build-model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "467e7549",
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "6d94e0b9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<module 'keras.api._v2.keras' from 'C:\\\\Users\\\\youne\\\\.conda\\\\envs\\\\dl\\\\lib\\\\site-packages\\\\keras\\\\api\\\\_v2\\\\keras\\\\__init__.py'>"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.keras"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "c529c1b5",
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_PATH = \"data.json\"\n",
    "SAVED_MODEL_PATH = \"model.h5\"\n",
    "EPOCHS = 40\n",
    "BATCH_SIZE = 32\n",
    "PATIENCE = 5\n",
    "LEARNING_RATE = 0.0001"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "9eab14bc",
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_data(data_path):\n",
    "    \"\"\"Loads training dataset from json file.\n",
    "    :param data_path (str): Path to json file containing data\n",
    "    :return X (ndarray): Inputs\n",
    "    :return y (ndarray): Targets\n",
    "    \"\"\"\n",
    "    with open(data_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "\n",
    "    X = np.array(data[\"MFCCs\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "    print(\"Training sets loaded!\")\n",
    "    return X, y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "aed67d95",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "def prepare_dataset(data_path, test_size=0.2, validation_size=0.2):\n",
    "    \"\"\"Creates train, validation and test sets.\n",
    "    :param data_path (str): Path to json file containing data\n",
    "    :param test_size (flaot): Percentage of dataset used for testing\n",
    "    :param validation_size (float): Percentage of train set used for cross-validation\n",
    "    :return X_train (ndarray): Inputs for the train set\n",
    "    :return y_train (ndarray): Targets for the train set\n",
    "    :return X_validation (ndarray): Inputs for the validation set\n",
    "    :return y_validation (ndarray): Targets for the validation set\n",
    "    :return X_test (ndarray): Inputs for the test set\n",
    "    :return X_test (ndarray): Targets for the test set\n",
    "    \"\"\"\n",
    "\n",
    "    # load dataset\n",
    "    X, y = load_data(data_path)\n",
    "\n",
    "    # create train, validation, 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 nd array\n",
    "    X_train = X_train[..., np.newaxis]\n",
    "    X_test = X_test[..., np.newaxis]\n",
    "    X_validation = X_validation[..., np.newaxis]\n",
    "\n",
    "    return X_train, y_train, X_validation, y_validation, X_test, y_test\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "7cd07fb1",
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_model(input_shape, loss=\"sparse_categorical_crossentropy\", learning_rate=0.0001):\n",
    "    \"\"\"Build neural network using keras.\n",
    "    :param input_shape (tuple): Shape of array representing a sample train. E.g.: (44, 13, 1)\n",
    "    :param loss (str): Loss function to use\n",
    "    :param learning_rate (float):\n",
    "    :return model: TensorFlow model\n",
    "    \"\"\"\n",
    "\n",
    "    # build network architecture using convolutional layers\n",
    "    model = tf.keras.models.Sequential()\n",
    "\n",
    "    # 1st conv layer\n",
    "    model.add(tf.keras.layers.Conv2D(64, (3, 3), activation='relu', input_shape=input_shape,\n",
    "                                     kernel_regularizer=tf.keras.regularizers.l2(0.001)))\n",
    "    model.add(tf.keras.layers.BatchNormalization())\n",
    "    model.add(tf.keras.layers.MaxPooling2D((3, 3), strides=(2,2), padding='same'))\n",
    "\n",
    "    # 2nd conv layer\n",
    "    model.add(tf.keras.layers.Conv2D(32, (3, 3), activation='relu',\n",
    "                                     kernel_regularizer=tf.keras.regularizers.l2(0.001)))\n",
    "    model.add(tf.keras.layers.BatchNormalization())\n",
    "    model.add(tf.keras.layers.MaxPooling2D((3, 3), strides=(2,2), padding='same'))\n",
    "\n",
    "    # 3rd conv layer\n",
    "    model.add(tf.keras.layers.Conv2D(32, (2, 2), activation='relu',\n",
    "                                     kernel_regularizer=tf.keras.regularizers.l2(0.001)))\n",
    "    model.add(tf.keras.layers.BatchNormalization())\n",
    "    model.add(tf.keras.layers.MaxPooling2D((2, 2), strides=(2,2), padding='same'))\n",
    "\n",
    "    # flatten output and feed into dense layer\n",
    "    model.add(tf.keras.layers.Flatten())\n",
    "    model.add(tf.keras.layers.Dense(64, activation='relu'))\n",
    "    tf.keras.layers.Dropout(0.3)\n",
    "\n",
    "    # softmax output layer\n",
    "    model.add(tf.keras.layers.Dense(10, activation='softmax'))\n",
    "\n",
    "    optimiser = tf.optimizers.Adam(learning_rate=learning_rate)\n",
    "\n",
    "    # compile model\n",
    "    model.compile(optimizer=optimiser,\n",
    "                  loss=loss,\n",
    "                  metrics=[\"accuracy\"])\n",
    "\n",
    "    # print model parameters on console\n",
    "    model.summary()\n",
    "\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "ae2fc31f",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "def train(model, epochs, batch_size, patience, X_train, y_train, X_validation, y_validation):\n",
    "    \"\"\"Trains model\n",
    "    :param epochs (int): Num training epochs\n",
    "    :param batch_size (int): Samples per batch\n",
    "    :param patience (int): Num epochs to wait before early stop, if there isn't an improvement on accuracy\n",
    "    :param X_train (ndarray): Inputs for the train set\n",
    "    :param y_train (ndarray): Targets for the train set\n",
    "    :param X_validation (ndarray): Inputs for the validation set\n",
    "    :param y_validation (ndarray): Targets for the validation set\n",
    "    :return history: Training history\n",
    "    \"\"\"\n",
    "\n",
    "    earlystop_callback = tf.keras.callbacks.EarlyStopping(monitor=\"accuracy\", min_delta=0.001, patience=patience)\n",
    "\n",
    "    # train model\n",
    "    history = model.fit(X_train,\n",
    "                        y_train,\n",
    "                        epochs=epochs,\n",
    "                        batch_size=batch_size,\n",
    "                        validation_data=(X_validation, y_validation),\n",
    "                        callbacks=[earlystop_callback])\n",
    "    return history\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "9dd2bc00",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_history(history):\n",
    "    \"\"\"Plots accuracy/loss for training/validation set as a function of the epochs\n",
    "    :param history: Training history of model\n",
    "    :return:\n",
    "    \"\"\"\n",
    "\n",
    "    fig, axs = plt.subplots(2)\n",
    "\n",
    "    # create accuracy subplot\n",
    "    axs[0].plot(history.history[\"accuracy\"], label=\"accuracy\")\n",
    "    axs[0].plot(history.history['val_accuracy'], label=\"val_accuracy\")\n",
    "    axs[0].set_ylabel(\"Accuracy\")\n",
    "    axs[0].legend(loc=\"lower right\")\n",
    "    axs[0].set_title(\"Accuracy evaluation\")\n",
    "\n",
    "    # create loss subplot\n",
    "    axs[1].plot(history.history[\"loss\"], label=\"loss\")\n",
    "    axs[1].plot(history.history['val_loss'], label=\"val_loss\")\n",
    "    axs[1].set_xlabel(\"Epoch\")\n",
    "    axs[1].set_ylabel(\"Loss\")\n",
    "    axs[1].legend(loc=\"upper right\")\n",
    "    axs[1].set_title(\"Loss evaluation\")\n",
    "\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "ae6ae3a1",
   "metadata": {},
   "outputs": [],
   "source": [
    "def main():\n",
    "    # generate train, validation and test sets\n",
    "    X_train, y_train, X_validation, y_validation, X_test, y_test = prepare_dataset(DATA_PATH)\n",
    "\n",
    "    # create network\n",
    "    input_shape = (X_train.shape[1], X_train.shape[2], X_train.shape[3])\n",
    "    model = build_model(input_shape, learning_rate=LEARNING_RATE)\n",
    "\n",
    "    # train network\n",
    "    history = train(model, EPOCHS, BATCH_SIZE, PATIENCE, X_train, y_train, X_validation, y_validation)\n",
    "\n",
    "    # plot accuracy/loss for training/validation set as a function of the epochs\n",
    "    plot_history(history)\n",
    "\n",
    "    # evaluate network on test set\n",
    "    test_loss, test_acc = model.evaluate(X_test, y_test)\n",
    "    print(\"\\nTest loss: {}, test accuracy: {}\".format(test_loss, 100*test_acc))\n",
    "\n",
    "    # save model\n",
    "    model.save(SAVED_MODEL_PATH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "88b3e0be",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training sets loaded!\n",
      "Model: \"sequential_4\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_10 (Conv2D)           (None, 42, 11, 64)        640       \n",
      "_________________________________________________________________\n",
      "batch_normalization_9 (Batch (None, 42, 11, 64)        256       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_9 (MaxPooling2 (None, 21, 6, 64)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_11 (Conv2D)           (None, 19, 4, 32)         18464     \n",
      "_________________________________________________________________\n",
      "batch_normalization_10 (Batc (None, 19, 4, 32)         128       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_10 (MaxPooling (None, 10, 2, 32)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_12 (Conv2D)           (None, 9, 1, 32)          4128      \n",
      "_________________________________________________________________\n",
      "batch_normalization_11 (Batc (None, 9, 1, 32)          128       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_11 (MaxPooling (None, 5, 1, 32)          0         \n",
      "_________________________________________________________________\n",
      "flatten_3 (Flatten)          (None, 160)               0         \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 64)                10304     \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 10)                650       \n",
      "=================================================================\n",
      "Total params: 34,698\n",
      "Trainable params: 34,442\n",
      "Non-trainable params: 256\n",
      "_________________________________________________________________\n",
      "Epoch 1/40\n",
      "427/427 [==============================] - 10s 22ms/step - loss: 2.1560 - accuracy: 0.2779 - val_loss: 1.7501 - val_accuracy: 0.4185\n",
      "Epoch 2/40\n",
      "427/427 [==============================] - 9s 20ms/step - loss: 1.4826 - accuracy: 0.5230 - val_loss: 1.2685 - val_accuracy: 0.6012\n",
      "Epoch 3/40\n",
      "427/427 [==============================] - 8s 19ms/step - loss: 1.1284 - accuracy: 0.6552 - val_loss: 1.0122 - val_accuracy: 0.6874\n",
      "Epoch 4/40\n",
      "427/427 [==============================] - 9s 21ms/step - loss: 0.9130 - accuracy: 0.7274 - val_loss: 0.8272 - val_accuracy: 0.7534\n",
      "Epoch 5/40\n",
      "427/427 [==============================] - 9s 21ms/step - loss: 0.7675 - accuracy: 0.7778 - val_loss: 0.7267 - val_accuracy: 0.7845\n",
      "Epoch 6/40\n",
      "427/427 [==============================] - 8s 19ms/step - loss: 0.6698 - accuracy: 0.8098 - val_loss: 0.6485 - val_accuracy: 0.8100\n",
      "Epoch 7/40\n",
      "427/427 [==============================] - 9s 21ms/step - loss: 0.5985 - accuracy: 0.8350 - val_loss: 0.5825 - val_accuracy: 0.8349\n",
      "Epoch 8/40\n",
      "427/427 [==============================] - 10s 23ms/step - loss: 0.5508 - accuracy: 0.8495 - val_loss: 0.5299 - val_accuracy: 0.8481\n",
      "Epoch 9/40\n",
      "427/427 [==============================] - 10s 23ms/step - loss: 0.5032 - accuracy: 0.8657 - val_loss: 0.5086 - val_accuracy: 0.8610\n",
      "Epoch 10/40\n",
      "427/427 [==============================] - 8s 20ms/step - loss: 0.4683 - accuracy: 0.8781 - val_loss: 0.4900 - val_accuracy: 0.8648\n",
      "Epoch 11/40\n",
      "427/427 [==============================] - 10s 23ms/step - loss: 0.4350 - accuracy: 0.8828 - val_loss: 0.4520 - val_accuracy: 0.8830\n",
      "Epoch 12/40\n",
      "427/427 [==============================] - 10s 22ms/step - loss: 0.4108 - accuracy: 0.8935 - val_loss: 0.4317 - val_accuracy: 0.8900\n",
      "Epoch 13/40\n",
      "427/427 [==============================] - 8s 20ms/step - loss: 0.3899 - accuracy: 0.9018 - val_loss: 0.4588 - val_accuracy: 0.8751\n",
      "Epoch 14/40\n",
      "427/427 [==============================] - 8s 19ms/step - loss: 0.3744 - accuracy: 0.9042 - val_loss: 0.4168 - val_accuracy: 0.8839\n",
      "Epoch 15/40\n",
      "427/427 [==============================] - 8s 19ms/step - loss: 0.3523 - accuracy: 0.9111 - val_loss: 0.3911 - val_accuracy: 0.8991\n",
      "Epoch 16/40\n",
      "427/427 [==============================] - 8s 20ms/step - loss: 0.3411 - accuracy: 0.9146 - val_loss: 0.3751 - val_accuracy: 0.8991\n",
      "Epoch 17/40\n",
      "427/427 [==============================] - 8s 20ms/step - loss: 0.3311 - accuracy: 0.9169 - val_loss: 0.3659 - val_accuracy: 0.9073\n",
      "Epoch 18/40\n",
      "427/427 [==============================] - 9s 22ms/step - loss: 0.3128 - accuracy: 0.9240 - val_loss: 0.3572 - val_accuracy: 0.9085\n",
      "Epoch 19/40\n",
      "427/427 [==============================] - 8s 20ms/step - loss: 0.3024 - accuracy: 0.9277 - val_loss: 0.3500 - val_accuracy: 0.9114\n",
      "Epoch 20/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.2909 - accuracy: 0.9298 - val_loss: 0.3582 - val_accuracy: 0.9056\n",
      "Epoch 21/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.2783 - accuracy: 0.9332 - val_loss: 0.3505 - val_accuracy: 0.9067\n",
      "Epoch 22/40\n",
      "427/427 [==============================] - 8s 19ms/step - loss: 0.2725 - accuracy: 0.9363 - val_loss: 0.3495 - val_accuracy: 0.9097\n",
      "Epoch 23/40\n",
      "427/427 [==============================] - 8s 19ms/step - loss: 0.2685 - accuracy: 0.9347 - val_loss: 0.3336 - val_accuracy: 0.9155\n",
      "Epoch 24/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.2531 - accuracy: 0.9402 - val_loss: 0.3397 - val_accuracy: 0.9100\n",
      "Epoch 25/40\n",
      "427/427 [==============================] - 8s 19ms/step - loss: 0.2500 - accuracy: 0.9435 - val_loss: 0.3159 - val_accuracy: 0.9170\n",
      "Epoch 26/40\n",
      "427/427 [==============================] - 10s 23ms/step - loss: 0.2407 - accuracy: 0.9467 - val_loss: 0.3347 - val_accuracy: 0.9111\n",
      "Epoch 27/40\n",
      "427/427 [==============================] - 9s 21ms/step - loss: 0.2333 - accuracy: 0.9477 - val_loss: 0.3686 - val_accuracy: 0.9038\n",
      "Epoch 28/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.2272 - accuracy: 0.9498 - val_loss: 0.3225 - val_accuracy: 0.9167\n",
      "Epoch 29/40\n",
      "427/427 [==============================] - 9s 20ms/step - loss: 0.2198 - accuracy: 0.9519 - val_loss: 0.3278 - val_accuracy: 0.9132\n",
      "Epoch 30/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.2159 - accuracy: 0.9519 - val_loss: 0.3364 - val_accuracy: 0.9161\n",
      "Epoch 31/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.2112 - accuracy: 0.9533 - val_loss: 0.3181 - val_accuracy: 0.9208\n",
      "Epoch 32/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.2058 - accuracy: 0.9547 - val_loss: 0.3164 - val_accuracy: 0.9176\n",
      "Epoch 33/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.1988 - accuracy: 0.9587 - val_loss: 0.3507 - val_accuracy: 0.9076\n",
      "Epoch 34/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.1948 - accuracy: 0.9589 - val_loss: 0.3434 - val_accuracy: 0.9082\n",
      "Epoch 35/40\n",
      "427/427 [==============================] - 7s 17ms/step - loss: 0.1883 - accuracy: 0.9613 - val_loss: 0.3008 - val_accuracy: 0.9267\n",
      "Epoch 36/40\n",
      "427/427 [==============================] - 7s 18ms/step - loss: 0.1811 - accuracy: 0.9641 - val_loss: 0.3176 - val_accuracy: 0.9220\n",
      "Epoch 37/40\n",
      "427/427 [==============================] - 8s 18ms/step - loss: 0.1843 - accuracy: 0.9606 - val_loss: 0.3120 - val_accuracy: 0.9223\n",
      "Epoch 38/40\n",
      "427/427 [==============================] - 7s 17ms/step - loss: 0.1738 - accuracy: 0.9661 - val_loss: 0.3170 - val_accuracy: 0.9208\n",
      "Epoch 39/40\n",
      "427/427 [==============================] - 7s 17ms/step - loss: 0.1729 - accuracy: 0.9660 - val_loss: 0.2964 - val_accuracy: 0.9267\n",
      "Epoch 40/40\n",
      "427/427 [==============================] - 8s 19ms/step - loss: 0.1723 - accuracy: 0.9651 - val_loss: 0.3200 - val_accuracy: 0.9167\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": [
      "134/134 [==============================] - 1s 9ms/step - loss: 0.3317 - accuracy: 0.9146\n",
      "\n",
      "Test loss: 0.3317473530769348, test accuracy: 91.46141409873962\n"
     ]
    }
   ],
   "source": [
    "main()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5f0e5f08",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8c598102",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8ed42ced",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}