02_evaluate_model_parameters.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Evaluate model parameters\n",
    "\n",
    "This notebook loads precomputed results for evaluating the general settings of the proposed CNN model, corresponding to the __first part of Section 4__ in the paper. In particular, we compare different _input frame rates_, numbers of _context frames, and _CNN kernel shapes_, tested on the Schubert Winterreise Dataset (SWD).\n",
    "\n",
    "&copy; Johannes Zeitler (johannes.zeitler@fau.de) and Christof Weiß (christof.weiss@audiolabs-erlangen.de), 2020/21"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import os\n",
    "import pandas"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1. Load proposed CNN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Name</th>\n",
       "      <th>Time</th>\n",
       "      <th>cqtRate</th>\n",
       "      <th>L</th>\n",
       "      <th>numFilt</th>\n",
       "      <th>size_filt1</th>\n",
       "      <th>zerosWeight</th>\n",
       "      <th>loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>binary_accuracy</th>\n",
       "      <th>binary_crossentropy</th>\n",
       "      <th>cosine_similarity</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "      <th>fMeasure</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF5_zer...</td>\n",
       "      <td>20201030-181538</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.210942</td>\n",
       "      <td>0.388589</td>\n",
       "      <td>0.911631</td>\n",
       "      <td>0.245484</td>\n",
       "      <td>0.799504</td>\n",
       "      <td>0.777052</td>\n",
       "      <td>0.837250</td>\n",
       "      <td>0.806028</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF5_zer...</td>\n",
       "      <td>20201030-182503</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>0.4</td>\n",
       "      <td>0.220503</td>\n",
       "      <td>0.391779</td>\n",
       "      <td>0.914166</td>\n",
       "      <td>0.225547</td>\n",
       "      <td>0.808697</td>\n",
       "      <td>0.805200</td>\n",
       "      <td>0.802816</td>\n",
       "      <td>0.804007</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF5_zer...</td>\n",
       "      <td>20201030-183414</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.220979</td>\n",
       "      <td>0.392209</td>\n",
       "      <td>0.911989</td>\n",
       "      <td>0.220979</td>\n",
       "      <td>0.809014</td>\n",
       "      <td>0.833936</td>\n",
       "      <td>0.747517</td>\n",
       "      <td>0.788365</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF9_zer...</td>\n",
       "      <td>20201030-184320</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>10</td>\n",
       "      <td>9</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.200687</td>\n",
       "      <td>0.388896</td>\n",
       "      <td>0.912694</td>\n",
       "      <td>0.237466</td>\n",
       "      <td>0.806015</td>\n",
       "      <td>0.767130</td>\n",
       "      <td>0.864223</td>\n",
       "      <td>0.812787</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF9_zer...</td>\n",
       "      <td>20201030-185230</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>10</td>\n",
       "      <td>9</td>\n",
       "      <td>0.4</td>\n",
       "      <td>0.216029</td>\n",
       "      <td>0.390000</td>\n",
       "      <td>0.916416</td>\n",
       "      <td>0.219876</td>\n",
       "      <td>0.811558</td>\n",
       "      <td>0.812199</td>\n",
       "      <td>0.804984</td>\n",
       "      <td>0.808575</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>103</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF9_ze...</td>\n",
       "      <td>20201031-143023</td>\n",
       "      <td>50</td>\n",
       "      <td>75</td>\n",
       "      <td>20</td>\n",
       "      <td>9</td>\n",
       "      <td>0.4</td>\n",
       "      <td>0.183079</td>\n",
       "      <td>0.390041</td>\n",
       "      <td>0.927909</td>\n",
       "      <td>0.191208</td>\n",
       "      <td>0.834889</td>\n",
       "      <td>0.821696</td>\n",
       "      <td>0.858592</td>\n",
       "      <td>0.839739</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>104</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF9_ze...</td>\n",
       "      <td>20201031-144747</td>\n",
       "      <td>50</td>\n",
       "      <td>75</td>\n",
       "      <td>20</td>\n",
       "      <td>9</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.187157</td>\n",
       "      <td>0.381007</td>\n",
       "      <td>0.928104</td>\n",
       "      <td>0.187157</td>\n",
       "      <td>0.838468</td>\n",
       "      <td>0.842465</td>\n",
       "      <td>0.827997</td>\n",
       "      <td>0.835168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>105</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF15_z...</td>\n",
       "      <td>20201031-150527</td>\n",
       "      <td>50</td>\n",
       "      <td>75</td>\n",
       "      <td>20</td>\n",
       "      <td>15</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.174674</td>\n",
       "      <td>0.398531</td>\n",
       "      <td>0.924258</td>\n",
       "      <td>0.210319</td>\n",
       "      <td>0.828334</td>\n",
       "      <td>0.793981</td>\n",
       "      <td>0.885434</td>\n",
       "      <td>0.837217</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF15_z...</td>\n",
       "      <td>20201031-152501</td>\n",
       "      <td>50</td>\n",
       "      <td>75</td>\n",
       "      <td>20</td>\n",
       "      <td>15</td>\n",
       "      <td>0.4</td>\n",
       "      <td>0.183576</td>\n",
       "      <td>0.375782</td>\n",
       "      <td>0.928299</td>\n",
       "      <td>0.193359</td>\n",
       "      <td>0.833551</td>\n",
       "      <td>0.819456</td>\n",
       "      <td>0.864529</td>\n",
       "      <td>0.841389</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>107</th>\n",
       "      <td>musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF15_z...</td>\n",
       "      <td>20201031-154416</td>\n",
       "      <td>50</td>\n",
       "      <td>75</td>\n",
       "      <td>20</td>\n",
       "      <td>15</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.188486</td>\n",
       "      <td>0.413007</td>\n",
       "      <td>0.927664</td>\n",
       "      <td>0.188486</td>\n",
       "      <td>0.835581</td>\n",
       "      <td>0.850483</td>\n",
       "      <td>0.814328</td>\n",
       "      <td>0.832013</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>108 rows × 15 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                  Name             Time  \\\n",
       "0    musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF5_zer...  20201030-181538   \n",
       "1    musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF5_zer...  20201030-182503   \n",
       "2    musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF5_zer...  20201030-183414   \n",
       "3    musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF9_zer...  20201030-184320   \n",
       "4    musicalCNN_Jo1_Schubert_r10_L5_C5_nF10_sF9_zer...  20201030-185230   \n",
       "..                                                 ...              ...   \n",
       "103  musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF9_ze...  20201031-143023   \n",
       "104  musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF9_ze...  20201031-144747   \n",
       "105  musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF15_z...  20201031-150527   \n",
       "106  musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF15_z...  20201031-152501   \n",
       "107  musicalCNN_Jo1_Schubert_r50_L75_C5_nF20_sF15_z...  20201031-154416   \n",
       "\n",
       "     cqtRate   L  numFilt  size_filt1  zerosWeight      loss  accuracy  \\\n",
       "0         10   5       10           5          0.3  0.210942  0.388589   \n",
       "1         10   5       10           5          0.4  0.220503  0.391779   \n",
       "2         10   5       10           5          0.5  0.220979  0.392209   \n",
       "3         10   5       10           9          0.3  0.200687  0.388896   \n",
       "4         10   5       10           9          0.4  0.216029  0.390000   \n",
       "..       ...  ..      ...         ...          ...       ...       ...   \n",
       "103       50  75       20           9          0.4  0.183079  0.390041   \n",
       "104       50  75       20           9          0.5  0.187157  0.381007   \n",
       "105       50  75       20          15          0.3  0.174674  0.398531   \n",
       "106       50  75       20          15          0.4  0.183576  0.375782   \n",
       "107       50  75       20          15          0.5  0.188486  0.413007   \n",
       "\n",
       "     binary_accuracy  binary_crossentropy  cosine_similarity  Precision  \\\n",
       "0           0.911631             0.245484           0.799504   0.777052   \n",
       "1           0.914166             0.225547           0.808697   0.805200   \n",
       "2           0.911989             0.220979           0.809014   0.833936   \n",
       "3           0.912694             0.237466           0.806015   0.767130   \n",
       "4           0.916416             0.219876           0.811558   0.812199   \n",
       "..               ...                  ...                ...        ...   \n",
       "103         0.927909             0.191208           0.834889   0.821696   \n",
       "104         0.928104             0.187157           0.838468   0.842465   \n",
       "105         0.924258             0.210319           0.828334   0.793981   \n",
       "106         0.928299             0.193359           0.833551   0.819456   \n",
       "107         0.927664             0.188486           0.835581   0.850483   \n",
       "\n",
       "       Recall  fMeasure  \n",
       "0    0.837250  0.806028  \n",
       "1    0.802816  0.804007  \n",
       "2    0.747517  0.788365  \n",
       "3    0.864223  0.812787  \n",
       "4    0.804984  0.808575  \n",
       "..        ...       ...  \n",
       "103  0.858592  0.839739  \n",
       "104  0.827997  0.835168  \n",
       "105  0.885434  0.837217  \n",
       "106  0.864529  0.841389  \n",
       "107  0.814328  0.832013  \n",
       "\n",
       "[108 rows x 15 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "path_results = os.path.join('PrecomputedResults', 'musicalCNN_Schubert_eval.csv')\n",
    "data = pandas.read_csv(path_results)\n",
    "data['fMeasure'] = 2*data['Recall']*data['Precision']/(data['Recall']+data['Precision']) \n",
    "pandas.set_option('display.max_rows', 10)\n",
    "data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2. Impact of the frame rate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Color settings\n",
    "colFM = 'tab:green'\n",
    "colCS = 'tab:red'\n",
    "colPrec = 'tab:blue'\n",
    "colRec = 'tab:orange'\n",
    "\n",
    "# Figure settings\n",
    "dxFM = .2\n",
    "dxCS = -.2\n",
    "height = .2\n",
    "marker='|'\n",
    "s = 100\n",
    "figsize = (10,4)\n",
    "\n",
    "d_cqtRate10 = data[(data['L']==15) & (data['numFilt']==20) & (data['size_filt1'] == 15) & (data['zerosWeight']==0.5)]\n",
    "d_cqtRate50 = data[(data['L']==75) & (data['numFilt']==20) & (data['size_filt1'] == 15) & (data['zerosWeight']==0.5)]\n",
    "\n",
    "plotData = {'50Hz': d_cqtRate50,\n",
    "            '10Hz': d_cqtRate10}\n",
    "\n",
    "plt.rcParams.update({'font.size': 12})\n",
    "plt.figure(figsize=figsize)\n",
    "\n",
    "for x, key in enumerate(plotData.keys()):\n",
    "    \n",
    "    if x is 0:\n",
    "        labelFM = 'F-Measure'\n",
    "        labelCS = 'Cosine Similarity'\n",
    "        labelPrec = 'Precision'\n",
    "        labelRec = 'Recall'\n",
    "    else:\n",
    "        labelFM=None\n",
    "        labelCS=None\n",
    "        labelPrec=None\n",
    "        labelRec=None\n",
    "    \n",
    "    d = plotData[key]\n",
    "    plt.barh(x+dxFM, d['fMeasure'], height=height, color=colFM, zorder=0, label=labelFM)\n",
    "    plt.scatter(d['Precision'], x+dxFM, marker=marker, s=s, color=colPrec, zorder=2)\n",
    "    plt.scatter(d['Recall'], x+dxFM, marker=marker, s=s, color=colRec, zorder=2)\n",
    "    plt.plot([d['fMeasure'].item(), d['Precision'].item()], [x+dxFM,x+dxFM], color=colPrec, zorder=2, label=labelPrec)\n",
    "    plt.plot([d['fMeasure'].item(), d['Recall'].item()], [x+dxFM,x+dxFM], color=colRec, zorder=2, label=labelRec)\n",
    "    plt.barh(x+dxCS, d['cosine_similarity'], height=height, color=colCS, label=labelCS)\n",
    "\n",
    "plt.yticks(np.arange(len(plotData)), plotData.keys())\n",
    "plt.legend(ncol=2, loc='upper right')\n",
    "plt.xlim([.7, .9])\n",
    "plt.ylim([-.5, len(plotData)])\n",
    "plt.grid(axis='x')\n",
    "\n",
    "plt.ylabel('Frame rate')\n",
    "plt.xlabel('Pitch class estimation quality')\n",
    "\n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3. Impact of the number of context frames"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "d_context05 = data[(data['L']==25) & (data['numFilt']==20) & (data['size_filt1'] == 15) & (data['zerosWeight']==0.5)]\n",
    "d_context1 = data[(data['L']==51) & (data['numFilt']==20) & (data['size_filt1'] == 15) & (data['zerosWeight']==0.5)]\n",
    "d_context15 = data[(data['L']==75) & (data['numFilt']==20) & (data['size_filt1'] == 15) & (data['zerosWeight']==0.5)]\n",
    "\n",
    "plotData = {'1.5 s': d_context15,\n",
    "            '1 s': d_context1,\n",
    "            '0.5 s': d_context05}\n",
    "plt.figure(figsize=figsize)\n",
    "\n",
    "for x, key in enumerate(plotData.keys()):\n",
    "    \n",
    "    if x is 0:\n",
    "        labelFM = 'F-Measure'\n",
    "        labelCS = 'Cosine Similarity'\n",
    "        labelPrec = 'Precision'\n",
    "        labelRec = 'Recall'\n",
    "    else:\n",
    "        labelFM=None\n",
    "        labelCS=None\n",
    "        labelPrec=None\n",
    "        labelRec=None\n",
    "    \n",
    "    d = plotData[key]\n",
    "    plt.barh(x+dxFM, d['fMeasure'], height=height, color=colFM, zorder=0, label=labelFM)\n",
    "    plt.scatter(d['Precision'], x+dxFM, marker=marker, s=s, color=colPrec, zorder=2)\n",
    "    plt.scatter(d['Recall'], x+dxFM, marker=marker, s=s, color=colRec, zorder=2)\n",
    "    plt.plot([d['fMeasure'].item(), d['Precision'].item()], [x+dxFM,x+dxFM], color=colPrec, zorder=2, label=labelPrec)\n",
    "    plt.plot([d['fMeasure'].item(), d['Recall'].item()], [x+dxFM,x+dxFM], color=colRec, zorder=2, label=labelRec)\n",
    "    plt.barh(x+dxCS, d['cosine_similarity'], height=height, color=colCS, label=labelCS)\n",
    "\n",
    "plt.yticks(np.arange(len(plotData)), plotData.keys())\n",
    "plt.legend(ncol=2, loc='upper right')\n",
    "plt.xlim([.7, .9])\n",
    "plt.ylim([-.5, len(plotData)+.3])\n",
    "plt.grid(axis='x')\n",
    "\n",
    "plt.ylabel('Total context (seconds)')\n",
    "plt.xlabel('Pitch class estimation quality')\n",
    "\n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 4. Impact of the kernel shape (first layer, \"prefiltering\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "d_size5 = data[(data['L']==75) & (data['numFilt']==20) & (data['size_filt1'] == 5) & (data['zerosWeight']==0.5)]\n",
    "d_size9 = data[(data['L']==75) & (data['numFilt']==20) & (data['size_filt1'] == 9) & (data['zerosWeight']==0.5)]\n",
    "d_size15 = data[(data['L']==75) & (data['numFilt']==20) & (data['size_filt1'] == 15) & (data['zerosWeight']==0.5)]\n",
    "\n",
    "plotData = {'15x15': d_size15,'9x9': d_size9, '5x5': d_size5,}\n",
    "np.savez('fig_5-2-a_data.npz', plotData=plotData)\n",
    "\n",
    "plt.figure(figsize=figsize)\n",
    "\n",
    "for x, key in enumerate(plotData.keys()):\n",
    "    \n",
    "    if x is 0:\n",
    "        labelFM = 'F-Measure'\n",
    "        labelCS = 'Cosine Similarity'\n",
    "        labelPrec = 'Precision'\n",
    "        labelRec = 'Recall'\n",
    "    else:\n",
    "        labelFM=None\n",
    "        labelCS=None\n",
    "        labelPrec=None\n",
    "        labelRec=None\n",
    "    \n",
    "    d = plotData[key]\n",
    "    plt.barh(x+dxFM, d['fMeasure'], height=height, color=colFM, zorder=0, label=labelFM)\n",
    "    plt.scatter(d['Precision'], x+dxFM, marker=marker, s=s, color=colPrec, zorder=2)\n",
    "    plt.scatter(d['Recall'], x+dxFM, marker=marker, s=s, color=colRec, zorder=2)\n",
    "    plt.plot([d['fMeasure'].item(), d['Precision'].item()], [x+dxFM,x+dxFM], color=colPrec, zorder=2, label=labelPrec)\n",
    "    plt.plot([d['fMeasure'].item(), d['Recall'].item()], [x+dxFM,x+dxFM], color=colRec, zorder=2, label=labelRec)\n",
    "    plt.barh(x+dxCS, d['cosine_similarity'], height=height, color=colCS, label=labelCS)\n",
    "\n",
    "plt.yticks(np.arange(len(plotData)), plotData.keys())\n",
    "plt.legend(ncol=2, loc='upper right')\n",
    "plt.xlim([.7, .9])\n",
    "plt.ylim([-.5, len(plotData)+.3])\n",
    "plt.grid(axis='x')\n",
    "\n",
    "plt.ylabel('Kernel size (bins)')\n",
    "plt.xlabel('Pitch class estimation quality')\n",
    "\n",
    "plt.tight_layout()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}