593 - accuracy: 0.7358 - val_loss: 0.7991 - val_accuracy: 0.6800\n",
      "Epoch 7/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.7048 - accuracy: 0.7506\n",
      "Epoch 00007: val_loss improved from 0.79460 to 0.66314, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 28s 6ms/sample - loss: 0.7040 - accuracy: 0.7515 - val_loss: 0.6631 - val_accuracy: 0.7809\n",
      "Epoch 8/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.6572 - accuracy: 0.7738\n",
      "Epoch 00008: val_loss improved from 0.66314 to 0.64370, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.6565 - accuracy: 0.7743 - val_loss: 0.6437 - val_accuracy: 0.7679\n",
      "Epoch 9/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.6131 - accuracy: 0.7876\n",
      "Epoch 00009: val_loss did not improve from 0.64370\n",
      "4909/4909 [==============================] - 27s 5ms/sample - loss: 0.6136 - accuracy: 0.7867 - val_loss: 0.7001 - val_accuracy: 0.7191\n",
      "Epoch 10/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.5895 - accuracy: 0.7932\n",
      "Epoch 00010: val_loss did not improve from 0.64370\n",
      "4909/4909 [==============================] - 27s 6ms/sample - loss: 0.5910 - accuracy: 0.7926 - val_loss: 0.6712 - val_accuracy: 0.7476\n",
      "Epoch 11/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.5726 - accuracy: 0.7934\n",
      "Epoch 00011: val_loss improved from 0.64370 to 0.61196, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.5724 - accuracy: 0.7938 - val_loss: 0.6120 - val_accuracy: 0.7736\n",
      "Epoch 12/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.5377 - accuracy: 0.8088\n",
      "Epoch 00012: val_loss did not improve from 0.61196\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.5395 - accuracy: 0.8077 - val_loss: 0.6427 - val_accuracy: 0.7630\n",
      "Epoch 13/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.5397 - accuracy: 0.8078\n",
      "Epoch 00013: val_loss improved from 0.61196 to 0.60048, saving model to sickweight4.h5\n",
      "4909/4909 [==============================] - 25s 5ms/sample - loss: 0.5389 - accuracy: 0.8083 - val_loss: 0.6005 - val_accuracy: 0.7752\n",
      "Epoch 14/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.5177 - accuracy: 0.8263\n",
      "Epoch 00014: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.5166 - accuracy: 0.8268 - val_loss: 0.6170 - val_accuracy: 0.7923\n",
      "Epoch 15/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.4891 - accuracy: 0.8361\n",
      "Epoch 00015: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 25s 5ms/sample - loss: 0.4895 - accuracy: 0.8364 - val_loss: 0.6680 - val_accuracy: 0.7761\n",
      "Epoch 16/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.4929 - accuracy: 0.8417\n",
      "Epoch 00016: ReduceLROnPlateau reducing learning rate to 0.00010000000474974513.\n",
      "\n",
      "Epoch 00016: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 24s 5ms/sample - loss: 0.4928 - accuracy: 0.8417 - val_loss: 0.6877 - val_accuracy: 0.7427\n",
      "Epoch 17/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.4442 - accuracy: 0.8740\n",
      "Epoch 00017: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.4437 - accuracy: 0.8741 - val_loss: 0.6252 - val_accuracy: 0.8013\n",
      "Epoch 18/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3930 - accuracy: 0.8892\n",
      "Epoch 00018: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 25s 5ms/sample - loss: 0.3932 - accuracy: 0.8890 - val_loss: 0.6404 - val_accuracy: 0.8021\n",
      "Epoch 19/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3654 - accuracy: 0.9050\n",
      "Epoch 00019: ReduceLROnPlateau reducing learning rate to 1.0000000474974514e-05.\n",
      "\n",
      "Epoch 00019: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 24s 5ms/sample - loss: 0.3650 - accuracy: 0.9053 - val_loss: 0.6615 - val_accuracy: 0.7972\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 20/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3414 - accuracy: 0.9194\n",
      "Epoch 00020: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 24s 5ms/sample - loss: 0.3417 - accuracy: 0.9189 - val_loss: 0.6661 - val_accuracy: 0.8005\n",
      "Epoch 21/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3435 - accuracy: 0.9147\n",
      "Epoch 00021: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.3438 - accuracy: 0.9149 - val_loss: 0.6717 - val_accuracy: 0.7989\n",
      "Epoch 22/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3396 - accuracy: 0.9180\n",
      "Epoch 00022: ReduceLROnPlateau reducing learning rate to 1.0000000656873453e-06.\n",
      "\n",
      "Epoch 00022: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 27s 5ms/sample - loss: 0.3390 - accuracy: 0.9185 - val_loss: 0.6755 - val_accuracy: 0.7980\n",
      "Epoch 23/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3363 - accuracy: 0.9186\n",
      "Epoch 00023: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 32s 6ms/sample - loss: 0.3362 - accuracy: 0.9187 - val_loss: 0.6769 - val_accuracy: 0.7980\n",
      "Epoch 24/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3392 - accuracy: 0.9159\n",
      "Epoch 00024: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.3389 - accuracy: 0.9161 - val_loss: 0.6779 - val_accuracy: 0.7980\n",
      "Epoch 25/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3375 - accuracy: 0.9176\n",
      "Epoch 00025: ReduceLROnPlateau reducing learning rate to 1.0000001111620805e-07.\n",
      "\n",
      "Epoch 00025: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.3373 - accuracy: 0.9173 - val_loss: 0.6782 - val_accuracy: 0.7972\n",
      "Epoch 26/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3341 - accuracy: 0.9192\n",
      "Epoch 00026: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.3337 - accuracy: 0.9193 - val_loss: 0.6787 - val_accuracy: 0.7980\n",
      "Epoch 27/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3340 - accuracy: 0.9190\n",
      "Epoch 00027: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 27s 6ms/sample - loss: 0.3336 - accuracy: 0.9187 - val_loss: 0.6788 - val_accuracy: 0.7989\n",
      "Epoch 28/50\n",
      "4864/4909 [============================>.] - ETA: 0s - loss: 0.3388 - accuracy: 0.9192\n",
      "Epoch 00028: ReduceLROnPlateau reducing learning rate to 1.000000082740371e-08.\n",
      "Restoring model weights from the end of the best epoch.\n",
      "\n",
      "Epoch 00028: val_loss did not improve from 0.60048\n",
      "4909/4909 [==============================] - 26s 5ms/sample - loss: 0.3379 - accuracy: 0.9197 - val_loss: 0.6789 - val_accuracy: 0.7989\n",
      "Epoch 00028: early stopping\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2046/2046 - 2s - loss: 0.6003 - accuracy: 0.7805\n",
      "\n",
      "Test accuracy: 0.7805474\n",
      "Target: 0, Predicted label: [0]\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "# import tensorflow.keras as keras\n",
    "import tensorflow as tf\n",
    "import tensorflow_hub as hub\n",
    "import matplotlib.pylab as plt\n",
    "\n",
    "DATA_PATH = \"sick_sound.json\"\n",
    "\n",
    "\n",
    "def load_data(data_path):\n",
    "    \"\"\"Loads training dataset from json file.\n",
    "\n",
    "        :param data_path (str): Path to json file containing data\n",
    "        :return X (ndarray): Inputs\n",
    "        :return y (ndarray): Targets\n",
    "    \"\"\"\n",
    "\n",
    "    with open(data_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "\n",
    "    X = np.array(data[\"mfcc\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "    return X, y\n",
    "\n",
    "\n",
    "def plot_history(history):\n",
    "    \"\"\"Plots accuracy/loss for training/validation set as a function of the epochs\n",
    "\n",
    "        :param history: Training history of model\n",
    "        :return:\n",
    "    \"\"\"\n",
    "\n",
    "    fig, axs = plt.subplots(2)\n",
    "\n",
    "    # create accuracy sublpot\n",
    "    axs[0].plot(history.history[\"accuracy\"], label=\"train accuracy\")\n",
    "    axs[0].plot(history.history[\"val_accuracy\"], label=\"test accuracy\")\n",
    "    axs[0].set_ylabel(\"Accuracy\")\n",
    "    axs[0].legend(loc=\"lower right\")\n",
    "    axs[0].set_title(\"Accuracy eval\")\n",
    "\n",
    "    # create error sublpot\n",
    "    axs[1].plot(history.history[\"loss\"], label=\"train error\")\n",
    "    axs[1].plot(history.history[\"val_loss\"], label=\"test error\")\n",
    "    axs[1].set_ylabel(\"Error\")\n",
    "    axs[1].set_xlabel(\"Epoch\")\n",
    "    axs[1].legend(loc=\"upper right\")\n",
    "    axs[1].set_title(\"Error eval\")\n",
    "\n",
    "    plt.show()\n",
    "\n",
    "\n",
    "def prepare_datasets(test_size, validation_size):\n",
    "    \"\"\"Loads data and splits it into train, validation and test sets.\n",
    "\n",
    "    :param test_size (float): Value in [0, 1] indicating percentage of data set to allocate to test split\n",
    "    :param validation_size (float): Value in [0, 1] indicating percentage of train set to allocate to validation split\n",
    "\n",
    "    :return X_train (ndarray): Input training set\n",
    "    :return X_validation (ndarray): Input validation set\n",
    "    :return X_test (ndarray): Input test set\n",
    "    :return y_train (ndarray): Target training set\n",
    "    :return y_validation (ndarray): Target validation set\n",
    "    :return y_test (ndarray): Target test set\n",
    "    \"\"\"\n",
    "\n",
    "    # load data\n",
    "    X, y = load_data(DATA_PATH)\n",
    "\n",
    "    # create train, validation and test split\n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)\n",
    "    X_train, X_validation, y_train, y_validation = train_test_split(X_train, y_train, test_size=validation_size)\n",
    "\n",
    "    # add an axis to input sets\n",
    "    X_train = X_train[..., np.newaxis]\n",
    "    X_validation = X_validation[..., np.newaxis]\n",
    "    X_test = X_test[..., np.newaxis]\n",
    "\n",
    "    return X_train, X_validation, X_test, y_train, y_validation, y_test\n",
    "\n",
    "\n",
    "def build_model(input_shape):\n",
    "    \"\"\"Generates CNN model\n",
    "\n",
    "    :param input_shape (tuple): Shape of input set\n",
    "    :return model: CNN model\n",
    "    \"\"\"\n",
    "\n",
    "    # build network topology\n",
    "    model = keras.Sequential()\n",
    "\n",
    "    # 1st conv layer\n",
    "    model.add(keras.layers.Conv2D(64, (2, 2), padding='same', activation='relu', input_shape=input_shape))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 2nd conv layer\n",
    "    model.add(keras.layers.Conv2D(64, (2, 2), padding='same', activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 3rd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (2, 2), padding='same', activation='relu'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "    \n",
    "    # 4th conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (2, 2), padding='same', activation='relu'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "    \n",
    "#     base_model = tf.keras.applications.MobileNetV2(input_shape=input_shape,\n",
    "#                                                include_top=False,\n",
    "#                                                weights='imagenet')\n",
    "    \n",
    "#     base_model.trainable = False\n",
    "    \n",
    "#     model = tf.keras.Sequential([\n",
    "#         base_model,\n",
    "#         tf.keras.layers.Dense(112, activation='relu'),\n",
    "#         tf.keras.layers.Dropout(0.3),\n",
    "#         tf.keras.layers.Dense(64, activation='relu'),\n",
    "#         tf.keras.layers.Dropout(0.3),\n",
    "#         tf.keras.layers.Dense(1, activation='sigmoid')\n",
    "#     ])\n",
    "\n",
    "#     URL = \"https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/2\"\n",
    "#     feature_extractor = hub.KerasLayer(URL,\n",
    "#                                    input_shape= input_shape)\n",
    "    \n",
    "#     feature_extractor.trainable = False\n",
    "    \n",
    "    \n",
    "#     model = tf.keras.Sequential([\n",
    "#         feature_extractor,\n",
    "#         tf.keras.layers.Dense(112, activation='relu'),\n",
    "#         tf.keras.layers.Dropout(0.3),\n",
    "#         tf.keras.layers.Dense(64, activation='relu'),\n",
    "#         tf.keras.layers.Dropout(0.3),\n",
    "#         tf.keras.layers.Dense(1, activation='sigmoid')\n",
    "#     ])\n",
    "\n",
    "    # flatten output and feed it into dense layer\n",
    "    model.add(keras.layers.Flatten())\n",
    "    \n",
    "    model.add(keras.layers.Dense(256, activation='relu', kernel_regularizer=keras.regularizers.l2(0.001)))\n",
    "    model.add(keras.layers.Dropout(0.4))\n",
    "    model.add(keras.layers.Dense(64, activation='relu', kernel_regularizer=keras.regularizers.l2(0.001)))\n",
    "    model.add(keras.layers.Dropout(0.3))\n",
    "\n",
    "    # output layer\n",
    "    model.add(keras.layers.Dense(2, activation='softmax'))\n",
    "\n",
    "    return model\n",
    "\n",
    "\n",
    "\n",
    "def predict(model, X, y):\n",
    "    \"\"\"Predict a single sample using the trained model\n",
    "\n",
    "    :param model: Trained classifier\n",
    "    :param X: Input data\n",
    "    :param y (int): Target\n",
    "    \"\"\"\n",
    "\n",
    "    # add a dimension to input data for sample - model.predict() expects a 4d array in this case\n",
    "    X = X[np.newaxis, ...] # array shape (1, 130, 13, 1)\n",
    "\n",
    "    # perform prediction\n",
    "    prediction = model.predict(X)\n",
    "\n",
    "    # get index with max value\n",
    "    predicted_index = np.argmax(prediction, axis=1)\n",
    "\n",
    "    print(\"Target: {}, Predicted label: {}\".format(y, predicted_index))\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    # get train, validation, test splits\n",
    "    X_train, X_validation, X_test, y_train, y_validation, y_test = prepare_datasets(0.25, 0.2)\n",
    "\n",
    "    # create network\n",
    "    input_shape = (X_train.shape[1], X_train.shape[2], 1)\n",
    "    model = build_model(input_shape)\n",
    "    \n",
    "#     lr_schedule = tf.keras.optimizers.schedules.InverseTimeDecay(\n",
    "#       0.001,\n",
    "#       decay_steps=STEPS_PER_EPOCH*1000,\n",
    "#       decay_rate=1,\n",
    "#       staircase=False\n",
    "#     )\n",
    "\n",
    "    reduce_lr = tf.keras.callbacks.ReduceLROnPlateau(\n",
    "                    monitor='val_loss', factor=0.1, patience=3, verbose=1, mode='auto',\n",
    "                    min_delta=0.0001, cooldown=0 #, min_lr=0.001\n",
    "                )\n",
    "    \n",
    "    earlystop = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=15, restore_best_weights = True, verbose=1)\n",
    "    checkpoint = tf.keras.callbacks.ModelCheckpoint(\n",
    "            \"sickweight4.h5\", monitor='val_loss', verbose=1, save_best_only=True,\n",
    "            save_weights_only=True, mode='min', save_freq='epoch' \n",
    "        )\n",
    "\n",
    "    callbacks = [reduce_lr, earlystop, checkpoint]\n",
    "    \n",
    "    # compile model\n",
    "    optimizer = keras.optimizers.Adam(learning_rate=0.001)\n",
    "    model.compile(optimizer=optimizer,\n",
    "                  loss='sparse_categorical_crossentropy',\n",
    "                  metrics=['accuracy'])\n",
    "    \n",
    "\n",
    "    model.summary()\n",
    "    \n",
    "    batch_size = 64\n",
    "    # train model\n",
    "    history = model.fit(X_train,\n",
    "                        y_train,\n",
    "                        batch_size=batch_size,\n",
    "                        steps_per_epoch = int(np.ceil(len(X_train) / float(batch_size))), #  len(X_train) // batch_size,\n",
    "                        validation_data=(X_validation, y_validation),\n",
    "                        validation_steps= int(np.ceil(len(X_validation) / float(batch_size))),\n",
    "                        epochs=50,\n",
    "                        callbacks=callbacks,\n",
    "                       )\n",
    "\n",
    "    # plot accuracy/error for training and validation\n",
    "    plot_history(history)\n",
    "\n",
    "    # evaluate model on test set\n",
    "    test_loss, test_acc = model.evaluate(X_test, y_test, verbose=2)\n",
    "    print('\\nTest accuracy:', test_acc)\n",
    "\n",
    "    # pick a sample to predict from the test set\n",
    "    X_to_predict = X_test[0]\n",
    "    y_to_predict = y_test[0]\n",
    "\n",
    "    # predict sample\n",
    "    predict(model, X_to_predict, y_to_predict)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Target: 1, Predicted label: [1]\n"
     ]
    }
   ],
   "source": [
    "# DOING MANUAL PREDICTIONS:\n",
    "\n",
    "input_shape = (X_train.shape[1], X_train.shape[2], 1)\n",
    "model = build_model(input_shape)\n",
    "\n",
    "X_to_predict = X_test[12]\n",
    "y_to_predict = y_test[12]\n",
    "\n",
    "predict(model, X_to_predict, y_to_predict)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ 126 1216]\n",
      " [  28  676]]\n"
     ]
    }
   ],
   "source": [
    "# PLOTTING CONFUSION MATRIX\n",
    "\n",
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "yhat_probs = model.predict(X_test, verbose=0)\n",
    "tt = np.argmax(yhat_probs, axis = 1)\n",
    "\n",
    "cm=confusion_matrix(y_test, tt)\n",
    "print(cm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ALERT\n",
    "### BELOW IS THE TESTING AREA, DON't TRY TO RUN THAT CODE."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(14, 197, 13, 1)\n",
      "(4, 197, 13, 1)\n",
      "(7, 197, 13, 1)\n",
      "Model: \"sequential_33\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_102 (Conv2D)          (None, 195, 11, 32)       320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_77 (MaxPooling (None, 98, 6, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_77 (Batc (None, 98, 6, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_103 (Conv2D)          (None, 96, 4, 32)         9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_78 (MaxPooling (None, 48, 2, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_78 (Batc (None, 48, 2, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_104 (Conv2D)          (None, 47, 1, 32)         4128      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_79 (MaxPooling (None, 24, 1, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_79 (Batc (None, 24, 1, 32)         128       \n",
      "_________________________________________________________________\n",
      "flatten_8 (Flatten)          (None, 768)               0         \n",
      "_________________________________________________________________\n",
      "dense_19 (Dense)             (None, 64)                49216     \n",
      "_________________________________________________________________\n",
      "dropout_11 (Dropout)         (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_20 (Dense)             (None, 2)                 130       \n",
      "=================================================================\n",
      "Total params: 63,426\n",
      "Trainable params: 63,234\n",
      "Non-trainable params: 192\n",
      "_________________________________________________________________\n",
      "Train on 14 samples, validate on 4 samples\n",
      "Epoch 1/30\n",
      "14/14 [==============================] - 1s 95ms/sample - loss: 0.8285 - accuracy: 0.4286 - val_loss: 0.7456 - val_accuracy: 0.5000\n",
      "Epoch 2/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.6866 - accuracy: 0.6429 - val_loss: 0.6370 - val_accuracy: 0.5000\n",
      "Epoch 3/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.5165 - accuracy: 0.8571 - val_loss: 0.6029 - val_accuracy: 0.5000\n",
      "Epoch 4/30\n",
      "14/14 [==============================] - 0s 8ms/sample - loss: 0.5710 - accuracy: 0.7857 - val_loss: 0.5909 - val_accuracy: 0.5000\n",
      "Epoch 5/30\n",
      "14/14 [==============================] - 0s 6ms/sample - loss: 0.5277 - accuracy: 0.9286 - val_loss: 0.5848 - val_accuracy: 0.5000\n",
      "Epoch 6/30\n",
      "14/14 [==============================] - 0s 7ms/sample - loss: 0.4473 - accuracy: 0.9286 - val_loss: 0.5834 - val_accuracy: 0.5000\n",
      "Epoch 7/30\n",
      "14/14 [==============================] - 0s 6ms/sample - loss: 0.4049 - accuracy: 1.0000 - val_loss: 0.5833 - val_accuracy: 0.5000\n",
      "Epoch 8/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.3122 - accuracy: 1.0000 - val_loss: 0.5823 - val_accuracy: 0.5000\n",
      "Epoch 9/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.3293 - accuracy: 0.9286 - val_loss: 0.5806 - val_accuracy: 0.5000\n",
      "Epoch 10/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.2727 - accuracy: 1.0000 - val_loss: 0.5782 - val_accuracy: 0.7500\n",
      "Epoch 11/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.2005 - accuracy: 1.0000 - val_loss: 0.5754 - val_accuracy: 1.0000\n",
      "Epoch 12/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.2637 - accuracy: 1.0000 - val_loss: 0.5725 - val_accuracy: 1.0000\n",
      "Epoch 13/30\n",
      "14/14 [==============================] - 0s 6ms/sample - loss: 0.2373 - accuracy: 1.0000 - val_loss: 0.5689 - val_accuracy: 1.0000\n",
      "Epoch 14/30\n",
      "14/14 [==============================] - 0s 8ms/sample - loss: 0.1690 - accuracy: 1.0000 - val_loss: 0.5653 - val_accuracy: 1.0000\n",
      "Epoch 15/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.1379 - accuracy: 1.0000 - val_loss: 0.5616 - val_accuracy: 1.0000\n",
      "Epoch 16/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.1522 - accuracy: 1.0000 - val_loss: 0.5576 - val_accuracy: 1.0000\n",
      "Epoch 17/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.1192 - accuracy: 1.0000 - val_loss: 0.5541 - val_accuracy: 1.0000\n",
      "Epoch 18/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.1377 - accuracy: 1.0000 - val_loss: 0.5508 - val_accuracy: 1.0000\n",
      "Epoch 19/30\n",
      "14/14 [==============================] - 0s 4ms/sample - loss: 0.1280 - accuracy: 1.0000 - val_loss: 0.5481 - val_accuracy: 1.0000\n",
      "Epoch 20/30\n",
      "14/14 [==============================] - 0s 4ms/sample - loss: 0.1182 - accuracy: 1.0000 - val_loss: 0.5453 - val_accuracy: 1.0000\n",
      "Epoch 21/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.1032 - accuracy: 1.0000 - val_loss: 0.5427 - val_accuracy: 1.0000\n",
      "Epoch 22/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.1358 - accuracy: 1.0000 - val_loss: 0.5403 - val_accuracy: 1.0000\n",
      "Epoch 23/30\n",
      "14/14 [==============================] - 0s 6ms/sample - loss: 0.0613 - accuracy: 1.0000 - val_loss: 0.5380 - val_accuracy: 1.0000\n",
      "Epoch 24/30\n",
      "14/14 [==============================] - 0s 6ms/sample - loss: 0.1491 - accuracy: 1.0000 - val_loss: 0.5356 - val_accuracy: 1.0000\n",
      "Epoch 25/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.0774 - accuracy: 1.0000 - val_loss: 0.5331 - val_accuracy: 1.0000\n",
      "Epoch 26/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.0586 - accuracy: 1.0000 - val_loss: 0.5305 - val_accuracy: 1.0000\n",
      "Epoch 27/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.0402 - accuracy: 1.0000 - val_loss: 0.5279 - val_accuracy: 1.0000\n",
      "Epoch 28/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.0413 - accuracy: 1.0000 - val_loss: 0.5252 - val_accuracy: 1.0000\n",
      "Epoch 29/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.0613 - accuracy: 1.0000 - val_loss: 0.5223 - val_accuracy: 1.0000\n",
      "Epoch 30/30\n",
      "14/14 [==============================] - 0s 5ms/sample - loss: 0.0313 - accuracy: 1.0000 - val_loss: 0.5201 - val_accuracy: 1.0000\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYkAAAEWCAYAAACT7WsrAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAgAElEQVR4nOzdd3gVVfrA8e+bEEgCJJCEHiCAIjWhhF7EAgKrCCIgRRcbVnR11cWCi22XBXX5qSuKiihKW7AhoIGVgCggoNRQE1oIhBBqAgkp5/fHTMJNSL9Jbsr7eZ773Kln3km5751zZs4RYwxKKaVUTtxcHYBSSqmyS5OEUkqpXGmSUEoplStNEkoppXKlSUIppVSuNEkopZTKlSYJpRQiYkTkGlfHocoeTRKqwhCRcBE5IyLVXB2LUhWFJglVIYhIENAHMMCQUj52ldI8nlKlSZOEqijuATYAc4A/O64QES8ReUtEDovIORFZJyJe9rreIvKriJwVkaMiMt5eHi4iDziUMV5E1jnMGxF5TET2A/vtZf9nl3FeRLaISB+H7d1F5AURiRSRC/b6xiLyHxF5K1u8S0XkLzmdpIi0EpGVInJaRPaKyEh7eXcROSEi7g7bDhOR7fZ0VxFZb5/ncRF5T0SqFuUHrSoXTRKqorgH+NJ+3SIi9RzWvQl0BnoCfsBzQLqINAFWAO8CdYAOwNZCHHMo0A1oY89vssvwA+YB/xURT3vd08BoYDDgA9wHXAQ+A0aLiBuAiAQANwHzsx9MRKoDK+2y69rlvS8ibY0xG4BE4EaHXcbY2wKkAU8BAUAP+xiPFuJcVSWlSUKVeyLSG2gKLDLGbAEisT4gsT987wOeNMYcM8akGWN+NcYkA2OBVcaY+caYFGNMvDGmMEnin8aY08aYSwDGmC/sMlKNMW8B1YDr7G0fAF4yxuw1lm32tr8B57A+tAHuAsKNMbE5HO9W4JAx5lP7GL8DS4A77fXzsRIHIlITKyHNt2PbYozZYO93CPgQuL4Q56oqKU0SqiL4MxBmjDllz8/jSpVTAOCJlTiya5zL8oI66jgjIn8Vkd12ldZZwNc+fn7H+gwYZ0+PA+bmsl1ToJtdZXTWPsZYoL69fh5wh91wfwfwuzHmsB1bSxH53q6SOg/8wyE2pXKlDW6qXLPbFkYC7iJywl5cDaglIiHADiAJaAFsy7b7UaBrLkUnAt4O8/Vz2CazC2W7/eFvWFcEu4wx6SJyBhCHY7UAduZQzhfATjve1sA3ucR0FFhjjOmf00pjTISIHAYGkbWqCWAm8Acw2hhzwW7zuDOHYpTKQq8kVHk3FKu+vQ1We0AHrA/an4F7jDHpwGzgbRFpaDcg97C/bX8J3CwiI0Wkioj4i0gHu9ytWN/Kve3nB+7PJ46aQCoQB1QRkZex2h4yfAy8JiLXiiVYRPwBjDHRWO0Zc4ElGdVXOfgeaCkid4uIh/3qIiKtHbaZBzwB9AX+my2+80CCiLQCHsnnfJQCNEmo8u/PwKfGmCPGmBMZL+A9YKx9e+ozWFcUm4DTwL8AN2PMEax6+7/ay7cCIXa5/wYuA7FY1UFf5hPHj1iN4PuAw1hXL47VUW8Di4AwrA/rTwAvh/WfAe3JvaoJY8wFYABWu0UMcMI+F8fnQuYD/YCfHKrfsH8GY4ALwEfAwnzORykARAcdUsr1RKQvVrVTkH31o1SZoFcSSrmYiHgATwIfa4JQZY0mCaVcyG5POAs0AGa4OBylrqLVTUoppXKlVxJKKaVyVWGekwgICDBBQUGuDkMppcqVLVu2nDLG1MltfYVJEkFBQWzevNnVYSilVLliP4CZK5dUN4nIbBE5KSI5PX2K/bDROyJyQES2i0in0o5RKaWU69ok5gAD81g/CLjWfk3A6lJAKaVUKXNJdZMxZq09SExubgc+N9atVxtEpJaINDDGHC+VACuJqLgEHp/3B6cSkgu0fduGPrw7phM1qjn/Z2OMYfqPe1m8JTrfbaubRGalvkQtLjh9XKUqohivloT8LaxEyi6rbRKNyNqlQbS9LEuSEJEJWFcaNGnSpNSCqwiOnb3EuI83kpyazoC29fLdPiXN8PUfx5jw+WZmj++Cp4d7vvvk5d8r9/F+eCT9rqtDA1/PPLdtd3Y11x45wnbfG0lyr+7UcZWqiNJ8S+7zr6wmCclh2VUPdBhjZgGzAEJDQ/WBjwI6lZDM3R9v5EJyKgsmdKdtQ98C7dfrGn+eWriNifP/YObYTlRxL1pt5cc/R/HOTwcYFdqYqcPbI5LTr9vB0tkQ60PwE/8F97L6J6tUxVRWn5OIxup/P0MgVodmyknnk1L48+zfiDl3iU/HdylwggAY1jGQV4a0ZWVELM8t3k56euHz8qJNR3l92W4Gt6/PP+4oQIIAiFoNQX00QSjlAmU1SXwH3GPf5dQdOKftEc67dDmN++dsYl/sBT4Y15nQIL9Cl/HnnkE83b8lX/1xjFe/j6AwT+wv33GcSV9tp8+1Afx7VAfc3QqQIE4fhDOHoHm/QseqlHKeS76aiUhGd8YBIhIN/B3wADDGfAAsx+rC+QDWOMD3uiLOiuRyajqPfLmFzYfP8O7ojvS7rm6Ry5p44zWcv5TCx+sO4uPlwdP9W+a7z5p9cTy54A86NqnNh3d3plqVArZpRIVb7y1uKHK8Sqmic9XdTaPzWW+Ax0opnAovLd3w1KKthO+NY+od7bk1uKFT5YkIL/6pNeeTUnjnf/vx8azCA32a57r9lsOneXjuFq6pW5PZ47vgXbUQf3ZR4eDTCPyvcSpmpVTRaCVvBWeM4aVvdrBs+3FeGNyKu7oWz10QIsI/7wjmQlIqry/bjY+nByO7NL5qu4iY84z/dBP1fT35/L6u+Hp5FPwg6elwcA1cNxgK0nahlCp2ZbVNQhUDYwxTV+xh/m9HeeyGFkzo26JYy3d3E2bc1YE+1wYw6avtrNiRtdno4KlE7pm9kRrVqjD3/q7UqVktl5JycWIbXDoDzbWqSSlX0SRRgb0fHsmHa6O4u3tTnhlwXYkco1oVdz68uzMdGtfiyQVb+Xl/HAAx9nMY6Qbm3t+NwNrehS88oz2i+fXFF7BSqlA0SVRQczccZvqPe7m9Q0NeGdK2YLeaFpF31Sp8Or4rzetUZ8LnW1gZEcvdn2zk3KUUPru3K9fUrVG0gqPCoW5bqFH0RnallHO0TaKcuJyazmvfRxCfmH8XGilphlW7Y7mpVV3eHBGCW0FuNXWSr7cHn9/flZEfrOfBzzdTrYobn9/XlfaBBX8OI4uUS3B4PXR9sHgDVUoViiaJcuKbrceYu+EwzQOqF+j5giEhDfnX8GA8ivhUdFHUrenJFw9044Wvd3JfryC6NfcvemFHNkBasj4foZSLaZIoB9LTDR+uiaRNAx+WPdG7RKuOnBVY25vP7+vqfEFR4eDmAU17Ol+WUqrItE2iHFi5O5bIuEQe7teiTCeIYhW1Ghp3g6raoZ9SrqRJoowzxjAzPJLGfl4Mblff1eGUjsR4OL5dq5qUKgM0SZRxGw+eZuvRs0zo26LIva6WO4fWAkaThFJlQCX51Cm/PlgTSUCNqozoHOjqUEpP5Gqo5gsNO7o6EqUqPU0SZVhEzHnC98Zxb69mTg/yU65EhUMz7RpcqbJAk0QZ9sGaSGpUq8K47k1dHUrpOR0FZw9rVZNSZYQmiTLqSPxFvt8ew5huTQrXKV55l9kVh/bXpFRZoEmijPro5yiquLlxf+9mrg6ldEWFg08g+BdvZ4RKqaLRJFFC1u0/xbmLKUXa91RCMos2H+WOTo2o5+NZzJGVYelpELXGqmqqLM+DKFXGaZIoAbtizjHuk42M/WQD55MKnyjm/HKIy2npTOib+0A+FdLxbZB0VkehU6oM0SRRAsJ2xSICe45f4IE5m7l0Oa3A+yYkp/L5+kMMbFuf5nWK2HtqeZXRHtGsr0vDUEpdoUmiBKyMiCW0aW3+PaoDmw6f5tEvt3A5Nb1A+87feITzSak8fH0lrJOPWg312mnX4EqVIU4lCRF5XERqF1cwFcHR0xeJOH6e/m3qcVtIQ/4xrD2r98bx1/9uIy3d5LlvcmoaH6+LomcLf0Ia1yqliMuIyxetnl+b93N1JEopB85eSdQHNonIIhEZKJWm97ncrYyIBaB/G6ufpdFdmzBpUCuWboth8rc7MSb3RPHtHzHEnk/mkX6V8Cri6AZIu6y3vipVxjiVJIwxLwHXAp8A44H9IvIPEamEn3KWsIgTtKxXg2YBV3ovffj6FjzSrwXzNh7hXz/szXG/tHTDB2sjadvQh97XBJRWuGVH5GpwrwpNe7g6EqWUA6fbJIz11fiE/UoFagOLRWRabvvYVx17ReSAiEzKYX0TEVktIn+IyHYRGexsnKXhTOJlNh06w4A2V/fW+twt1zG2WxM+WBPJzPDIq9avjDhBVFwij1Sm7sAdRYVr1+BKlUHOtkk8ISJbgGnAL0B7Y8wjQGdgeC77uAP/AQYBbYDRItIm22YvAYuMMR2Bu4D3nYmztPy05yRp6Yb+bepdtU5EeO32dtaIcT/s4cuNhzPXGWOYuSaKpv7eDGrXoDRDLhsS4+HEdmh+vasjUUpl42wPagHAHcaYw44LjTHpInJrLvt0BQ4YY6IARGQBcDsQ4VgE4GNP+wIxTsZZKsIiTlDfx5P2jXIe19nNTXhrZAgJyam89M1OalSrwu0dGrE+Kp5tR8/yxrB2BRqatMI5GG69a3uEUmWOs9VNy4HTGTMiUlNEugEYY3bnsk8j4KjDfLS9zNEUYJyIRNvHmJhTQSIyQUQ2i8jmuLi4op1BMUlKSWPtvlP0b1MPtzw+6D3c3Xh/bCe6BPnx10Xb+GlPLB+siSKgRjWGd6pE3YE7igrXrsGVKqOcTRIzgQSH+UR7WV5y+gTNfsvPaGCOMSYQGAzMFZGrYjXGzDLGhBpjQuvUqVOIsIvfuv2nuJSSxoC2V1c1Zefp4c4nfw6lVYOaPPzF76zdF8d9vYMqV3fgGYyByHCra3C3Snj+SpVxziYJMQ73dBpj0sm/CisaaOwwH8jV1Un3A4vsMtcDnlhVW2VWWMQJalarQrdm/gXavqanB5/d25XGtb2o6VnJugN3dDoKzh3R5yOUKqOcbZOIEpEnuHL18CgQlc8+m4BrRaQZcAyrYXpMtm2OADcBc0SkNVaScG19Uh7S0g3/232SG1rVpWqVgudd/xrV+OaxXpy7lIKPZyXqDtxRRlccLW50aRhKqZw5eyXxMNAT68M+GugGTMhrB2NMKvA48COwG+supl0i8qqIDLE3+yvwoIhsA+YD401eT6G52O9HzhCfeLlAVU3Z1fT0ILC2dwlEVU5ErQbfxuBXyTozVKqccOpKwhhzEutKoLD7LcdqkHZc9rLDdATQy5nYSlPYrhNUdXfj+paubRcpd9LT4OBaaH2bdg2uVBnlVJIQEU+s9oO2WFVCABhj7nMyrnLDGENYRCw9WvhTs7JWGRXV8a2QdE5vfVWqDHO2umkuVv9NtwBrsBqhLzgbVHmyLzaBw/EXi1TVVOlFrrbem+lDdEqVVc4miWuMMZOBRGPMZ8CfgPbOh1V+rIw4AUD/1pokCi0qHOq1hxpaTadUWeVsksgYdu2siLTDejo6yMkyy5WwiFg6NqlF3co0zGhxuHwRjm6EFv1cHYlSKg/OJolZ9ngSLwHfYXWt8S+noyonjp+7xPboczn21aTyceRXu2vwfq6ORCmVhyI3XNtPQJ83xpwB1gKV7h7GVfbYETn1+qryERVudQ3epKerI1FK5aHIVxL209WPF2Ms5U5YRCzN61TnmrqVbCzq4pDZNXglfkZEqXLA2eqmlSLyjIg0FhG/jFexRFbGnbuUwvrIeL2KKIqEODixQ6ualCoHnO2WI+N5iMcclhkqQdVT+N6TpOYydoTKx8E11rs+H6FUmefsE9fNiiuQ8iYsIpaAGtXo2LiWq0Mpf6LCwdMXGnZwdSRKqXw4+8T1PTktN8Z87ky5ZV1yahrhe04ypEOjPMeOUDkwxkoSzfpq1+BKlQPOVjd1cZj2xOq59XegQieJXyPjSbxcsLEjVDano+DcUej9F1dHopQqAGerm7KMGCcivlhddVRoYbtiqV7VnZ4tCjZ2hHIQ+ZP1ru0RSpULzt7dlN1F4NpiLrNMSU83rNodS7/r6lKtilaXFFpUOPg20a7BlSonnG2TWMqVoUfdgDbYI8pVVFujzxJ3IVmrmooiPQ0O/gxthmjX4EqVE862SbzpMJ0KHDbGRDtZZpkWtiuWKm5Cv+vqujqU8ifmD0g+By20qkmp8sLZJHEEOG6MSQIQES8RCTLGHHI6sjJqZcQJujf3x9dLx44otCjtGlyp8sbZNon/AukO82n2sgrpwMkEIuMStaqpqKLWQP1gqB7g6kiUUgXkbJKoYoy5nDFjT1d1sswya6Xdod/NOnZE4V1OhCMbtCsOpcoZZ6ub4kRkiDHmOwARuR045XxYZVNYxAnaN/KlYS0vV4dS/hxeD+kpmiQqoJSUFKKjo0lKSnJ1KCoPnp6eBAYG4uFRuKpyZ5PEw8CXIvKePR8N5PgUdnl38nwSW4+e5embW7o6lPIpajW4V4Om2jV4RRMdHU3NmjUJCgpC9K61MskYQ3x8PNHR0TRrVrjelJx9mC4S6C4iNQAxxlTY8a1X7T6JMTCgrfb6WiRR4dCkG3joVVhFk5SUpAmijBMR/P39iYuLK/S+TrVJiMg/RKSWMSbBGHNBRGqLyOvOlFlWhUWcoKm/Ny3r6dgRhZZwEmJ3alVTBaYJouwr6u/I2YbrQcaYsxkz9ih1g/PbSUQGisheETkgIpNy2WakiESIyC4RmedknE5JSE7l1wPxDGhTT/8ZiuLgWutdu+JQqtxxNkm4i0i1jBkR8QKq5bE9IuIO/AcYhPWE9mgRaZNtm2uB54Fexpi2gEt7g1uzN47Laen01wGGiiZyNXjWggYhro5EVUBnz57l/fffL9K+gwcP5uzZs/lvWIk5myS+AP4nIveLyP3ASuCzfPbpChwwxkTZt8wuAG7Pts2DwH/sKxOMMSedjNMpYREn8Ktelc5Na7syjPJJuwZXJSyvJJGWlpbnvsuXL6dWrbI3JowxhvT09Pw3LAVOJQljzDTgdaA11lXBD0DTfHZrBBx1mI+2lzlqCbQUkV9EZIOIDMypIBGZICKbRWRzURpkCuJyajo/7TnJza3r4q5jRxRefCScj9auOFSJmTRpEpGRkXTo0IFnn32W8PBwbrjhBsaMGUP79u0BGDp0KJ07d6Zt27bMmjUrc9+goCBOnTrFoUOHaN26NQ8++CBt27ZlwIABXLp06apjLV26lG7dutGxY0duvvlmYmOtZ6cSEhK49957ad++PcHBwSxZsgSAH374gU6dOhESEsJNN90EwJQpU3jzzSs9GrVr145Dhw5lxvDoo4/SqVMnjh49yiOPPEJoaCht27bl73//e+Y+mzZtomfPnoSEhNC1a1cuXLhAnz592Lp1a+Y2vXr1Yvv27U7/fJ29BRbgBNZT1yOBg8CSfLbP6ZPWZJuvgtWbbD8gEPhZRNo5tn8AGGNmAbMAQkNDs5dRLDYejOdCUqqOZV1UGV1xNO/nyihUKXll6S4iYs4Xa5ltGvrw99va5rp+6tSp7Ny5M/MDMjw8nN9++42dO3dm3u45e/Zs/Pz8uHTpEl26dGH48OH4+2ft6n///v3Mnz+fjz76iJEjR7JkyRLGjRuXZZvevXuzYcMGRISPP/6YadOm8dZbb/Haa6/h6+vLjh07ADhz5gxxcXE8+OCDrF27lmbNmnH69Ol8z3Xv3r18+umnmVdGb7zxBn5+fqSlpXHTTTexfft2WrVqxahRo1i4cCFdunTh/PnzeHl58cADDzBnzhxmzJjBvn37SE5OJjg4uOA/6FwUKUmISEvgLmA0EA8sxLoFtiBfF6OBxg7zgUBMDttsMMakAAdFZC9W0thUlHidsTIiFi8Pd3pfq11JFElUONRqArUr7Ui3ygW6du2a5XmAd955h6+//hqAo0ePsn///quSRLNmzejQwRpSt3Pnzhw6dOiqcqOjoxk1ahTHjx/n8uXLmcdYtWoVCxYsyNyudu3aLF26lL59+2Zu4+fnl2/cTZs2pXv37pnzixYtYtasWaSmpnL8+HEiIiIQERo0aECXLtaYbz4+PgCMGDGC1157jenTpzN79mzGjx+f7/EKoqhXEnuAn4HbjDEHAETkqQLuuwm4VkSaAcewks2YbNt8g5WA5ohIAFb1U1QRYy0yYwxhu2Lp2zIATw+tTy+0tFTrzqa2w7Rr8Eoir2/8pal69eqZ0+Hh4axatYr169fj7e1Nv379cnw6vFq1K/fcuLu751jdNHHiRJ5++mmGDBlCeHg4U6ZMAazPiux3Pua0DKBKlSpZ2hscY3GM++DBg7z55pts2rSJ2rVrM378eJKSknIt19vbm/79+/Ptt9+yaNEiNm/enNOPptCK2iYxHKuaabWIfCQiN5FzNdJVjDGpwOPAj8BuYJExZpeIvCoiQ+zNfgTiRSQCWA08a4yJL2KsRbbj2DlOnE/SqqaiivkDks9rVZMqUTVr1uTChdyf4z137hy1a9fG29ubPXv2sGHDhiIf69y5czRqZDWhfvbZlXt0BgwYwHvvvZc5f+bMGXr06MGaNWs4ePAgQGZ1U1BQEL///jsAv//+e+b67M6fP0/16tXx9fUlNjaWFStWANCqVStiYmLYtMmqWLlw4QKpqakAPPDAAzzxxBN06dKlQFcuBVGkJGGM+doYMwpoBYQDTwH1RGSmiAwowP7LjTEtjTEtjDFv2MtezugDylieNsa0Mca0N8YsyLvEkhG2KxZ3N+HGVjp2RJFEhQOiXYOrEuXv70+vXr1o164dzz777FXrBw4cSGpqKsHBwUyePDlLdU5hTZkyhREjRtCnTx8CAq5UQb/00kucOXOGdu3aERISwurVq6lTpw6zZs3ijjvuICQkhFGjRgEwfPhwTp8+TYcOHZg5cyYtW+bc1U9ISAgdO3akbdu23HffffTq1QuAqlWrsnDhQiZOnEhISAj9+/fPvBrp3LkzPj4+3HvvvUU+x+zEmOJp7xURP2AEMMoYc2OxFFoIoaGhprgurzLc8u+11K7uwYIJPYq13Erj08FwOQEeWuvqSFQJ2r17N61bt3Z1GAqIiYmhX79+7NmzBze3q68BcvpdicgWY0xobmUW2xjXxpjTxpgPXZEgSsKhU4nsjb2gVU1FlZwAR3/TqialSsnnn39Ot27deOONN3JMEEVVHLfAVkgZY0f0b6NjRxTJkYyuwfX5CKVKwz333MM99xR/J9zFl24qmLCIE7Ru4ENjP29Xh1I+Rdpdgzcpev2vUsr1NEnk4FRCMlsOn2GAXkUUXVS4lSC0a3ClyjVNEjn4afdJ0g06lnVRXYiFk7u0Kw6lKgBNEjkIizhBo1petGng4+pQyqeDa6z35v1cGYVSqhhoksjm4uVUft5/iv46dkTRRYWDV22o73y/MUrlx5muwgFmzJjBxYsXizGiikXvbkq9DIeu3McfcegM3dL3M7L2JThw3IWBlWORP1kP0GnX4KoUZCSJRx99tEj7z5gxg3HjxuHt7bqbVFJTU6lSpWx+HJfNqEpT8nn4YnjmbCjweVXgfy6LqGK4tr+rI1CVhGNX4f3792f69OlMnz6dRYsWkZyczLBhw3jllVdITExk5MiRREdHk5aWxuTJk4mNjSUmJoYbbriBgIAAVq9enaXsV199laVLl3Lp0iV69uzJhx9+iIhw4MABHn74YeLi4nB3d+e///0vLVq0YNq0acydOxc3NzcGDRrE1KlT6devH2+++SahoaGcOnWK0NBQDh06xJw5c1i2bBlJSUkkJiby3Xffcfvtt3PmzBlSUlJ4/fXXuf12a6idzz//nDfffBMRITg4mPfff5/g4GD27duHh4cH58+fJzg4mP379+Ph4VGsP19NEp6+cP9KAFLT07lvziY6N63Nkzfl/Ki8KgB3D61qqqxWTIITO4q3zPrtYdDUXFdn7yo8LCyM/fv389tvv2GMYciQIaxdu5a4uDgaNmzIsmXLAKsfJl9fX95++21Wr16dpZuNDI8//jgvv/wyAHfffTfff/89t912G2PHjmXSpEkMGzaMpKQk0tPTWbFiBd988w0bN27E29u7QF2Dr1+/nu3bt+Pn50dqaipff/01Pj4+nDp1iu7duzNkyBAiIiJ44403+OWXXwgICOD06dPUrFmTfv36sWzZMoYOHcqCBQsYPnx4sScI0CRhfaA17grApsh41l6KZ0xoJ2jcwMWBKaWKIiwsjLCwMDp27AhYAwLt37+fPn368Mwzz/C3v/2NW2+9lT59+uRb1urVq5k2bRoXL17k9OnTtG3bln79+nHs2DGGDRsGgKenJ2B1F37vvfdmVlsVpIO9/v37Z25njOGFF15g7dq1uLm5cezYMWJjY/npp5+48847M5NYxvYPPPAA06ZNY+jQoXz66ad89NFHhfxJFYwmCQdhESeoVsWNvi3ruDoUpcqnPL7xlxZjDM8//zwPPfTQVeu2bNnC8uXLef755xkwYEDmVUJOkpKSePTRR9m8eTONGzdmypQpmV1153bc/LoGz95FuWPX4F9++SVxcXFs2bIFDw8PgoKC8uwavFevXhw6dIg1a9aQlpZGu3btcj0XZ+jdTbaMsSP6XBuAd1XNnUqVF9m7Cr/llluYPXs2CQkJABw7doyTJ08SExODt7c348aN45lnnsnsrju3rsYzPtADAgJISEhg8eLFgDXIT2BgIN988w0AycnJXLx4kQEDBjB79uzMO6UcuwbfsmULQGYZOTl37hx169bFw8OD1atXc/jwYQBuuukmFi1aRHx8fJZyweqKY/To0cXa62t2miRsEcfPc+zsJe3QT6lyJntX4QMGDGDMmDH06NGD9u3bc+edd3LhwgV27NhB165d6dChA2+88QYvvfQSABMmTGDQoEHccEPWhz9r1arFgw8+SPv27Rk6dGjmSHAAc+fO5Z133iE4OJiePXty4sQJBg4cyJAhQwgNDaVDhw6Z41g/88wzzJw5k549e3Lq1I/0aEoAACAASURBVKlcz2Ps2LFs3ryZ0NBQvvzyS1q1agVA27ZtefHFF7n++usJCQnh6aefzrLPmTNnGD16dLH9PLMrtq7CXc3ZrsJnrNrH//1vP5tevJmAGtXy30EpBWhX4a60ePFivv32W+bOnVug7YvSVbjWq9jCdsUS2rS2JgilVLkwceJEVqxYwfLly0v0OJokgKOnLxJx/DwvDtZvQ0qp8uHdd98tleNomwQ6doRSzqoo1dYVWVF/R5oksJJEy3o1CAqonv/GSqksPD09iY+P10RRhhljiI+Pz3ymozAqfXXTmcTL/HboNI9c38LVoShVLgUGBhIdHU1cXJyrQ1F58PT0JDAwsND7Vfok4e4uvDC4NX2vvfqRfKVU/jw8PGjWrJmrw1AlpNInCR9PD+7vrX/gSimVE22TUEoplStNEkoppXJVYZ64FpE44LATRQQAuT8zX/5UtPOBindOFe18oOKdU0U7H7j6nJoaY3Lt1bTCJAlnicjmvB5NL28q2vlAxTuninY+UPHOqaKdDxT+nLS6SSmlVK40SSillMqVJokrZrk6gGJW0c4HKt45VbTzgYp3ThXtfKCQ56RtEkpVICLSD/jCGFP4R2uVyoFeSahKR0QOicglEUlweL3n6riUKosq/RPXqtK6zRizKr+NRKSKMSY1v2WFLUOp8qLSX0mIyEAR2SsiB0RkkqvjKQ72N+UdIrJVRIo+XJ+LiMhsETkpIjsdlvmJyEoR2W+/1y6hY48XkV9E5N8ichqYkssyNxF5SUQO27F+LiK+dhlBImJE5H4ROQL8lMs5zRORyyKSJiKJIvKovXySiCzOFtf/icg79vS9IrJbRC6ISJSIPFQSP4u8iEhjEVltx7FLRJ60l5fK76m45XE+U0TkmP2/tFVEBrs61oISEU8R+U1Ettnn9Iq9vJmIbLR/RwtFpGqeBRljKu0LcAcigeZAVWAb0MbVcRXDeR0CAlwdhxPx9wU6ATsdlk0DJtnTk4B/OfnzuTmXdeOBVGAi1pW2Vy7L7gMO2H87NYCvgLl2GUGAAT4HqtvbZzknezoReMf+O/yzHVc1oClwEfBx+Ds9DnS35/8EtAAEuN7etpO9rh8QXQq/owYOx6wJ7APaFOfvqZT/5nI7nynAM66Or4jnJEANe9oD2Ah0BxYBd9nLPwAeyaucyn4l0RU4YIyJMsZcBhYAt7s4pkrPGLMWOJ1t8e3AZ/b0Z8BQJw/zjYicdXg96LAuxhjzrjEm1RhzKZdlY4G37b+dBOB54C4RcazCnWKMSTTGXMrhnB4ENgNHjDFpxpjPgGSsRHAY+N3hHG8ELhpjNgAYY5YZYyKNZQ0QBvRx8udRKMaY48aY3+3pC8BuoBHF/3sqFXmcT7ll/30k2LMe9stg/T1lXKnm+zuq7EmiEXDUYT6acv6HYTNAmIhsEZEJrg6mmNQzxhwH6x8aqOtkeUONMbUcXh85rDuaw/bZlzUkazcwh7GuMhyHN8ypnAxNgZ7ANLu6KRlobJcLMA8YbU+PsecBEJFBIrJBRE6LyFlgMFZXCy4hIkFAR6xvqsX9eyp12c4H4HER2W5XGZaL6rMMIuIuIluBk8BKrJqTs+ZKG1m+n3mVPUlIDssqwj3BvYwxnYBBwGMi0tfVAZUzOf0NZF8Wg/VBn6EJVpVUbD7lZDgKTOfKN7w3gQXGmPn2+v8C/UQkEBiGnSREpBqwxN6+njGmFrCcnP+WS5yI1LDj+Ysx5rwrYihOOZzPTKyqvQ5YVX5vuTC8QrOvUjsAgVg1J61z2iyvMip7kojG+vaWIRDrn79cM8bE2O8nga+x/jjKu1gRaQBgv590cTzzgafsRsAawD+AhabgdzF9BNwNhGL9k34B3CgiNQGMMXFAOPApcNAYs9verypWu0UckCoig4ABxXNKhSMiHlgfqF8aY76yF5e131OB5XQ+xphY+4M2Het3Vi7/l4wxZ7H+nroDtRyqRfP9zKvsSWITcK39j14VuAv4zsUxOUVEqmd80IhIdawPkJ1571UufIfVuIv9/q2T5S2VrM9JfF3I/WcDc4G1wEEgCathu0CMMZuB54D3gDNYVRvZ/x/nATfjUNVk15c/gdX4eAarKqrU/2ZFRIBPgN3GmLcdVhX376lU5HY+GQnPNoxy9L8kInVEpJY97YX1t7QbWA3caW+W7++o0j9xbd/SNgPrDpLZxpg3XBySU0SkOdbVA1h15PPK2zmJyHysu3QCsKpv/g58g/XB2AQ4AowwxmRv3C6zcjmnfljVGAbrzqaHMurzyzoR6Q38DOwA0u3FL2Alu3L3e8rjfEZTfn9HwVgN0+5YX0AWGWNetT8jFgB+wB/AOGNMcq7lVPYkoZRSKneVvbpJKaVUHjRJKKWUypUmCaWUUrmqMB38BQQEmKCgIFeHoZRS5cqWLVtOmTzGuK4wSSIoKIjNm8tdX3ZKKeVSInI4r/Va3aSUUipXlT5JGGP4YecJLiSluDoUpZQqcyp9kog6lcijX25h6oo9rg5FKaXKnArTJlFULerU4P7ezfjo54P8KbgBPVu4rDNNpRSQkpJCdHQ0SUlJrg6lQvH09CQwMBAPD49C7VfpkwTA0/2vY2VELJOW7OCHv/TBu6r+WJRylejoaGrWrElQUBBWl0rKWcYY4uPjiY6OplmzZoXat9JXNwF4VXXnX8ODOXL6ItN/3OvqcJSq1JKSkvD399cEUYxEBH9//yJdnWmSsHVr7s89PZoy59dDbD5U5vsjU6pC0wRR/Ir6M9Uk4eC5ga1o6OvFc4u3k5SS5upwlFLK5Uo0SYjIQBHZKyIHRGRSDuv/LSJb7dc+eyjGjHVpDutKpb/8GtWqMHV4e6JOJTJj1f7SOKRSqow5e/Ys77//fpH2HTx4MGfPns1/w3KkxJKEiLgD/8EaQrMNMFpE2jhuY4x5yhjTwR5e713gK4fVlzLWGWOGlFSc2fW5tg6jQhsza20k245WrF+2Uip/eSWJtLS8axiWL19OrVq1ijWe1NTUPOdzk1+sBVWSt/F0BQ4YY6IARGQBcDsQkcv2o7EGYil9F0+Dt1/m7Iu3tmbNvjieW7ydpRN7U7WK1sop5QqvLN1FREzxDp3dpqEPf7+tba7rJ02aRGRkJB06dKB///786U9/4pVXXqFBgwZs3bqViIgIhg4dytGjR0lKSuLJJ59kwoQJwJXugRISEhg0aBC9e/fm119/pVGjRnz77bd4eXllOVZcXBwPP/wwR44cAWDGjBn06tWLKVOmEBMTw6FDhwgICGDAgAEsW7aMpKQkEhMT+d///sdzzz3HihUrEBFeeuklRo0aRXh4+FWxOqskk0QjrMHeM0QD3XLaUESaAs2AnxwWe4rIZqzB5acaY77JYb8JwASAJk2aFC3K+Ej4pD/c+BKE3geAj6cHbwxrx/2fbeY/qw/wVP+WRStbKVXuTJ06lZ07d7J161YAwsPD+e2339i5c2fm7aOzZ8/Gz8+PS5cu0aVLF4YPH46/v3+Wcvbv38/8+fP56KOPGDlyJEuWLGHcuHFZtnnyySd56qmn6N27N0eOHOGWW25h925rOPMtW7awbt06vLy8mDNnDuvXr2f79u34+fmxZMkStm7dyrZt2zh16hRdunShb9++AFfF6qySTBI5NaXnNgzeXcBiY4zj9VETY0yMPdTeTyKywxgTmaUwY2YBswBCQ0OLNsRerabQsBMsewZqNYFrbgbgptb1GNaxEf9ZfYCB7erTuoFPkYpXShVdXt/4S1PXrl2zfOi+8847fP21NUrw0aNH2b9//1VJolmzZnTo0AGAzp07c+jQoavKXbVqVZZv++fPn+fChQsADBkyJMuVR//+/fHzs2o81q1bx+jRo3F3d6devXpcf/31bNq0CR8fn6tidVZJ1qNEA40d5gOBmFy2vQuY77jAGBNjv0cB4UDH4g8RcK8CIz6Fuq1h0XiI3ZW56uVb21DL24NnF28jNS099zKUUhVa9erVM6fDw8NZtWoV69evZ9u2bXTs2DHH5w+qVauWOe3u7p5jW0J6ejrr169n69atbN26lWPHjlGzZs2rjpl9Pq9hp7Pv56ySTBKbgGtFpJmIVMVKBFfdpSQi1wG1gfUOy2qLSDV7OgDoRe5tGc6rVhPGLIKq1WHeKLgQC0Dt6lV57fZ27Dx2nlk/R5XY4ZVSZUfNmjUzv83n5Ny5c9SuXRtvb2/27NnDhg0binysAQMG8N5772XOZ1Rx5adv374sXLiQtLQ04uLiWLt2LV27di1yHHkpsSRhjEkFHgd+BHYDi4wxu0TkVRFxvFtpNLDAZE2NrYHNIrINWI3VJlFySQLAtxGMWQgX42H+KLh8EYBB7RswuH19Zqzaz4GTuf/hKKUqBn9/f3r16kW7du149tlnr1o/cOBAUlNTCQ4OZvLkyXTv3r3Ix3rnnXfYvHkzwcHBtGnThg8++KBA+w0bNozg4GBCQkK48cYbmTZtGvXr1y9yHHmRvC5bypPQ0FBTLIMO7VkOC8ZAqz/ByLng5kbchWT6/3sNzQKqs/jhnri76dOgSpWU3bt307p1a1eHUSHl9LMVkS3GmNDc9tF7O7NrNRgG/hP2fA+rXgagTs1qTLmtLX8cOcunvxx0cYBKKVV6NEnkpNvD0OVB+PVd2DwbgNs7NOSmVnV5K2wfJ89rF8ZKqcpBk0RORGDgVLimv3Vr7IFViAgv39aG1PR0/q1ddiilKglNErm56tbYCJr6V2dst6Ys3HREG7GVUpWCJom8ZLk1diRciGXijddQvWoVpq7QcSeUUhWfJon8ZLk19i78q6bxcL8WrNody8aoeFdHp5RSJUqTREE07ADDP4GYP2DeSB5oeIiGNT34x4o9eT75qJQqf5zpKhysTvouXrxYjBG5liaJgmo1GG6bASe2U23+cFa5Pcag4zNZu26tqyNTShUjVyeJonYNXtDtCqskO/ireDqPh+C7YN8PeG2bzwP7llPlf0tJ3xWMW8hd0H4E1Kjr6iiVqjhWTIITO4q3zPrtYdDUXFdn7yp8+vTpTJ8+nUWLFpGcnMywYcN45ZVXSExMZOTIkURHR5OWlsbkyZOJjY0lJiaGG264gYCAAFavXp2l7C1btvD000+TkJBAQEAAc+bMoUGDBvTr14+ePXvyyy+/MGTIEHbs2IGfnx9//PEHnTp14sUXX+S+++4jKioKb29vZs2aRXBw8FVdis+bN694f1Zokig8D09oOxRpO5QN23azctFMJl7cRMCPL0DYZGhxI4TcZT2x7eGVf3lKqTIle1fhYWFh7N+/n99++w1jDEOGDGHt2rXExcXRsGFDli1bBlh9Ovn6+vL222+zevVqAgICspSbkpLCxIkT+fbbb6lTpw4LFy7kxRdfZPZs61mss2fPsmbNGgDGjx/Pvn37WLVqFe7u7kycOJGOHTvyzTff8NNPP3HPPfdkxufYpXhJ0CThhF7BrZi5aQz9Y27l5wcaUWPPYti+EJbcD1VrQr02VlfktZteea8dBD6NwM3d1eErVfbl8Y2/tISFhREWFkbHjlZH1AkJCezfv58+ffrwzDPP8Le//Y1bb72VPn365FnO3r172blzJ/379weskeMaNGiQuX7UqFFZth8xYgTu7tbnxLp161iyZAkAN954I/Hx8Zw7dw64ukvx4qZJwgkiwvODWnPru+t4f6c7zw38O9w4GQ6vg11fw6n9cGQD7FwMxqGrcbcq4Bt4JXH4NALPWuBVCzx9r5728LIe8FNKlTpjDM8//zwPPfTQVeu2bNnC8uXLef755xkwYAAvv/xynuW0bduW9evX57i+sF2Di/2ZUNxdg2enScJJ7Rr5MrRDQz5Zd5C7ezSlga8XNOtrvTKkpcC5aDh7GM4chjOHrkzvXQGJcXkfxM3jStJwr2Zdhbi5W8km85VtXoTMcZ+yJBiH6Yzl4gZilynu4OZ29TJxu3IMd4+sx8pp3t3DirVKNYfpqg7Lqjq8e1rTenWlyoDsXYXfcsstTJ48mbFjx1KjRg2OHTuGh4cHqamp+Pn5MW7cOGrUqMGcOXOy7J+9uum6664jLi6O9evX06NHD1JSUti3bx9t2+Y/sFLfvn358ssvmTx5MuHh4QQEBODjUzoDoWmSKAZ/HXAdy3ec4O2wfUwfEXL1Bu4e4NfMemVz7mIK5xIv0aR6Klw6A0lnIekcXLLfk846TJ+DtMuQngbpqVdeJh1SLjssS7NeQJbBALN8GzFXlpl0MGmQbr+bdGt/Y5djjD3tcMyS4FblShKp4mkllYwEkmV5xrtnDsuqWVdeHl5QxevKdObL++p1mpyUA8euwgcNGsT06dPZvXs3PXr0AKBGjRp88cUXHDhwgGeffRY3Nzc8PDyYOXMmABMmTGDQoEE0aNAgS8N11apVWbx4MU888QTnzp0jNTWVv/zlLwVKElOmTOHee+8lODgYb29vPvvss5I5+RxoV+HF5I1lEXy87iArnuxDq/r5Z/jUtHS+2HCYt1fu43xSKn2uDeCR61vQo4V/5mVkmWWMQ6JKsd7TUq/Mp2W8kq331GRrOvWyleTSLmdblmzNpybZ7w7TGetSLtn7JWXb1uG9qMnLvdqVBFLVO2syyXzPviy37ez3qt7gUd1+99ZEVAjaVXjJKUpX4XolUUweu+EaFm46ytQVe5hzb94jRP164BSvLI1gb+wFel8TQNdmfny+/jBjPt5IcKAvD1/fglva1i+741aIWH1buVcBPF0dzRVpqVZSSUmC1EtWYkm5aL87zKcmWdOXE+3pizlse9EaeCox3l7usC4tufCxVfG0k1B1h4RUPWsCytgmS9LxdFjnlf+7u0fx/1xVpaZJopjU8q7K4zdewz+W7+GXA6fodU3AVdscPX2RfyzfzYqdJwis7cWHd3dmQJt6iAgT+jbnq9+PMWttJI9++TvNAqrzYJ/m3NGpEZ4e+i20QDISV9WSbcgjPS1rMnFMIpcd5i8nOixPvJJ4UhKvbHfxdLZEZW9r0vKPIyfiZiWMzLYe+z2zuq5aDtV3VbPtk30bhzakzO2ytSdlWWa/69VThaDVTcUoKSWNm95aQ+3qHnz3WG/c7CuBS5fTmLkmkg/XROImwqP9WvBg3+Y5fvinpRt+3HWCD9ZEsj36HHVqVuPeXkGM694UH0/9llhppKXkfPWT57t9BZWa7FA1Z79nn3esyssynZT1TjxniLudNOwbFjKnHV6ZNzZcWba7+X20at4EETfrqlUEcJwW+8aKjGl7Psfp7Ps6Tlcuxhj27Nmj1U2u5OnhzjO3tOSphdtYuj2GISENWb7jBG8siyDmXBK3hTTk+UGtaFgr93ua3d2Ewe0bMKhdfdZHxjNzTSTTftjL+6sjGdutCY/ecA2+XposKryMO8Q8S+cOliwyqu2ytA05JJmrkovDssw2p4z2J8d2qYx2Ksf1l60kd+ls5npPvwPEx1fH39vd+iw3hiw3YBSLHBJO9kRz1frcEk4O2zpul1lu9u1LL2EZY4iPj8fTs/DVwyV6JSEiA4H/A9yBj40xU7OtHw9MB47Zi94zxnxsr/sz8JK9/HVjTJ7N+WXhSgIgPd1w23vrOHsxhcDaXmw8eJrWDXyYclsbujX3L1KZO4+d48O1USzbHkPzOjX4dHwXGvt5F3PkSpUNKSkpREdHk5SUbQTIjGRhsN4xDnfsmavXZyYXh+2yLMu+X17bZltWrInLMWFlf7fXO05nJhWH7dzcraEN8uDp6UlgYCAeHlm/ZOZ3JVFiSUJE3IF9QH8gGtgEjDbGRDhsMx4INcY8nm1fP2AzEIr1m9gCdDbGnMnteGUlSQD8cuAUYz/eSG1vD/464DpGd21SLI3Q6yPjeWjuZqpWcePjP3ehQ+NaxRCtUqpI0tOvXC3leIWV27xj9V8ud+o5bpeWcuWqK3M6xeHKLQUahMB9K4p0Gk5XN9kf9lONMc8W8thdgQPGmCi7nAXA7UBEnntZbgFWGmNO2/uuBAYC8wsZg0v0uiaARQ/1oGW9GtTyrlps5fZo4c9Xj/bi3jm/cdes9cwY1ZGB7eoXW/lKqUJwcwM3T+sOtAos367CjTFpQGcp/M37jYCjDvPR9rLshovIdhFZLCKNC7OviEwQkc0isjkuLp+nlktZ12Z+xZogMlxTtwZfP9qLVvV9eOTLLXz8c5SOaaGUKjEFHU/iD+BbEblbRO7IeOWzT05JJfun2VIgyBgTDKwCMtodCrIvxphZxphQY0xonTp18gmn4gioUY0FE7ozsG19Xl+2m5e/3UVqWjHdkaKUUg4KmiT8gHjgRuA2+3VrPvtEA40d5gOBGMcNjDHxxpiMJ5M+AjoXdN/KztPDnf+M6cRDfZszd8NhJszdQmJyCXWXoZSqtAp0C6wx5t4ilL0JuFZEmmHdvXQXMMZxAxFpYIw5bs8OAXbb0z8C/xCR2vb8AOD5IsRQobm5Cc8Pbk0Tf29e/nYXIz5Yz+zxXajvW7HrSJVSpadAVxIiEigiX4vISRGJFZElIhKY1z7GmFTgcawP/N3AImPMLhF5VUSG2Js9ISK7RGQb8AQw3t73NPAaVqLZBLya0Yitrja2W1M++XMoh+MTGfb+L0TEnHd1SEqpCqJAt8DadxfNA+bai8YBY40x/UswtkIpS7fAukpEzHnum7OJC0kpzLirIze3rlv2OwtUSrlUfrfAFrRNoo4x5lNjTKr9mgNUnpbicqJNQx++eawXTf2r8+Dnmxn6/q8s236ctHS9+0kpVTQFTRKnRGSciLjbr3FYDdmqjKnv68lXj/bktaHtOHvxMo/N+50b3gzn8/WHuHhZG7aVUoVT0OqmJsB7QA+sW1F/BZ40xhwu2fAKTqubrpaWblgZcYIP10bxx5Gz1PL24J7uTbmnZxABNaq5OjylVBngdLcc9hPXTxhj/l3cwRUnTRK5M8aw5fAZPlwbxardsXi4uzG8UyAP9mlG8zo1XB2eUsqFiqXvJhEJN8b0K87AipsmiYKJjEvg458PsuT3aFLS0rm5dT3+flsbAmtrh4FKVUbFlSTeAHyBhUBixnJjzO/FEWRx0CRROHEXkpm7/hCf/nIIHy8PFkzorj3LKlUJFVeSWJ3DYmOMudGZ4IqTJomi2XnsHGM/3kiNalU0UShVCTl9C6yIuAEzjTE3ZHuVmQShiq5dI1++fKAbCcmpjPpwPUfiL7o6JKVUGVKQXmDTsZ6cVhVURqK4mJLGXbPWczg+Mf+dlFKVQkGfk1gpIs+ISGMR8ct4lWhkqlRlTRQbNFEopYCCJ4n7gMeAtVijxG3BGjlOVSBtG/oy74HuJKWkMerDDRw6pYlCqcquQEnCGNMsh1fzkg5Olb42DX2Y92B3Lqelc9esDRzURKFUpZZnkhCR5xymR2Rb94+SCkq5VusGPsx7sJudKNYTFZdQqP11pDylKo48b4EVkd+NMZ2yT+c072p6C2zx23viAmM+2oC7m7BgQverns42xhB7Ppndx88Tcfw8e05cYPfx8xw5fZFbgxvw/KDW1Kmp3X8oVZbldwtsfoMOSS7TOc2rCua6+jWZ92B3xny0gbtmbWD6iBDiLlhJIeN15mJK5vaNannRuoEPnZrU4us/jrEyIpbnbrmOMd2a4u6mfy5KlUd6JaHytS/WuqI4lXAZgGpV3GhVvyat6vvQukFNWjfwoVUDH3y9PDL3OXAygZe/3cmvkfG0b+TL60PbEdK4lqtOQSmVC6eeuBaRNKxuOATwAjKetBLA0xjjkdu+pU2TRMk6fu4SW4+c5dp6NWkWUL1AVwbGGJZuP87r30cQl5DM6K5NeO6W66jlXbUUIlZKFUSxdMtRHmiSKLsuJKXw75X7+Wz9IXy9PJg0qBV3dgrETauglHK54hqZTqkiq+npwcu3tWHp471pFlCd5xZvZ8SH69l9XMfiVqqsK9EkISIDRWSviBwQkUk5rH9aRCJEZLuI/E9EmjqsSxORrfbru5KMU5WONg19+O9DPZh2ZzAHTyVy67vr+Ofy3VxOTXd1aEqpXOR3d1OR2YMV/QfoD0QDm0TkO2NMhMNmfwChxpiLIvIIMA0YZa+7ZIzpUFLxKddwcxNGhjZmQJt6/OuHPXy4NorNh8/wnzGdqO/r6VTZxhi+2xZDxPHz9GoRQNdmfnh6uBdT5EpVTiXWJiEiPYApxphb7PnnAYwx/8xl+47Ae8aYXvZ8gjGmwMOmaZtE+bR0Wwx/W7Id76ruvDO6Iz1bBBSpnJMXknjhqx2s2n0SETAGPD3c6NHcn37X1eX6lnUICqhezNErVf45+5yEMxoBRx3mo4FueWx/P7DCYd5TRDYDqcBUY8w3xR+icrXbQhrSqn5NHv5iC+M+3shzA1vxUN/miBS8UXvpthgmf7uTS5fTmHxrG+7q0pjfDp5mzb44wveeZPXeXQAE+Xtzfcs69LuuLt2b++NVVa8ylMpPSV5JjABuMcY8YM/fDXQ1xkzMYdtxWN2RX2+MSbaXNTTGxIhIc+An4CZjTGS2/SYAEwCaNGnS+fDhwyVyLqrkJSSn8tzibSzfcYIBberx5sgQfDzzvsP6dOJlJn+7k2XbjxPSuBZvjQjhmrpXX3weOpWYmTDWR8WTlJJO1SpudG/uz1/7t9TnN1Sl5rJbYAta3SQiNwPvYiWIk7mUNQf43hizOLfjaXVT+WeM4ZN1B/nnij008fNm5rhOtKrvk+O2KyNief6rHZy7dJm/3NySh/o2p4p7/vdhJKWkZV5lfL89hviEyzw38Doe6N1cb8lVlZIrk0QVYB9wE3AM2ASMMcbsctimI7AYGGiM2e+wvDZw0RiTLCIBwHrg9myN3llokqg4fjt4msfm/c6FpBT+eUd7hnUMzFx3PimFV5dGsHhLNK0b+PDWiBDaNMw5keTn3MUU/rZkOz/sOsH1Levw1sgQAmpoX1OqcnHpw3QiMhiYAbgDs40xb4jIq8BmY8x38EPtMQAADr1JREFUIrIKaA8ct3c5YowZIiI9gQ+BdKzbdGcYYz7J61iaJCqWkxeSeHzeH/x28DR3d2/KS7e2ZmPUaf62ZDsnLyTzaL8WTLzxWqpWce4ubmMMX2w8wmvfR+Dr5cGMUR3odU3RGs+VKo/0iWtVbqWkpTP9x73MWhtFYG0vos9cokWd6rw9skOxtyPsPn6eifP/IDIugUf7teCpm1sWqPpKqfJOn7hW5ZaHuxsvDG7NzLGduJyazgO9m7HsiT4l0tDcuoEP3z3ei5GdG/Of1ZGMmrWB6DMX899RqQpOrySUyua7bTG88NUO3ASm3RnMwHYNXB2SUiVGrySUKqQhIQ1Z/kQfmgVU5+Evfuelb3aQlJLm6rCUcomSfJhOqXKrib83/324J2+F7eXDtVGs3hPHrcEN+FNwA9o38i3Uw35KlWda3aRUPn7eH8cn6w6ybv8pUtMNgbW9+FN75xJGalo6BqvdRSlX0rublComZ/+/vfsPrqo8Ezj+fZKQQH4n5BcEEAhSfgQIaNlWd6y/WrDbgt21FUe3rnVG17XVbncc207HrZ3tTu20u9Wp1tLqbKm11qko7IyLuIpWt1ZUCCSASEgpkITkhpCb3Py6ufc++8c5hEvMjblJbsK5eT4zmXvy3nPOfV/eyX047/ue53QH2XmwmRdrmkYcMNq7gxz1BTjq66Le10W9L8BRX4Djbd1MT0tlXWUZ11eV88mKmfaI1zHY8tYxWjr6uOeasS+LnmosSBiTALECxrVLS+kOhpyA0NpFW1dw4JhpqcJFM7NYWJRFRUk2rZ197Kg9RWdfiOKcDD6/cjYbq2azck58VyeRiPKXtm5qG/x09oYozc2gNHc6pbnTmZmVfsHdSX6mK0haqpDzEWlXRkJVeWjHYR5/3cnYUzU3n8duXsPs/BljPvdUYUHCmAQbHDDyM6exsCibipIsFhZls7A4i4XF2cwtmPGhey96+8O8+n4L26ob2PW+j2A4wsKiLDZUzWZjVTkLBmWujUSUY6e7qGnwU9vgp6bBz4GGDjr7QkPWbVqqUJIz/bzAUZY3nbKo7dLcDDLTEzs92dnbz84DzWzb18j/1bWSmZ7Kw5uquHpJ6ajPGYkoD2yv5ak/Hefmv5rHZRVF3P/cfqalCg9vWs0Vi4vHsQXJy4KEMRMoHNFRDxv5u/v5n9omtlU38qc/n0YVVs3JY11lGW2BIDUNfg42ngsI6WkpLC3LobI8jxXleVSW51GYlU5zR6/708epjl6a/b3Oq1sWGCKg5ExPo2wgaLhBxA0mcwpmsKAoK+5nc/T2h3ntcAvb9zXyyqEW+kIRyvNn8PlVs3njiI+DTR1849rF3H3VorivdkLhCPf9fj/P723gzk8t5JvrlyAi1PsC3PXUHj5o6eTr1yzma1fHf+6pxoKEMR7U5O/hv/c18sJe5yFKGWkpLJ2V6waDXCrL81hcmjOqie9AX4hT/t6BYBIdSE519NHs78UX6CMcOffdkCIwtzCTRcXZLCrJpqLEeV1Ukn1ett5QOMJb9afZVt3IS+5QWlF2On+zYhYbqspZMy8fEaEnGOZbW/fzQnUj65eX8aMvrSI7Y2RXM32hMF97ei87DzZz37qP8U9XVpw3PNcdDPGd52vZureBKxYX85MbqyjMSo/732mqsCBhjMe1dPZSkJk+oSuhwhHldMC5Ejne1s2R5gB1vgBHWwLU+7oIhs89crY0N4NFJdmU5EznjSM+WgNBcjLSWFdZxoZVs7msYuaQKU6is/4uLMpi85cv/dDw2mDdwRB3bHmPN+taeXDDcm69bP6Q+6kqT+8+zoPbD1Kck8GjN6+hylLCD8mChDFmXIXCEU6c6aGuJRD100lDey9rFxSwYdVsrvxYyYiHp/5Y18rdT+8hFFEe2bSaq5aUDLmfv6efr/zXO+w9foYf3rCKGy6ZM+R+0fafbOeup/bQ0tnLA59bxi2fuMjucRnEgoQx5oJ38kw3d/76PQ42dfAvn3bmKaK/zFsDfXz5id0caenkkU2ruW7FyFOltHcH+effVbPrsI/rq2bz73+7IuET9V5iQcIY4wmx5ika23u45Ym3aWzv4ed/fymfGsWqpUhEeey1On788gdUFGdzwyVzWFmex/LyPPJmjH0prpdZkDDGeMbgeYrvfG4Z395aQ0dPP0/e9nE+Pr9wTOd/80grD2yrpb61a6Bs/sxMKsvzWDnHWSFWWZ73kY/OTSYWJIwxnnN2nuJMdz+FWels+cpaKsvzxu38Z7qcJcU1DX5qTjqvDe09A+8vKMpylxbnTljgCIUjHG7uZO/xdqpPtFPvC7BmXgHrK8tYM68gYUt5LUgYYzzpRFs3j79+lNsun8+ikpyEf97pQB+1jR3UnGynpsHP/pN+mvy9A++fveIYuC9ldh55maMLHKpKk7+X6hNOQKg+7nxmj5ttuDArnfkzM6lt6CAYjlCck8G65aVcVzmLtQsKx3WlmwUJY4wZpdZAH7VRd7fXNnScd8UxrzCTFeV5LCnLIT0thbAqkYgSjnBue6DM2W4400P1iXZaOvsASE9NYXl5LlVz86mam8/quQXMLZyBiNDZ28+r77fw0oFT7HrfR09/mPzMaVy7tJTrKsu4fFFR3Dc5DmZBwhhjxlFbVzAqaPipbfRzoq1nyH1TBFJTBBEhVYTUFKEoO30gIFTNK2DprBwy0j76i74nGOYPR3zsqD3F/x5qprM3RHZGGlctKeGzlWVxrfiKZkHCGGMSrCcYJqJOSpYUNxikCAm7JyMYcu5s31HbxM4DzVQUZ/PsP35yVOea1CAhIuuBh4FU4Jeq+oNB72cAW4BLgNPAjap6zH3vW8DtQBi4R1VfGu6zLEgYY6ais3fHl+ROH9Xxk/b4UhFJBR4FrgOWATeJyLJBu90OnFHVRcB/Ag+5xy4DNgHLgfXAY+75jDHGRElNkVEHiJFIZDKYtUCdqtarahB4Btg4aJ+NwK/c7d8D14hzfbYReEZV+1T1z0Cdez5jjDETKJFBohw4EfX7SbdsyH1UNQT4gZkjPBYRuUNE3hWRd30+3zhW3RhjDEAiE5gMNWMzeAIk1j4jORZV3QxsBhARn4j8Jd5KRikCWsdw/IUm2doDydemZGsPJF+bkq098OE2XTTczokMEieBuVG/zwEaY+xzUkTSgDygbYTHnkdVx/QYKhF5d7jJG69JtvZA8rUp2doDydemZGsPxN+mRA43vQNcLCILRCQdZyJ6+6B9tgO3uts3AK+qs9xqO7BJRDJEZAFwMbA7gXU1xhgzhIRdSahqSES+CryEswT2SVU9ICLfA95V1e3AE8CvRaQO5wpik3vsARF5FjgIhIC7VTWcqLoaY4wZWkKTqqvqi8CLg8oeiNruBb4Y49jvA99PZP0G2TyBnzURkq09kHxtSrb2QPK1KdnaA3G2KWnuuDbGGDP+Ju6hucYYYzzHgoQxxpiYpnyQEJH1InJYROpE5JuTXZ/xICLHRKRGRKpFxHMJrUTkSRFpEZHaqLJCEXlZRI64rwWTWcd4xWjTd0Wkwe2nahH57GTWMR4iMldEdonIIRE5ICL3uuWe7Kdh2uPlPpouIrtFZJ/bpgfd8gUi8rbbR79zV5/GPs9UnpNw80F9AHwa596Md4CbVPXgpFZsjETkGHCpqnryJiARuQIIAFtUtdIt+yHQpqo/cIN5gareP5n1jEeMNn0XCKjqjyazbqMhIrOAWaq6R0RygPeA64F/wIP9NEx7voR3+0iALFUNiMg04E3gXuAbwFZVfUZEHgf2qerPYp1nql9JjCS/lJlgqvoHnCXR0aLzfP0K5w/YM2K0ybNUtUlV97jbncAhnNQ5nuynYdrjWeoIuL9Oc38UuBonVx6MoI+mepAYUY4oD1Jgp4i8JyJ3THZlxkmpqjaB8wcNlExyfcbLV0Vkvzsc5YmhmcFEZD6wGnibJOinQe0BD/eRiKSKSDXQArwMHAXa3Vx5MILvvKkeJEaUI8qDLlfVNThp2u92hzrMhednQAVQBTQBP57c6sRPRLKB54Cvq2rHZNdnrIZoj6f7SFXDqlqFk9poLbB0qN2GO8dUDxJx54jyAlVtdF9bgOdJjjTrze648dnx45ZJrs+YqWqz+0ccAX6Bx/rJHed+DviNqm51iz3bT0O1x+t9dJaqtgOvAZ8A8t1ceTCC77ypHiRGkl/KU0Qky514Q0SygM8AtcMf5QnReb5uBbZNYl3GxdkvU9cX8FA/uZOiTwCHVPU/ot7yZD/Fao/H+6hYRPLd7RnAtThzLbtwcuXBCPpoSq9uAnCXtP2Ec/mlJjIVyLgTkYU4Vw/gpF152mttEpHfAlfipDRuBv4VeAF4FpgHHAe+qKqemQiO0aYrcYYxFDgG3Hl2PP9CJyJ/DbwB1AARt/jbOOP4nuunYdpzE97to5U4E9OpOBcEz6rq99zviGeAQmAvcIuq9sU8z1QPEsYYY2Kb6sNNxhhjhmFBwhhjTEwWJIwxxsRkQcIYY0xMFiSMMcbEZEHCmDiISDgqI2j1eGYOFpH50VlijbkQJPTxpcYkoR43zYExU4JdSRgzDtxneDzk5u/fLSKL3PKLROQVN0HcKyIyzy0vFZHn3Vz/+0TkMvdUqSLyCzf//073TlljJo0FCWPiM2PQcNONUe91qOpa4Kc4d/Hjbm9R1ZXAb4BH3PJHgNdVdRWwBjjgll8MPKqqy4F24O8S3B5jhmV3XBsTBxEJqGr2EOXHgKtVtd5NFHdKVWeKSCvOw2z63fImVS0SER8wJzodgpui+mVVvdj9/X5gmqr+W+JbZszQ7ErCmPGjMbZj7TOU6Bw6YWze0EwyCxLGjJ8bo17fcrf/iJNdGOBmnEdIArwC3AUDD4bJnahKGhMP+1+KMfGZ4T7p66wdqnp2GWyGiLyN85+vm9yye4AnReQ+wAfc5pbfC2wWkdtxrhjuwnmojTEXFJuTMGYcuHMSl6pq62TXxZjxZMNNxhhjYrIrCWOMMTHZlYQxxpiYLEgYY4yJyYKEMcaYmCxIGGOMicmChDHGmJj+H9IQAsS3f5BvAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7/7 - 0s - loss: 0.5612 - accuracy: 0.8571\n",
      "\n",
      "Test accuracy: 0.85714287\n",
      "Target: 1, Predicted label: [1]\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "import tensorflow.keras as keras\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "DATA_PATH = \"breath_sound.json\"\n",
    "\n",
    "\n",
    "def load_data(data_path):\n",
    "    \"\"\"Loads training dataset from json file.\n",
    "\n",
    "        :param data_path (str): Path to json file containing data\n",
    "        :return X (ndarray): Inputs\n",
    "        :return y (ndarray): Targets\n",
    "    \"\"\"\n",
    "\n",
    "    with open(data_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "\n",
    "    X = np.array(data[\"mfcc\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "    return X, y\n",
    "\n",
    "\n",
    "def plot_history(history):\n",
    "    \"\"\"Plots accuracy/loss for training/validation set as a function of the epochs\n",
    "\n",
    "        :param history: Training history of model\n",
    "        :return:\n",
    "    \"\"\"\n",
    "\n",
    "    fig, axs = plt.subplots(2)\n",
    "\n",
    "    # create accuracy sublpot\n",
    "    axs[0].plot(history.history[\"accuracy\"], label=\"train accuracy\")\n",
    "    axs[0].plot(history.history[\"val_accuracy\"], label=\"test accuracy\")\n",
    "    axs[0].set_ylabel(\"Accuracy\")\n",
    "    axs[0].legend(loc=\"lower right\")\n",
    "    axs[0].set_title(\"Accuracy eval\")\n",
    "\n",
    "    # create error sublpot\n",
    "    axs[1].plot(history.history[\"loss\"], label=\"train error\")\n",
    "    axs[1].plot(history.history[\"val_loss\"], label=\"test error\")\n",
    "    axs[1].set_ylabel(\"Error\")\n",
    "    axs[1].set_xlabel(\"Epoch\")\n",
    "    axs[1].legend(loc=\"upper right\")\n",
    "    axs[1].set_title(\"Error eval\")\n",
    "\n",
    "    plt.show()\n",
    "\n",
    "\n",
    "def prepare_datasets(test_size, validation_size):\n",
    "    \"\"\"Loads data and splits it into train, validation and test sets.\n",
    "\n",
    "    :param test_size (float): Value in [0, 1] indicating percentage of data set to allocate to test split\n",
    "    :param validation_size (float): Value in [0, 1] indicating percentage of train set to allocate to validation split\n",
    "\n",
    "    :return X_train (ndarray): Input training set\n",
    "    :return X_validation (ndarray): Input validation set\n",
    "    :return X_test (ndarray): Input test set\n",
    "    :return y_train (ndarray): Target training set\n",
    "    :return y_validation (ndarray): Target validation set\n",
    "    :return y_test (ndarray): Target test set\n",
    "    \"\"\"\n",
    "\n",
    "    # load data\n",
    "    X, y = load_data(DATA_PATH)\n",
    "\n",
    "    # create train, validation and test split\n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)\n",
    "    X_train, X_validation, y_train, y_validation = train_test_split(X_train, y_train, test_size=validation_size)\n",
    "\n",
    "    # add an axis to input sets\n",
    "    X_train = X_train[..., np.newaxis]\n",
    "    print(X_train.shape)\n",
    "    X_validation = X_validation[..., np.newaxis]\n",
    "    print(X_validation.shape)\n",
    "    X_test = X_test[..., np.newaxis]\n",
    "    print(X_test.shape)\n",
    "\n",
    "    return X_train, X_validation, X_test, y_train, y_validation, y_test\n",
    "\n",
    "\n",
    "def build_model(input_shape):\n",
    "    \"\"\"Generates CNN model\n",
    "\n",
    "    :param input_shape (tuple): Shape of input set\n",
    "    :return model: CNN model\n",
    "    \"\"\"\n",
    "\n",
    "    # build network topology\n",
    "    model = keras.Sequential()\n",
    "\n",
    "    # 1st conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=input_shape))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 2nd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 3rd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (2, 2), activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((2, 2), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # flatten output and feed it into dense layer\n",
    "    model.add(keras.layers.Flatten())\n",
    "    model.add(keras.layers.Dense(64, activation='relu'))\n",
    "    model.add(keras.layers.Dropout(0.3))\n",
    "\n",
    "    # output layer\n",
    "    model.add(keras.layers.Dense(2, activation='sigmoid'))\n",
    "    # is me problem hau blc sigmoid hamesha ak hi single output deta hai so dense should be 1. \n",
    "\n",
    "    return model\n",
    "\n",
    "\n",
    "def predict(model, X, y):\n",
    "    \"\"\"Predict a single sample using the trained model\n",
    "\n",
    "    :param model: Trained classifier\n",
    "    :param X: Input data\n",
    "    :param y (int): Target\n",
    "    \"\"\"\n",
    "\n",
    "    # add a dimension to input data for sample - model.predict() expects a 4d array in this case\n",
    "    X = X[np.newaxis, ...] # array shape (1, 130, 13, 1)\n",
    "\n",
    "    # perform prediction\n",
    "    prediction = model.predict(X)\n",
    "\n",
    "    # get index with max value\n",
    "    predicted_index = np.argmax(prediction, axis=1)\n",
    "\n",
    "    print(\"Target: {}, Predicted label: {}\".format(y, predicted_index))\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    # get train, validation, test splits\n",
    "    X_train, X_validation, X_test, y_train, y_validation, y_test = prepare_datasets(0.25, 0.2)\n",
    "\n",
    "    # create network\n",
    "    input_shape = (X_train.shape[1], X_train.shape[2], 1)\n",
    "    model = build_model(input_shape)\n",
    "\n",
    "    # compile model\n",
    "    optimizer = keras.optimizers.Adam(learning_rate=0.0001)\n",
    "    model.compile(optimizer=optimizer,\n",
    "                  loss='sparse_categorical_crossentropy',\n",
    "                  metrics=['accuracy'])\n",
    "    \n",
    "\n",
    "    model.summary()\n",
    "\n",
    "    # train model\n",
    "    history = model.fit(X_train, y_train, validation_data=(X_validation, y_validation), batch_size=32, epochs=30)\n",
    "\n",
    "    # plot accuracy/error for training and validation\n",
    "    plot_history(history)\n",
    "\n",
    "    # evaluate model on test set\n",
    "    test_loss, test_acc = model.evaluate(X_test, y_test, verbose=2)\n",
    "    print('\\nTest accuracy:', test_acc)\n",
    "\n",
    "    # pick a sample to predict from the test set\n",
    "    X_to_predict = X_test[0]\n",
    "    y_to_predict = y_test[0]\n",
    "\n",
    "    # predict sample\n",
    "    predict(model, X_to_predict, y_to_predict)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(14, 197, 13, 1)\n",
      "(4, 197, 13, 1)\n",
      "(7, 197, 13, 1)\n",
      "Target: 1, Predicted label: [1]\n"
     ]
    }
   ],
   "source": [
    "X_train, X_validation, X_test, y_train, y_validation, y_test = prepare_datasets(0.25, 0.2)\n",
    "\n",
    "# create network\n",
    "input_shape = (X_train.shape[1], X_train.shape[2], 1)\n",
    "model = build_model(input_shape)\n",
    "\n",
    "X_to_predict = X_test[6]\n",
    "y_to_predict = y_test[6]\n",
    "\n",
    "predict(model, X_to_predict, y_to_predict)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tensorflow_env",
   "language": "python",
   "name": "tensorflow_env"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}