pn2_visualization_with_multiple_radius_v3.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "34c51bd628d64c82a3c29cdbedc32b71",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HTML(value='<h1>Interactive Dashboards for Exploring Deep Learning Models on Point Cloud</h1>', layout=Layout(…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "db99c0c3a99443f8b3ce8612ca1822c1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Tab(children=(VBox(children=(HBox(children=(Box(children=(FigureWidget({\n",
       "    'data': [{'hovertemplate': 'Radiu…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import pickle\n",
    "from ipyvolume import pylab as pb\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import ipywidgets as widgets\n",
    "from plotly import graph_objects as go\n",
    "from plotly.subplots import make_subplots\n",
    "from collections import Counter\n",
    "import pandas as pd\n",
    "import glob\n",
    "%matplotlib inline \n",
    "color_bar = {'clutter':[200,200,200],'chair':[0,0,255],\n",
    "                 'table':[0,255,255],\n",
    "                 'bookcase':[255,255,0],\n",
    "                 'sofa':[255,0,255],\n",
    "                 'column':[100,100,255],\n",
    "                 'board':[200,200,100],\n",
    "                 'window':[170,120,200],\n",
    "                 'door':[255,0,0],\n",
    "                 'beam':[200,100,100],\n",
    "                 'wall':[10,200,100],\n",
    "                 'floor':[0,255,0],\n",
    "                 'ceiling':[50,50,50]} \n",
    "\n",
    "with open('pred_for_dif_rad.pickle','rb') as f:\n",
    "    pts = pickle.load(f)\n",
    "    rad_list = pickle.load(f)\n",
    "with open('test_split.txt','r') as f:\n",
    "    rooms = f.readlines()\n",
    "\n",
    "rooms = [ r for r in rooms if 'hallway' not in r and 'WC' not in r and 'storage' not in r]\n",
    "ori_rooms = rooms\n",
    "rooms = list(zip(rooms,range(len(rooms))))\n",
    "x_axis = [int(s)/1000 for s in ['025','050','075','100','125','150','175','200','225','250','300','350','400']]\n",
    "rad_acc_list = []\n",
    "conf_mat_list = []\n",
    "for i in range(len(rad_list)):\n",
    "    rad_acc_list.append((np.sum(np.concatenate(pts['gt'],axis=0)==np.concatenate(rad_list[i],axis=0))/np.concatenate(pts['gt'],axis=0).shape[0],np.histogram(np.concatenate(pts['gt'],axis=0)[np.concatenate(pts['gt'],axis=0)==np.concatenate(rad_list[i],axis=0)],range(len(color_bar)+1))[0]/np.histogram(np.concatenate(pts['gt'],axis=0),range(len(color_bar)+1))[0]))\n",
    "#Overall Line Chart\n",
    "overall_acc_plt = go.FigureWidget()\n",
    "overall_acc_plt.add_traces([go.Scatter(x=x_axis,y=[i[0] for i in rad_acc_list],mode='lines',hovertemplate=\"Radius:%{x}  Accuracy:%{y:.3f}\",name='per_point(8192)')])\n",
    "overall_acc_plt.update_layout(title='Overall accuracy',\n",
    "                              xaxis_title='Radius',\n",
    "                              yaxis_title='Accuracy')\n",
    "class_acc_plt = go.FigureWidget()\n",
    "for i in range(len(color_bar)):\n",
    "    class_acc_plt.add_trace(go.Scatter(x=x_axis,y=np.array([i[1] for i in rad_acc_list])[:,i],mode='lines',name=list(color_bar.keys())[i],marker=dict(color='rgb('+', '.join(map(str,list(color_bar.values())[i]))+')')))\n",
    "class_acc_plt.update_layout(title='Class accuracy',xaxis_title='Radius',yaxis_title='Accuracy',height=600)\n",
    "\n",
    "#Bar, confusion Matrix, and Pie by radius###############################\n",
    "i = 0\n",
    "#Confusion matrix\n",
    "pre_all_pts_pn2 = np.concatenate(rad_list[i])\n",
    "gt_all_pts_pn2 = np.concatenate(pts['gt'])\n",
    "df_confusion_pn2 = pd.crosstab(pre_all_pts_pn2,gt_all_pts_pn2,rownames=['Predict'],colnames=['GT'])\n",
    "df_confusion_pn2 = df_confusion_pn2/df_confusion_pn2.sum(axis=0)\n",
    "for i in range(len(color_bar)):\n",
    "    try:\n",
    "        df_confusion_pn2[i]\n",
    "    except KeyError:\n",
    "        df_confusion_pn2[i]=np.nan\n",
    "    try:\n",
    "        df_confusion_pn2.loc[i]\n",
    "    except KeyError:\n",
    "        df_confusion_pn2.loc[i]=np.nan\n",
    "df_confusion_pn2=df_confusion_pn2[range(len(color_bar))].reindex(range(len(color_bar)))\n",
    "conf_matrix_pn2 = go.FigureWidget(layout=dict(title={'text':\"Confusion Matrix\"}))\n",
    "conf_matrix_pn2.add_traces([go.Heatmap(\n",
    "    name='pn2',\n",
    "    z=df_confusion_pn2.values,\n",
    "    x=list(color_bar.keys()),\n",
    "    y=list(color_bar.keys()),\n",
    "    autocolorscale=True)])\n",
    "conf_matrix_pn2.update_layout(\n",
    "    yaxis_title=\"Prediction\",\n",
    "    xaxis_title=\"Ground Truth\",\n",
    "    width = 400,\n",
    "    height = 400\n",
    ")\n",
    "def click_event(trace,points,selector):\n",
    "    #print(points.xs[0])\n",
    "    radius_selector.children[0].label = str(points.xs[0])\n",
    "    radius_selector.update()\n",
    "overall_acc_plt.data[0].on_click(click_event)\n",
    "\n",
    "def update_by_radius(rd):\n",
    "    pre_all_pts_pn2 = np.concatenate(rad_list[rd])\n",
    "    gt_all_pts_pn2 = np.concatenate(pts['gt'])\n",
    "    df_confusion_pn2 = pd.crosstab(pre_all_pts_pn2,gt_all_pts_pn2,rownames=['Predict'],colnames=['GT'])\n",
    "    df_confusion_pn2 = df_confusion_pn2/df_confusion_pn2.sum(axis=0)\n",
    "    for i in range(len(color_bar)):\n",
    "        try:\n",
    "            df_confusion_pn2[i]\n",
    "        except KeyError:\n",
    "            df_confusion_pn2[i]=np.nan\n",
    "        try:\n",
    "            df_confusion_pn2.loc[i]\n",
    "        except KeyError:\n",
    "            df_confusion_pn2.loc[i]=np.nan\n",
    "    df_confusion_pn2=df_confusion_pn2[range(len(color_bar))].reindex(range(len(color_bar)))\n",
    "    pn2_acc=np.round([np.sum((pre_all_pts_pn2==i) & (gt_all_pts_pn2==i))/np.sum(gt_all_pts_pn2==i) if np.sum(gt_all_pts_pn2==i)!=0 else np.nan for i in range(len(color_bar))],decimals=4)\n",
    "#     acc_plot.update_traces(selector={'name':'All Points'},patch={'y':pn2_acc,'text':pn2_acc})\n",
    "    conf_matrix_pn2.update_traces(selector={'name':'pn2'},patch={'z':df_confusion_pn2})\n",
    "    radius_selector_cur.children[0].label = radius_selector.children[0].label\n",
    "    radius_selector_cur.update()\n",
    "radius_selector = widgets.interactive(update_by_radius,rd=widgets.Dropdown(options=[(a,b) for (b,a) in enumerate(x_axis)],values=0,description=\"Radius: \"))\n",
    "\n",
    "#########################################################\n",
    "\n",
    "#Bar, confusion Matrix, and Pie by radius and room###############################\n",
    "rd_cur = 0\n",
    "rm_cur = 0\n",
    "#Bar by room\n",
    "selected_cls = 0\n",
    "pn2_acc_by_rms = np.round([np.sum((rad_list[rd_cur][j]==selected_cls) & (pts['gt'][j]==selected_cls))/np.sum(pts['gt'][j]==selected_cls) if np.sum(pts['gt'][j]==selected_cls)!=0 else np.nan for j in range(len(rooms))],decimals=4)\n",
    "rms_idx = pn2_acc_by_rms.argsort()\n",
    "acc_by_rms_fig = go.FigureWidget()\n",
    "acc_by_rms_fig.add_traces([\n",
    "    go.Bar(\n",
    "        x=np.array(ori_rooms)[rms_idx],\n",
    "        y=pn2_acc_by_rms[rms_idx],\n",
    "        name='All Points')\n",
    "])\n",
    "acc_by_rms_fig.update_layout(height=400,width=800,title='Accuracy by Room')\n",
    "\n",
    "#Confusion matrix\n",
    "pre_all_pts_pn2_cur = rad_list[rd_cur][rm_cur]\n",
    "gt_all_pts_pn2_cur = pts['gt'][rm_cur]\n",
    "\n",
    "df_confusion_pn2_cur = pd.crosstab(pre_all_pts_pn2_cur,gt_all_pts_pn2_cur,rownames=['Predict'],colnames=['GT'])\n",
    "df_confusion_pn2_cur = df_confusion_pn2_cur/df_confusion_pn2_cur.sum(axis=0)\n",
    "for i in range(len(color_bar)):\n",
    "    try:\n",
    "        df_confusion_pn2_cur[i]\n",
    "    except KeyError:\n",
    "        df_confusion_pn2_cur[i]=np.nan\n",
    "    try:\n",
    "        df_confusion_pn2_cur.loc[i]\n",
    "    except KeyError:\n",
    "        df_confusion_pn2_cur.loc[i]=np.nan\n",
    "df_confusion_pn2_cur=df_confusion_pn2_cur[range(len(color_bar))].reindex(range(len(color_bar)))\n",
    "conf_matrix_pn2_cur = go.FigureWidget(layout=dict(title={'text':\"Confusion Matrix\"}))\n",
    "conf_matrix_pn2_cur.add_traces([go.Heatmap(\n",
    "    name='pn2',\n",
    "    z=df_confusion_pn2_cur.values,\n",
    "    x=list(color_bar.keys()),\n",
    "    y=list(color_bar.keys()),\n",
    "    autocolorscale=True)])\n",
    "conf_matrix_pn2_cur.update_layout(\n",
    "    yaxis_title=\"Prediction\",\n",
    "    xaxis_title=\"Ground Truth\",\n",
    "    width = 400,\n",
    "    height = 400\n",
    ")\n",
    "\n",
    "#Bar\n",
    "pn2_acc_cur=np.round([np.sum((pre_all_pts_pn2_cur==i) & (gt_all_pts_pn2_cur==i))/np.sum(gt_all_pts_pn2_cur==i) if np.sum(gt_all_pts_pn2_cur==i)!=0 else np.nan for i in range(len(color_bar))],decimals=4)\n",
    "acc_plot_cur = go.FigureWidget()\n",
    "acc_plot_cur.add_traces([\n",
    "    go.Bar(\n",
    "        x=list(color_bar.keys()),\n",
    "        y=pn2_acc_cur,\n",
    "        name='All Points',\n",
    "        text=pn2_acc_cur,\n",
    "        textposition='auto',\n",
    "        marker=dict(color=['rgb('+', '.join(map(str,list(color_bar.values())[k]))+')' for k in range(len(color_bar))]))])\n",
    "acc_plot_cur.update_layout(height=400,width=400,title='Accuracy by Class')\n",
    "\n",
    "#pie\n",
    "pts_dist_cur = np.histogram(gt_all_pts_pn2_cur,range(len(color_bar)+1))[0]/gt_all_pts_pn2_cur.shape[0]\n",
    "dist_plot_cur = go.FigureWidget(layout=dict(title={'text':\"Distrbution\"}))\n",
    "dist_plot_cur.add_traces([\n",
    "    go.Pie(\n",
    "        title='Distribution',\n",
    "        values=pts_dist_cur,\n",
    "        labels=list(color_bar.keys()),\n",
    "        marker=dict(colors=['rgb('+', '.join(map(str,list(color_bar.values())[k]))+')'  for k in range(len(color_bar))]),\n",
    "        name='All Points')])\n",
    "dist_plot_cur.update_layout(height=400,width=500 )\n",
    "def update_by_class(selected_cls):\n",
    "    global rd_cur\n",
    "    pn2_acc_by_rms = np.round([np.sum((rad_list[rd_cur][j]==selected_cls) & (pts['gt'][j]==selected_cls))/np.sum(pts['gt'][j]==selected_cls) if np.sum(pts['gt'][j]==selected_cls)!=0 else np.nan for j in range(len(rooms))],decimals=4)\n",
    "    rms_idx = pn2_acc_by_rms.argsort()\n",
    "    acc_by_rms_fig.update_traces(selector={'name':'All Points'},patch={'x':np.array(ori_rooms)[rms_idx],'y':pn2_acc_by_rms[rms_idx]})\n",
    "\n",
    "def update_by_radius_cur(rd):\n",
    "    global rd_cur\n",
    "    global rm_cur\n",
    "    rm = rm_cur\n",
    "    rd_cur = rd\n",
    "    pre_all_pts_pn2_cur = rad_list[rd][rm]\n",
    "    gt_all_pts_pn2_cur = pts['gt'][rm]\n",
    "    df_confusion_pn2_cur = pd.crosstab(pre_all_pts_pn2_cur,gt_all_pts_pn2_cur,rownames=['Predict'],colnames=['GT'])\n",
    "    df_confusion_pn2_cur = df_confusion_pn2_cur/df_confusion_pn2_cur.sum(axis=0)\n",
    "    for i in range(len(color_bar)):\n",
    "        try:\n",
    "            df_confusion_pn2_cur[i]\n",
    "        except KeyError:\n",
    "            df_confusion_pn2_cur[i]=np.nan\n",
    "        try:\n",
    "            df_confusion_pn2_cur.loc[i]\n",
    "        except KeyError:\n",
    "            df_confusion_pn2_cur.loc[i]=np.nan\n",
    "    df_confusion_pn2_cur=df_confusion_pn2_cur[range(len(color_bar))].reindex(range(len(color_bar)))\n",
    "    pn2_acc_cur=np.round([np.sum((pre_all_pts_pn2_cur==i) & (gt_all_pts_pn2_cur==i))/np.sum(gt_all_pts_pn2_cur==i) if np.sum(gt_all_pts_pn2_cur==i)!=0 else np.nan for i in range(len(color_bar))],decimals=4)\n",
    "    acc_plot_cur.update_traces(selector={'name':'All Points'},patch={'y':pn2_acc_cur,'text':pn2_acc_cur})\n",
    "    conf_matrix_pn2_cur.update_traces(selector={'name':'pn2'},patch={'z':df_confusion_pn2_cur})\n",
    "    pn2_acc_by_rms = np.round([np.sum((rad_list[rd_cur][j]==selected_cls) & (pts['gt'][j]==selected_cls))/np.sum(pts['gt'][j]==selected_cls) if np.sum(pts['gt'][j]==selected_cls)!=0 else np.nan for j in range(len(rooms))],decimals=4)\n",
    "    rms_idx = pn2_acc_by_rms.argsort()\n",
    "    acc_by_rms_fig.update_traces(selector={'name':'All Points'},patch={'x':np.array(ori_rooms)[rms_idx],'y':pn2_acc_by_rms[rms_idx]})\n",
    "    if not gt_check:\n",
    "        update_color(gt_check)\n",
    "    cls_selector_cur.update()\n",
    "def update_by_room_cur(rm):\n",
    "    global rd_cur\n",
    "    global rm_cur\n",
    "    rm_cur = rm\n",
    "    rd = rd_cur\n",
    "    pre_all_pts_pn2_cur = rad_list[rd][rm]\n",
    "    gt_all_pts_pn2_cur = pts['gt'][rm]\n",
    "    df_confusion_pn2_cur = pd.crosstab(pre_all_pts_pn2_cur,gt_all_pts_pn2_cur,rownames=['Predict'],colnames=['GT'])\n",
    "    df_confusion_pn2_cur = df_confusion_pn2_cur/df_confusion_pn2_cur.sum(axis=0)\n",
    "    for i in range(len(color_bar)):\n",
    "        try:\n",
    "            df_confusion_pn2_cur[i]\n",
    "        except KeyError:\n",
    "            df_confusion_pn2_cur[i]=np.nan\n",
    "        try:\n",
    "            df_confusion_pn2_cur.loc[i]\n",
    "        except KeyError:\n",
    "            df_confusion_pn2_cur.loc[i]=np.nan\n",
    "    df_confusion_pn2_cur=df_confusion_pn2_cur[range(len(color_bar))].reindex(range(len(color_bar)))\n",
    "    pn2_acc_cur=np.round([np.sum((pre_all_pts_pn2_cur==i) & (gt_all_pts_pn2_cur==i))/np.sum(gt_all_pts_pn2_cur==i) if np.sum(gt_all_pts_pn2_cur==i)!=0 else np.nan for i in range(len(color_bar))],decimals=4)\n",
    "    pts_dist_cur = np.histogram(gt_all_pts_pn2_cur,range(len(color_bar)+1))[0]/gt_all_pts_pn2_cur.shape[0]\n",
    "    acc_plot_cur.update_traces(selector={'name':'All Points'},patch={'y':pn2_acc_cur,'text':pn2_acc_cur})\n",
    "    conf_matrix_pn2_cur.update_traces(selector={'name':'pn2'},patch={'z':df_confusion_pn2_cur})\n",
    "    dist_plot_cur.update_traces(selector={'name':'All Points'},patch={'values':pts_dist_cur})\n",
    "    \n",
    "    hide_classes(gt_check)\n",
    "radius_selector_cur = widgets.interactive(update_by_radius_cur,rd=widgets.Dropdown(options=[(a,b) for (b,a) in enumerate(x_axis)],values=0,description=\"Radius: \"))\n",
    "room_selector_cur = widgets.interactive(update_by_room_cur,rm=widgets.Dropdown(options=rooms,description='Room: ',style={'description_width': 'initial'}))\n",
    "cls_selector_cur = widgets.interactive(update_by_class,selected_cls=widgets.Dropdown(options=list(zip(list(color_bar.keys()),range(len(color_bar)))),description='Class: '))\n",
    "def click_event_rm(trace,points,selector):\n",
    "    room_selector_cur.children[0].label = str(points.xs[0])\n",
    "    room_selector_cur.update()\n",
    "acc_by_rms_fig.data[0].on_click(click_event_rm)\n",
    "########################################################################\n",
    "\n",
    "####Point Cloud#########################################################\n",
    "pn2_points=pts['points'][rm_cur]\n",
    "pn2_fig=pb.figure(width=600,heigth=500,title='Point Cloud')\n",
    "scatter_pn2_gt = pb.scatter(pn2_points[:,0]-pn2_points[:,0].min(), pn2_points[:,1]-pn2_points[:,1].min(), pn2_points[:,2]-pn2_points[:,2].min(),size=1,marker='sphere')\n",
    "pb.xyzlim(-1,6)\n",
    "pb.squarelim()\n",
    "pn2_fig.animation = 300\n",
    "pb.style.box_off()\n",
    "\n",
    "gt_check=True\n",
    "points_filter = pts['gt'][rm_cur] != -1\n",
    "def update_color(g):\n",
    "    global gt_check\n",
    "    gt_check = g\n",
    "    if gt_check:\n",
    "        scatter_pn2_gt.color = np.array([list(color_bar.values())[i] for i in pts['gt'][rm_cur][points_filter].astype(int)])/255\n",
    "    else:\n",
    "        scatter_pn2_gt.color = np.array([list(color_bar.values())[i] for i in rad_list[rd_cur][rm_cur][points_filter].astype(int)])/255\n",
    "\n",
    "# def hide_classes(*args):\n",
    "#     global points_filter\n",
    "#     points_filter = np.isin(pts['gt'][rm_cur], class_selector.value, invert=True)\n",
    "#     pn2_points=pts['points'][rm_cur][points_filter,:]\n",
    "    \n",
    "#     scatter_pn2_gt.x = pn2_points[:,0]-pn2_points[:,0].min()\n",
    "#     scatter_pn2_gt.y = pn2_points[:,1]-pn2_points[:,1].min()\n",
    "#     scatter_pn2_gt.z = pn2_points[:,2]-pn2_points[:,2].min()\n",
    "#     update_color(gt_check)\n",
    "def hide_classes(*args):\n",
    "    global points_filter\n",
    "    if scatter_pn2_gt.visible == False:\n",
    "        scatter_pn2_gt.visible = True\n",
    "    cur_hinding_cls = [not c.value for c in class_check_list[:-1]]\n",
    "    vfunc = np.vectorize(lambda i: cur_hinding_cls[i])\n",
    "    points_filter = vfunc(pts['gt'][rm_cur])\n",
    "    pn2_points=pts['points'][rm_cur][points_filter,:]\n",
    "    if pn2_points.shape[0]==0:\n",
    "        scatter_pn2_gt.visible = False\n",
    "        return\n",
    "    \n",
    "    scatter_pn2_gt.x = pn2_points[:,0]-pn2_points[:,0].min()\n",
    "    scatter_pn2_gt.y = pn2_points[:,1]-pn2_points[:,1].min()\n",
    "    scatter_pn2_gt.z = pn2_points[:,2]-pn2_points[:,2].min()\n",
    "    update_color(gt_check)\n",
    "    \n",
    "gt_checkbox = widgets.interactive(update_color,g=widgets.Checkbox(value=True,description='Show Ground Truth',style={'description_width': 'initial'},layout={'width':'99%'}))\n",
    "# class_selector = widgets.SelectMultiple(options=list(zip(color_bar.keys(),range(len(color_bar)))),value=[],description='Classes to hide: ',style={'description_width': 'initial'})\n",
    "# class_selector.observe(hide_classes,names='value')\n",
    "class_check_list = [widgets.Checkbox(description=i,value=False,style={'description_width': 'initial'}) for i in color_bar.keys()]\n",
    "class_selector  = widgets.Button(description='Apply')#,layout={'width':'80px'})\n",
    "class_check_list.append(class_selector)\n",
    "class_dropdown = widgets.Accordion(children=[widgets.Box([widgets.GridBox(class_check_list,layout=widgets.Layout(grid_template_columns=\"repeat(3, 80px)\"))],layout={'border':'1px solid black','width':'250px'})],layout={'width':'300px'})\n",
    "class_dropdown.set_title(0,'Classes to hide')\n",
    "class_selector.on_click(hide_classes)\n",
    "########################################################################\n",
    "header = widgets.HTML(value=\"<h1>Interactive Dashboards for Exploring Deep Learning Models on Point Cloud</h1>\",layout={'width':'1000px','align_self':'center'})\n",
    "visual_tab = widgets.Tab(layout={'width':'1030px','align_self':'center'})\n",
    "visual_tab.children = [\n",
    "    widgets.VBox([widgets.HBox([widgets.Box([overall_acc_plt],layout=widgets.Layout(flex='4 1 0%', width='auto')),widgets.VBox([radius_selector,conf_matrix_pn2],layout=widgets.Layout(flex='3 1 0%', width='auto'))],layout={'margin':'0'}),class_acc_plt,widgets.HTML(value='<b>*Click legends on the right to hide/show</b>',layout={'align_self':'flex-end'})],layout={'width':'1000px','align-items':'flex-start'}),\n",
    "    widgets.VBox([widgets.HBox([radius_selector_cur,cls_selector_cur]),acc_by_rms_fig,widgets.VBox([widgets.HBox([widgets.VBox([widgets.HBox([room_selector_cur,gt_checkbox],layout={'width':'600px'}),pn2_fig],layout={'height':'600px'}),widgets.VBox([class_dropdown,conf_matrix_pn2_cur])]),widgets.HBox([acc_plot_cur,widgets.VBox([dist_plot_cur,widgets.HTML(value='<b>*Click legends on the right to hide/show</b>',layout={'align_self':'flex-end'})])]),],layout={'border': '1px solid black'})],layout={'width':'1000px'})]\n",
    "visual_tab.set_title(0,'Overall Analysis')\n",
    "visual_tab.set_title(1,'Room Analysis')\n",
    "display(header,visual_tab)"
   ]
  }
 ],
 "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": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}