pretrain_planet.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Understanding the Amazon from Space\n",
    "\n",
    "Pretrain neural network using Amazon satellite images from old Kaggle [competition](https://www.kaggle.com/c/planet-understanding-the-amazon-from-space). In order to used the trained weight for the Oil Palm [competition](https://www.kaggle.com/c/widsdatathon2019/leaderboard). This problem is originally a multi-class classification problem. I changed it to a single class classification problem to make the pretrained weights similar to the Oil Palm compeition."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "#setup the notebook\n",
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline\n",
    "%config IPCompleter.greedy=True\n",
    "\n",
    "import sys\n",
    "import feather\n",
    "from pathlib import Path\n",
    "from glob import glob\n",
    "sys.path.append(r'C:\\Users\\Benny\\fastai\\old')\n",
    "\n",
    "from fastai.imports import *\n",
    "from fastai.transforms import *\n",
    "from fastai.conv_learner import *\n",
    "from fastai.model import *\n",
    "from fastai.dataset import *\n",
    "from fastai.plots import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.cuda.is_available()\n",
    "torch.backends.cudnn.enabled"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " Volume in drive C is Windows\n",
      " Volume Serial Number is FC06-B52C\n",
      "\n",
      " Directory of C:\\Users\\Benny\\Documents\\Fern\\planet\n",
      "\n",
      "03/02/2019  02:42 PM    <DIR>          .\n",
      "03/02/2019  02:42 PM    <DIR>          ..\n",
      "03/02/2019  02:39 PM    <DIR>          .ipynb_checkpoints\n",
      "03/02/2019  02:34 PM           600,968 lesson2-image_models.ipynb\n",
      "03/02/2019  02:42 PM             1,962 planet.ipynb\n",
      "03/02/2019  01:01 PM    <DIR>          train\n",
      "02/28/2019  02:34 PM         1,432,177 train_v2.csv\n",
      "               3 File(s)      2,035,107 bytes\n",
      "               4 Dir(s)  13,857,865,728 bytes free\n"
     ]
    }
   ],
   "source": [
    "!dir"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "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>image_name</th>\n",
       "      <th>tags</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>train_0</td>\n",
       "      <td>haze primary</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>train_1</td>\n",
       "      <td>agriculture clear primary water</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>train_2</td>\n",
       "      <td>clear primary</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>train_3</td>\n",
       "      <td>clear primary</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>train_4</td>\n",
       "      <td>agriculture clear habitation primary road</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  image_name                                       tags\n",
       "0    train_0                               haze primary\n",
       "1    train_1            agriculture clear primary water\n",
       "2    train_2                              clear primary\n",
       "3    train_3                              clear primary\n",
       "4    train_4  agriculture clear habitation primary road"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv('train_v2.csv')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Modify Classes\n",
    "\n",
    "There are 17 classes in this problems. Some are related to the weather, and some are not related to agriculture vs forest. These classes are dropped."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# drop all irrelvant classes \n",
    "to_drop = ['clear', 'cloudy', 'haze', 'partly_cloudy',\n",
    "          'conventional_mine', 'artisinal_mine', 'habitation',\n",
    "          'road', 'water', 'selective_logging','slash_burn']\n",
    "\n",
    "#change all argriculture area to 1\n",
    "to_arg = ['cultivation']\n",
    "\n",
    "# change all non-argriculture to 0\n",
    "to_primary = ['bare_ground', 'blooming', 'blow_down']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_mod = df.copy()\n",
    "    \n",
    "for word in to_arg:\n",
    "    df_mod['tags'] = df_mod['tags'].str.replace(word,'agriculture')\n",
    "\n",
    "for word in to_primary:\n",
    "    df_mod['tags'] = df_mod['tags'].str.replace(word,'primary')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "40479"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(df_mod)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "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>image_name</th>\n",
       "      <th>tags</th>\n",
       "      <th>is_arg</th>\n",
       "      <th>is_forest</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>train_0</td>\n",
       "      <td>haze primary</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>train_1</td>\n",
       "      <td>agriculture clear primary water</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>train_2</td>\n",
       "      <td>clear primary</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>train_3</td>\n",
       "      <td>clear primary</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>train_4</td>\n",
       "      <td>agriculture clear habitation primary road</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  image_name                                       tags  is_arg  is_forest\n",
       "0    train_0                               haze primary   False       True\n",
       "1    train_1            agriculture clear primary water    True       True\n",
       "2    train_2                              clear primary   False       True\n",
       "3    train_3                              clear primary   False       True\n",
       "4    train_4  agriculture clear habitation primary road    True       True"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_mod.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_mod['is_arg'] = df_mod['tags'].str.contains('agriculture')\n",
    "df_mod['is_forest'] = df_mod['tags'].str.contains('primary')\n",
    "idxs = df_mod[(df_mod['is_arg'] == False) & (df_mod['is_forest'] == False)].index.values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(None, 37995)"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_mod.drop(idxs, axis=0, inplace=True), len(df_mod)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_mod = df_mod.drop('tags', axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_mod = df_mod.drop('is_forest', axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_mod['is_arg'] = df_mod['is_arg'].astype(int)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 37995 entries, 0 to 40478\n",
      "Data columns (total 2 columns):\n",
      "image_name    37995 non-null object\n",
      "is_arg        37995 non-null int32\n",
      "dtypes: int32(1), object(1)\n",
      "memory usage: 742.1+ KB\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(None,   image_name  is_arg\n",
       " 0    train_0       0\n",
       " 1    train_1       1\n",
       " 2    train_2       0\n",
       " 3    train_3       0\n",
       " 4    train_4       1)"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_mod.info(), df_mod.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_mod.to_csv('train_single.csv',index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train a NN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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>image_name</th>\n",
       "      <th>is_arg</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>train_0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>train_1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>train_2</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>train_3</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>train_4</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  image_name  is_arg\n",
       "0    train_0       0\n",
       "1    train_1       1\n",
       "2    train_2       0\n",
       "3    train_3       0\n",
       "4    train_4       1"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_mod = pd.read_csv('train_single.csv')\n",
    "df_mod.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1899"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "idxs = get_cv_idxs(len(df_mod), val_pct=0.05, seed=52)\n",
    "len(idxs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0    2428\n",
      "1    1371\n",
      "Name: is_arg, dtype: int64\n",
      "0    22152\n",
      "1    12044\n",
      "Name: is_arg, dtype: int64\n"
     ]
    }
   ],
   "source": [
    "print(df_mod.iloc[idxs]['is_arg'].value_counts())\n",
    "print(df_mod.loc[~df_mod.index.isin(idxs)]['is_arg'].value_counts())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "arch = resnext101 #plan with resnext101 or dn201\n",
    "sz = 128\n",
    "bs = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "aug_tfms = transforms_top_down"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "tfms = tfms_from_model(arch, sz, aug_tfms=aug_tfms, max_zoom=1.05)\n",
    "data = ImageClassifierData.from_csv('', 'train', 'train_single.csv', bs=bs, tfms=tfms,suffix='.jpg',\n",
    "                                    val_idxs=idxs, continuous=False, skip_header=True, num_workers=8, cat_separator=',')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "%config IPCompleter.greedy=True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "x, _ = next(iter(data.trn_dl))\n",
    "img = data.trn_ds.denorm(to_np(x))[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(128, 128, 3)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100%|██████████████████████████████████████████████████████████████████████████████| 3610/3610 [06:07<00:00, 11.05it/s]\n",
      "100%|████████████████████████████████████████████████████████████████████████████████| 190/190 [00:19<00:00,  7.80it/s]\n"
     ]
    }
   ],
   "source": [
    "learn = ConvLearner.pretrained(arch, data, precompute=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a748d9716df0476d9d4f4819b029bd9f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1, style=ProgressStyle(description_width='initial…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      " 69%|█████████████████████████████████████████████▍                    | 2487/3610 [00:11<00:07, 153.43it/s, loss=1.29]"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#learning rate finder\n",
    "lrf=learn.lr_find()\n",
    "learn.sched.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5a84251bd667439ebfc6c8ed157dca51",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=7, style=ProgressStyle(description_width='initial…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                                                                              \n",
      "    0      0.476359   0.349768   0.874671  \n",
      "    1      0.391839   0.316619   0.88415                                                                               \n",
      "    2      0.48146    0.304126   0.888362                                                                              \n",
      "    3      0.433937   0.29617    0.888889                                                                              \n",
      "    4      0.370368   0.288049   0.888889                                                                              \n",
      "    5      0.359131   0.289188   0.889415                                                                              \n",
      "    6      0.401924   0.29479    0.888889                                                                              \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.29479039786457073, 0.8888888821719885]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lr = 2e-5\n",
    "learn.fit(lr, 3, cycle_len=1, cycle_mult=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "lrs = np.array([lr/9,lr/3,lr])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "539dcc57384443aeb4bb4fd7e44d1cc6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=7, style=ProgressStyle(description_width='initial…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "epoch      trn_loss   val_loss   accuracy                                                                              \n",
      "    0      0.348769   0.274213   0.886783  \n",
      "    1      0.350279   0.268811   0.889942                                                                              \n",
      "    2      0.351396   0.267629   0.893628                                                                              \n",
      "    3      0.315701   0.2637     0.890469                                                                              \n",
      "    4      0.333657   0.265173   0.895208                                                                              \n",
      "    5      0.3177     0.260399   0.895208                                                                              \n",
      "    6      0.322538   0.270528   0.892048                                                                              \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.27052777644645043, 0.8920484390178437]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.unfreeze()\n",
    "learn.fit(lrs, 3, cycle_len=1, cycle_mult=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "eedc1063dcde41c7a32232ad998d50c7",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=7, style=ProgressStyle(description_width='initial…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                                                                              \n",
      "    0      0.439524   0.410666   0.800474  \n",
      "    1      0.441772   0.342032   0.849171                                                                              \n",
      "    2      0.42314    0.359413   0.828902                                                                              \n",
      "    3      0.395329   0.332802   0.848908                                                                              \n",
      "    4      0.350468   0.323857   0.857068                                                                              \n",
      "    5      0.391015   0.313847   0.866017                                                                              \n",
      "    6      0.384389   0.31262    0.861543                                                                              \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.31261991118758753, 0.8615425057899453]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lr = 1e-5\n",
    "lrs = np.array([lr/9,lr/3,lr])\n",
    "learn.unfreeze()\n",
    "learn.fit(lrs, 3, cycle_len=1, cycle_mult=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save(arch.__name__+'planet_single')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In summary, I use the Amazon satellite image problem to pretrain a CNN architecture. In order to make the weights the same, I convert the multi-class classification problem into a binary classification problem by dropping some labels and data. Only label images with or without agriculture. The pretrain weight should be useful for training the oil palm satellite images. "
   ]
  },
  {
   "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}