Experimental Datasets Unique Speakers.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Influence of Number of Unique Speakers in Dataset on Estimating Number of Speakers\n",
    "\n",
    "This notebook contains all code used for creating the datasets used to investigate the effect of restricting the number of unique speakers in a dataset on the performance on a machine learning model trained for estimating the number of speakers in an audio file."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import tensorflow as tf\n",
    "import shutil\n",
    "import numpy as np\n",
    "from subprocess import call\n",
    "import sys\n",
    "sys.path.insert(0, './src/')\n",
    "from src import util\n",
    "import IPython.display as ipd\n",
    "import json\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn.metrics import mean_absolute_error as mae\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "import statistics\n",
    "import random\n",
    "import csv\n",
    "experimental_conditions = [750, 500, 250]\n",
    "data_dir = \"/vol/tensusers3/camghane/ASR/LibriSpeech_train_clean_360/data/train-clean-360/normalized\"\n",
    "\n",
    "\n",
    "copy_subsets = False\n",
    "merge_files = False"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Copy Subsets of LibriSpeech 360"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def cp_dir(source, target):\n",
    "    call(['cp', '-a', source, target]) # Linux"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "if copy_subsets:\n",
    "    speakers = np.array(os.listdir(data_dir))\n",
    "    for nr_speakers in experimental_conditions:\n",
    "\n",
    "        new_data_dir = \"/vol/tensusers3/camghane/ASR/jdy/data/trainset{}/normalized\".format(nr_speakers)\n",
    "\n",
    "        if not os.path.exists(new_data_dir):\n",
    "            os.makedirs(new_data_dir)\n",
    "\n",
    "        speakers_subset = speakers[np.random.choice(len(speakers), size=nr_speakers, replace=False)]\n",
    "\n",
    "        for speaker_id in speakers_subset:\n",
    "            source_loc = data_dir + \"/{}\".format(speaker_id)\n",
    "            target_loc = new_data_dir + \"/{}\".format(speaker_id)\n",
    "            cp_dir(source_loc, target_loc)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Merging Files for smallest Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "\n",
    "if merge_files:\n",
    "    max_nr_of_speakers = 10\n",
    "\n",
    "    for nr_speakers in experimental_conditions:\n",
    "        data_dir = \"/vol/tensusers3/camghane/ASR/jdy/data/trainset{}/normalized/\".format(nr_speakers)\n",
    "        new_dir = \"/vol/tensusers3/camghane/ASR/jdy/data/trainset{}/merged/\".format(nr_speakers)\n",
    "        if not os.path.exists(new_dir):\n",
    "            os.makedirs(new_dir)\n",
    "        util.merge_audiofiles(data_dir, new_dir, max_nr_of_speakers)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Move Files for Larger Datasets\n",
    "\n",
    "For the 500 and 750 speaker dataset, files per speaker need to be moved, such that in the end, all three datasets are of equal size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "# data_dir = '/vol/tensusers3/camghane/ASR/jdy/data/trainset750/normalized/'\n",
    "# for base, dirs, files in os.walk(data_dir):\n",
    "#     speaker_files = []\n",
    "#     for f in files:\n",
    "#         sid = f.split(\"_\")[0]\n",
    "#         #print(f.split(\"_\")[0])\n",
    "# #         print(data_dir+sid+\"/\"+f)\n",
    "#         speaker_files.append(data_dir+sid+\"/\"+f)\n",
    "#     print(len(speaker_files))\n",
    "# #     speaker_files = np.array(speaker_files)\n",
    "# #     random = speaker_files[np.random.choice(len(speaker_files), size=int(len(speaker_files)/3), replace=False)]\n",
    "# #     print(len(random))\n",
    "    \n",
    "# #     for source_path in random:\n",
    "# #         new_filepath = source_path.replace('normalized', 'backup')\n",
    "# #         filename = new_filepath.split(\"/\")[-1]\n",
    "# #         new_path = new_filepath.replace(filename, \"\")\n",
    "# #         if not os.path.exists(new_path):\n",
    "# #             os.makedirs(new_path)\n",
    "# #         shutil.move(source_path, new_filepath)\n",
    "#     print(\"----\")\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Counting Size per Class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total Files in Class 1 for 250 speakers: 654\n",
      "Total Files in Class 1 for 500 speakers: 662\n",
      "Total Files in Class 1 for 750 speakers: 656\n",
      "Total Files in Class 1 for original set: 2410\n",
      "\n",
      "Total Files in Class 2 for 250 speakers: 651\n",
      "Total Files in Class 2 for 500 speakers: 663\n",
      "Total Files in Class 2 for 750 speakers: 656\n",
      "Total Files in Class 2 for original set: 2410\n",
      "\n",
      "Total Files in Class 3 for 250 speakers: 648\n",
      "Total Files in Class 3 for 500 speakers: 662\n",
      "Total Files in Class 3 for 750 speakers: 657\n",
      "Total Files in Class 3 for original set: 2407\n",
      "\n",
      "Total Files in Class 4 for 250 speakers: 649\n",
      "Total Files in Class 4 for 500 speakers: 662\n",
      "Total Files in Class 4 for 750 speakers: 657\n",
      "Total Files in Class 4 for original set: 2407\n",
      "\n",
      "Total Files in Class 5 for 250 speakers: 651\n",
      "Total Files in Class 5 for 500 speakers: 662\n",
      "Total Files in Class 5 for 750 speakers: 656\n",
      "Total Files in Class 5 for original set: 2407\n",
      "\n",
      "Total Files in Class 6 for 250 speakers: 648\n",
      "Total Files in Class 6 for 500 speakers: 661\n",
      "Total Files in Class 6 for 750 speakers: 656\n",
      "Total Files in Class 6 for original set: 2407\n",
      "\n",
      "Total Files in Class 7 for 250 speakers: 647\n",
      "Total Files in Class 7 for 500 speakers: 662\n",
      "Total Files in Class 7 for 750 speakers: 655\n",
      "Total Files in Class 7 for original set: 2406\n",
      "\n",
      "Total Files in Class 8 for 250 speakers: 647\n",
      "Total Files in Class 8 for 500 speakers: 661\n",
      "Total Files in Class 8 for 750 speakers: 655\n",
      "Total Files in Class 8 for original set: 2405\n",
      "\n",
      "Total Files in Class 9 for 250 speakers: 647\n",
      "Total Files in Class 9 for 500 speakers: 661\n",
      "Total Files in Class 9 for 750 speakers: 655\n",
      "Total Files in Class 9 for original set: 2405\n",
      "\n",
      "Total Files in Class 10 for 250 speakers: 647\n",
      "Total Files in Class 10 for 500 speakers: 660\n",
      "Total Files in Class 10 for 750 speakers: 655\n",
      "Total Files in Class 10 for original set: 2405\n",
      "\n",
      "Total Files in Class 0 for 250 speakers: 648\n",
      "Total Files in Class 0 for 500 speakers: 661\n",
      "Total Files in Class 0 for 750 speakers: 655\n",
      "Total Files in Class 0 for original set: 2406\n",
      "\n",
      "7137\n",
      "7277\n",
      "7213\n",
      "26475\n"
     ]
    }
   ],
   "source": [
    "data_dir = \"/vol/tensusers3/camghane/ASR/jdy/data/trainset250/merged/train/\"\n",
    "set500 = \"/vol/tensusers3/camghane/ASR/jdy/data/trainset500/merged/train/\"\n",
    "set750 = \"/vol/tensusers3/camghane/ASR/jdy/data/trainset750/merged/train/\"\n",
    "original_set = \"/vol/tensusers3/camghane/ASR/LibriSpeech_train_clean_360/data/train-clean-360/merged/train/\"\n",
    "\n",
    "totalDir = 0\n",
    "total250 = 0\n",
    "total500 = 0\n",
    "total750 = 0\n",
    "\n",
    "for (base250, dirs250, files250), (base500, dirs500, files500), (base750, dirs750, files750), (base_or, dirs_or, files_or) in zip(os.walk(data_dir), os.walk(set500), os.walk(set750), os.walk(original_set)):\n",
    "    if not base500.endswith('/'):\n",
    "        total_or = 0\n",
    "        totalFiles250 = 0\n",
    "        totalFiles500 = 0\n",
    "        totalFiles750 = 0\n",
    "        #print('Searching in : ',base)\n",
    "        for Files in files250:\n",
    "            totalFiles250 += 1\n",
    "        for Files in files500:\n",
    "            totalFiles500 +=1\n",
    "        for Files in files750:\n",
    "            totalFiles750 +=1\n",
    "        for Files in files_or:\n",
    "            total_or += 1\n",
    "        print(\"Total Files in Class {} for 250 speakers: {}\".format(base250.split('/')[-1], totalFiles250))\n",
    "        print(\"Total Files in Class {} for 500 speakers: {}\".format(base500.split('/')[-1], totalFiles500))\n",
    "        print(\"Total Files in Class {} for 750 speakers: {}\".format(base750.split('/')[-1], totalFiles750))\n",
    "        print(\"Total Files in Class {} for original set: {}\".format(base750.split('/')[-1], total_or))\n",
    "        print(\"\")\n",
    "        \n",
    "        total250 += totalFiles250\n",
    "        total500 += totalFiles500\n",
    "        total750 += totalFiles750\n",
    "        totalDir += total_or\n",
    "print(total250)\n",
    "print(total500)\n",
    "print(total750)\n",
    "print(totalDir)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('/vol/tensusers3/camghane/ASR/predictions.json') as json_file:\n",
    "    data = json.load(json_file)\n",
    "with open('/vol/tensusers3/camghane/ASR/jdy/predictions_250.json') as json_file:\n",
    "    data250 = json.load(json_file)\n",
    "with open('/vol/tensusers3/camghane/ASR/jdy/predictions_500.json') as json_file:\n",
    "    data500 = json.load(json_file)\n",
    "with open('/vol/tensusers3/camghane/ASR/jdy/predictions_750.json') as json_file:\n",
    "    data750 = json.load(json_file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_true = data['y_true']\n",
    "y_pred = data['y_pred']\n",
    "\n",
    "cm_train = confusion_matrix(y_true, y_pred, labels=range(11))\n",
    "\n",
    "# Create figure\n",
    "fig, ax = plt.subplots(1,1, figsize=(5,5))\n",
    "\n",
    "\n",
    "\n",
    "# Plot confusion matrix for training data\n",
    "sns.heatmap(cm_train, annot=True, fmt='g', ax=ax, cmap=\"Blues\")\n",
    "\n",
    "ax.set_xlabel('Predicted labels')\n",
    "ax.set_ylabel('True labels')\n",
    "ax.set_title('Confusion Matrix for Baseline Model')\n",
    "\n",
    "b, t = plt.ylim() # discover the values for bottom and top\n",
    "b += 0.5 # Add 0.5 to the bottom\n",
    "t -= 0.5 # Subtract 0.5 from the top\n",
    "plt.ylim(b, t) # update the ylim(bottom, top) values\n",
    "plt.savefig(\"confusion_baseline.png\")\n",
    "plt.show() # ta-da!\n",
    "\n",
    "# Show the result\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MAE: 0.3471882640586797\n",
      "MAE St.Dev.: 0.548391736200062\n",
      "Relative MAE: 0.05568448882503808\n",
      "Relative MAE St.Dev.: 0.09394000696264884\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array(y_true)\n",
    "y_pred = np.array(y_pred)\n",
    "\n",
    "print(\"MAE: {}\".format(mae(y_true, y_pred)))\n",
    "print(\"MAE St.Dev.: {}\".format(statistics.stdev(abs(y_true.astype(np.float)-y_pred.astype(np.float)))))\n",
    "print(\"Relative MAE: {}\".format(np.mean(abs(y_true-y_pred)/ (y_pred+0.00001))))\n",
    "print(\"Relative MAE St.Dev.: {}\".format(statistics.stdev( abs(y_true-y_pred)/ (y_pred+0.00001) )))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "results = abs(y_true-y_pred)\n",
    "\n",
    "for r in results:\n",
    "    with open('results.csv','a') as fd:\n",
    "        writer = csv.writer(fd)\n",
    "        writer.writerow([r, 'baseline'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "with open('results_relative.csv','a') as fd:\n",
    "        writer = csv.writer(fd)\n",
    "        writer.writerow(['abs_error', 'group'])\n",
    "        \n",
    "results = abs(y_true-y_pred)/ (y_pred+0.00001)\n",
    "\n",
    "for r in results:\n",
    "    with open('results_relative.csv','a') as fd:\n",
    "        writer = csv.writer(fd)\n",
    "        writer.writerow([r, 'baseline'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_true = data250['y_true']\n",
    "y_pred = data250['y_pred']\n",
    "\n",
    "cm_train = confusion_matrix(y_true, y_pred, labels=range(11))\n",
    "\n",
    "# Create figure\n",
    "fig, ax = plt.subplots(1,1, figsize=(5,5))\n",
    "fig.set_facecolor('white')\n",
    "\n",
    "\n",
    "# Plot confusion matrix for training data\n",
    "sns.heatmap(cm_train, annot=True, fmt='g', ax=ax, cmap=\"Blues\")\n",
    "\n",
    "ax.set_xlabel('Predicted labels')\n",
    "ax.set_ylabel('True labels')\n",
    "ax.set_title('Confusion Matrix for 250-Speaker Model')\n",
    "\n",
    "b, t = plt.ylim() # discover the values for bottom and top\n",
    "b += 0.5 # Add 0.5 to the bottom\n",
    "t -= 0.5 # Subtract 0.5 from the top\n",
    "plt.ylim(b, t) # update the ylim(bottom, top) values\n",
    "plt.savefig(\"confusion_250.png\")\n",
    "plt.show() # ta-da!\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MAE: 0.42053789731051344\n",
      "MAE St.Dev.: 0.589257345954426\n",
      "Relative MAE: 0.07274308676451735\n",
      "Relative MAE St.Dev.: 0.10877958266631683\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array(y_true)\n",
    "y_pred = np.array(y_pred)\n",
    "\n",
    "print(\"MAE: {}\".format(mae(y_true, y_pred)))\n",
    "print(\"MAE St.Dev.: {}\".format(statistics.stdev(abs(y_true.astype(np.float)-y_pred.astype(np.float)))))\n",
    "print(\"Relative MAE: {}\".format(np.mean(abs(y_true-y_pred)/ (y_pred+0.00001))))\n",
    "print(\"Relative MAE St.Dev.: {}\".format(statistics.stdev( abs(y_true-y_pred)/ (y_pred+0.00001) )))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# results = abs(y_true-y_pred)\n",
    "\n",
    "# for r in results:\n",
    "#     with open('results.csv','a') as fd:\n",
    "#         writer = csv.writer(fd)\n",
    "#         writer.writerow([r, '250 Speakers'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "results = abs(y_true-y_pred)/ (y_pred+0.00001)\n",
    "\n",
    "for r in results:\n",
    "    with open('results_relative.csv','a') as fd:\n",
    "        writer = csv.writer(fd)\n",
    "        writer.writerow([r, '250 Speakers'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_true = data500['y_true']\n",
    "y_pred = data500['y_pred']\n",
    "\n",
    "cm_train = confusion_matrix(y_true, y_pred, labels=range(11))\n",
    "\n",
    "# Create figure\n",
    "fig, ax = plt.subplots(1,1, figsize=(5,5))\n",
    "fig.set_facecolor('white')\n",
    "\n",
    "\n",
    "# Plot confusion matrix for training data\n",
    "sns.heatmap(cm_train, annot=True, fmt='g', ax=ax, cmap=\"Blues\")\n",
    "\n",
    "ax.set_xlabel('Predicted labels')\n",
    "ax.set_ylabel('True labels')\n",
    "ax.set_title('Confusion Matrix for 500-Speaker Model')\n",
    "\n",
    "b, t = plt.ylim() # discover the values for bottom and top\n",
    "b += 0.5 # Add 0.5 to the bottom\n",
    "t -= 0.5 # Subtract 0.5 from the top\n",
    "plt.ylim(b, t) # update the ylim(bottom, top) values\n",
    "plt.savefig(\"confusion_500.png\")\n",
    "plt.show() # ta-da!\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MAE: 0.4156479217603912\n",
      "MAE St.Dev.: 0.5801872865737389\n",
      "Relative MAE: 0.07169619299803857\n",
      "Relative MAE St.Dev.: 0.11379483896360802\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array(y_true)\n",
    "y_pred = np.array(y_pred)\n",
    "\n",
    "print(\"MAE: {}\".format(mae(y_true, y_pred)))\n",
    "print(\"MAE St.Dev.: {}\".format(statistics.stdev(abs(y_true.astype(np.float)-y_pred.astype(np.float)))))\n",
    "print(\"Relative MAE: {}\".format(np.mean(abs(y_true-y_pred)/ (y_pred+0.00001))))\n",
    "print(\"Relative MAE St.Dev.: {}\".format(statistics.stdev( abs(y_true-y_pred)/ (y_pred+0.00001) )))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# results = abs(y_true-y_pred)\n",
    "# for r in results:\n",
    "#     with open('results.csv','a') as fd:\n",
    "#         writer = csv.writer(fd)\n",
    "#         writer.writerow([r, '500 Speakers'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "results = abs(y_true-y_pred)/ (y_pred+0.00001)\n",
    "\n",
    "for r in results:\n",
    "    with open('results_relative.csv','a') as fd:\n",
    "        writer = csv.writer(fd)\n",
    "        writer.writerow([r, '500 Speakers'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_true = data750['y_true']\n",
    "y_pred = data750['y_pred']\n",
    "\n",
    "cm_train = confusion_matrix(y_true, y_pred, labels=range(11))\n",
    "\n",
    "# Create figure\n",
    "fig, ax = plt.subplots(1,1, figsize=(5,5))\n",
    "fig.set_facecolor('white')\n",
    "\n",
    "\n",
    "# Plot confusion matrix for training data\n",
    "sns.heatmap(cm_train, annot=True, fmt='g', ax=ax, cmap=\"Blues\")\n",
    "\n",
    "ax.set_xlabel('Predicted labels')\n",
    "ax.set_ylabel('True labels')\n",
    "ax.set_title('Confusion Matrix for 750-Speaker Model')\n",
    "\n",
    "b, t = plt.ylim() # discover the values for bottom and top\n",
    "b += 0.5 # Add 0.5 to the bottom\n",
    "t -= 0.5 # Subtract 0.5 from the top\n",
    "plt.ylim(b, t) # update the ylim(bottom, top) values\n",
    "plt.savefig(\"confusion_750.png\")\n",
    "plt.show() # ta-da!\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MAE: 0.39364303178484106\n",
      "MAE St.Dev.: 0.5891658108321721\n",
      "Relative MAE: 0.0669042259430424\n",
      "Relative MAE St.Dev.: 0.10984922658566777\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array(y_true)\n",
    "y_pred = np.array(y_pred)\n",
    "\n",
    "print(\"MAE: {}\".format(mae(y_true, y_pred)))\n",
    "print(\"MAE St.Dev.: {}\".format(statistics.stdev(abs(y_true.astype(np.float)-y_pred.astype(np.float)))))\n",
    "print(\"Relative MAE: {}\".format(np.mean(abs(y_true-y_pred)/ (y_pred+0.00001))))\n",
    "print(\"Relative MAE St.Dev.: {}\".format(statistics.stdev( abs(y_true-y_pred)/ (y_pred+0.00001) )))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# results = abs(y_true-y_pred)\n",
    "# for r in results:\n",
    "#     with open('results.csv','a') as fd:\n",
    "#         writer = csv.writer(fd)\n",
    "#         writer.writerow([r, '750 Speakers'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "results = abs(y_true-y_pred)/ (y_pred+0.00001)\n",
    "\n",
    "for r in results:\n",
    "    with open('results_relative.csv','a') as fd:\n",
    "        writer = csv.writer(fd)\n",
    "        writer.writerow([r, '750 Speakers'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}