experiment_augmentation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# This notebook builds a model using Bayesian optimization and mel-spectrograms as input feature to the model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from methods_audio import data_handling, model_performance_training"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 1. Get data (file names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = data_handling.get_data()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 2. Read data (transforming file names into waves) <br>\n",
    "Additionally, the mean is removed and the data is normalized. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = data.map(data_handling.read_in_data)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 3. Get input for model training "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "samples, labels = data_handling.extract_samples_labels(data)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Set arguments from baseline: "
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 4. Denoising (on train and validation set)\n",
    "### 5. Transform data into Spectogram\n",
    "### 6. Set batch size \n",
    "### 7. Choose model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define type of denoising to do. Options: spectral_gating, low_pass, none\n",
    "type_denoising = \"none\"\n",
    "differentiation = False\n",
    "\n",
    "# For low pass filter: not used\n",
    "low_pass_cutoff = 0\n",
    "low_pass_order = 0\n",
    "\n",
    "# Define type of transformation to do. Options = spectrogram, mel_spectrogram, db_mel_spectrogram, mfcc, mfcc_delta\n",
    "type_transformation = \"spectrogram\"\n",
    "\n",
    "batch_size = 8\n",
    "epoch = 30\n",
    "\n",
    "number_model = 1"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 8. Select type of augmentation "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "type_augmentation = \"signal\"  # Options: {signal, none}\n",
    "probability = 0.3  # Options tested: {0, 0.3, 0.5, 0.7, 0.9, 1}"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 9. Evaluate performance of the model by training it with K-Fold Validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "confusion_matrices, histories = model_performance_training.train_performance_k_fold(\n",
    "    number_model,\n",
    "    samples,\n",
    "    labels,\n",
    "    epoch,\n",
    "    batch_size,\n",
    "    type_augmentation,\n",
    "    type_denoising,\n",
    "    differentiation,\n",
    "    low_pass_cutoff,\n",
    "    low_pass_order,\n",
    "    type_transformation,\n",
    "    probability,\n",
    ")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 10. Save results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "(\n",
    "    list_loss,\n",
    "    list_val_loss,\n",
    "    list_precision,\n",
    "    list_val_precision,\n",
    "    list_recall,\n",
    "    list_val_recall,\n",
    "    list_accuracy,\n",
    "    list_val_accuracy,\n",
    ") = model_performance_training.get_metrics(epoch, histories)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "if type_augmentation == \"none\":\n",
    "    temp_title = \"No data augmentation applied\"\n",
    "else:\n",
    "    temp_title = \"Data augmentation with probability = \" + str(probability)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "general_path = \"experiments_results/2_data_augmentation/\"\n",
    "\n",
    "specific_exeriment_path = \"0_3_p/\"  # this is changed based on the probability used\n",
    "general_path = general_path + specific_exeriment_path"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 10.1 Loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "destination = general_path + \"loss.png\"\n",
    "title_plot = \"Loss: \" + temp_title\n",
    "model_performance_training.plot_performance(\n",
    "    list_loss, list_val_loss, title_plot, title_y_label=\"Loss\", destination_file=destination\n",
    ")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 10.2 Precision"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "destination = general_path + \"precision.png\"\n",
    "title_plot = \"Precision: \" + temp_title\n",
    "model_performance_training.plot_performance(\n",
    "    list_precision, list_val_precision, title_plot, title_y_label=\"Precision\", destination_file=destination\n",
    ")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 10.3 Recall"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "destination = general_path + \"recall.png\"\n",
    "title_plot = \"Recall: \" + temp_title\n",
    "model_performance_training.plot_performance(\n",
    "    list_recall, list_val_recall, title_plot, title_y_label=\"Recall\", destination_file=destination\n",
    ")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 10.4 Accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "destination = general_path + \"accuracy.png\"\n",
    "title_plot = \"Accuracy: \" + temp_title\n",
    "model_performance_training.plot_performance(\n",
    "    list_accuracy, list_val_accuracy, title_plot, title_y_label=\"Accuracy\", destination_file=destination\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  10.5 Confusion matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "destination = general_path + \"confusion_matrix.png\"\n",
    "title_plot = \"Confusion Matrix: \" + temp_title\n",
    "model_performance_training.plot_confusion_matrix(confusion_matrices, destination, title_plot)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(list_val_accuracy[-1])\n",
    "print(list_val_precision[-1])\n",
    "print(list_val_recall[-1])\n",
    "print(list_val_loss[-1])"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "bach_thesis_4",
   "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": "3.9.16"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}