pytorch_event.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from __future__ import print_function\n",
    "import pandas as pd\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import accuracy_score, precision_score, recall_score\n",
    "import torch\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import torch.nn.init as init\n",
    "import torch.nn.functional as F\n",
    "from torch.autograd import Variable\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# read in the dataset Gluon\n",
    "file_name_g = open('data/g1M.out','r')\n",
    "data_gluon=[]\n",
    "for line in file_name_g.readlines():\n",
    "    data_gluon.append(line.strip().split())\n",
    "file_name_g.close()\n",
    "# read in the dataset Quark\n",
    "file_name_q = open('data/q1M.out','r')\n",
    "data_quark=[]\n",
    "for line in file_name_q.readlines():\n",
    "    data_quark.append(line.strip().split())\n",
    "file_name_q.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "#create a DataFrame, we transform the imported data in Dataframe\n",
    "df_gluon=pd.DataFrame(data_gluon)\n",
    "df_quark=pd.DataFrame(data_quark)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>1256966</td>\n",
       "      <td>1256966</td>\n",
       "      <td>1256966</td>\n",
       "      <td>1256966</td>\n",
       "      <td>1237968</td>\n",
       "      <td>1237968</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>unique</th>\n",
       "      <td>735801</td>\n",
       "      <td>10010</td>\n",
       "      <td>625283</td>\n",
       "      <td>613577</td>\n",
       "      <td>708139</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>top</th>\n",
       "      <td>Event</td>\n",
       "      <td>0.1</td>\n",
       "      <td>(1</td>\n",
       "      <td>elapsed)</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>freq</th>\n",
       "      <td>19736</td>\n",
       "      <td>678143</td>\n",
       "      <td>738</td>\n",
       "      <td>18998</td>\n",
       "      <td>62</td>\n",
       "      <td>699717</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              0        1        2         3        4        5\n",
       "count   1256966  1256966  1256966   1256966  1237968  1237968\n",
       "unique   735801    10010   625283    613577   708139        3\n",
       "top       Event      0.1       (1  elapsed)        0        1\n",
       "freq      19736   678143      738     18998       62   699717"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#concatenate the both dataset. we combine the quark set and gluon set in the same data set\n",
    "data=pd.concat([df_gluon,df_quark])\n",
    "#A Brief description of the dataset\n",
    "data.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    19736\n",
       "1    19736\n",
       "2    19736\n",
       "3    19736\n",
       "4      738\n",
       "5      738\n",
       "dtype: int64"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#From the above description, we observed taht the data set containt a string \"Event\", we have to clean. \n",
    "data[data[0].str.contains('Event')].count()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "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>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.0755577</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.189465</td>\n",
       "      <td>0.104326</td>\n",
       "      <td>0.0544762</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.0639362</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.184013</td>\n",
       "      <td>0.058119</td>\n",
       "      <td>0.0475469</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.0369286</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.139499</td>\n",
       "      <td>0.446972</td>\n",
       "      <td>0.0292832</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.167616</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.383354</td>\n",
       "      <td>0.309005</td>\n",
       "      <td>0.11008</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.218167</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.345461</td>\n",
       "      <td>0.128316</td>\n",
       "      <td>0.110414</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           0    1         2         3          4  5\n",
       "0  0.0755577  0.2  0.189465  0.104326  0.0544762  0\n",
       "1  0.0639362  0.1  0.184013  0.058119  0.0475469  0\n",
       "2  0.0369286  0.1  0.139499  0.446972  0.0292832  0\n",
       "3   0.167616  0.2  0.383354  0.309005    0.11008  0\n",
       "4   0.218167  0.2  0.345461  0.128316   0.110414  0"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Clean data\n",
    "data=data[data[0]!='Event']\n",
    "#we print the first 5 elements of the data set\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "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>nmass</th>\n",
       "      <th>nsub</th>\n",
       "      <th>lha</th>\n",
       "      <th>zg</th>\n",
       "      <th>ci</th>\n",
       "      <th>origin</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.0755577</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.189465</td>\n",
       "      <td>0.104326</td>\n",
       "      <td>0.0544762</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.0639362</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.184013</td>\n",
       "      <td>0.058119</td>\n",
       "      <td>0.0475469</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.0369286</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.139499</td>\n",
       "      <td>0.446972</td>\n",
       "      <td>0.0292832</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.167616</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.383354</td>\n",
       "      <td>0.309005</td>\n",
       "      <td>0.11008</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.218167</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.345461</td>\n",
       "      <td>0.128316</td>\n",
       "      <td>0.110414</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       nmass nsub       lha        zg         ci origin\n",
       "0  0.0755577  0.2  0.189465  0.104326  0.0544762      0\n",
       "1  0.0639362  0.1  0.184013  0.058119  0.0475469      0\n",
       "2  0.0369286  0.1  0.139499  0.446972  0.0292832      0\n",
       "3   0.167616  0.2  0.383354  0.309005    0.11008      0\n",
       "4   0.218167  0.2  0.345461  0.128316   0.110414      0"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#rename column by using the name of variables that describe the event\n",
    "data.rename(columns={0:'nmass',1:'nsub',2:'lha',3:'zg',4:'ci',5:'origin'},inplace=True)\n",
    "#we print the five first element\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "       nmass nsub       lha         zg         ci\n",
      "0  0.0755577  0.2  0.189465   0.104326  0.0544762\n",
      "1  0.0639362  0.1  0.184013   0.058119  0.0475469\n",
      "2  0.0369286  0.1  0.139499   0.446972  0.0292832\n",
      "3   0.167616  0.2  0.383354   0.309005    0.11008\n",
      "4   0.218167  0.2  0.345461   0.128316   0.110414\n",
      "5   0.526424  0.2  0.705519   0.423706   0.280389\n",
      "6  0.0913633  0.2  0.190319  0.0550395  0.0569352\n",
      "7   0.112613  0.2  0.229782   0.125332  0.0655758\n",
      "8   0.503168  0.4  0.697395   0.113624   0.291295\n",
      "9  0.0267763  0.1  0.120125   0.100771  0.0247467\n",
      "0    0\n",
      "1    0\n",
      "2    0\n",
      "3    0\n",
      "4    0\n",
      "5    0\n",
      "6    0\n",
      "7    0\n",
      "8    0\n",
      "9    0\n",
      "Name: origin, dtype: object\n"
     ]
    }
   ],
   "source": [
    "features = data.drop('origin', axis=1)\n",
    "targets = data['origin']\n",
    "print(features[:10])\n",
    "print(targets[:10])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "features=features.astype('float')\n",
    "targets=targets.astype('float')\n",
    " "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "#We split the data set by taking 80% of data for training and 20% for testing. Note we use a rondom sampling\n",
    "X_train, X_test, y_train, y_test = train_test_split(features, targets, test_size=0.2, random_state=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# wrap up with Variable in pytorch\n",
    "train_X = Variable(torch.Tensor(X_train.values).float())\n",
    "test_X = Variable(torch.Tensor(X_test.values).float())\n",
    "train_y = Variable(torch.Tensor(y_train.values).float())\n",
    "test_y = Variable(torch.Tensor(y_test.values).float())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sequential(\n",
      "  (0): Linear(in_features=5, out_features=4, bias=True)\n",
      "  (1): ELU(alpha=1.0)\n",
      "  (2): Linear(in_features=4, out_features=1, bias=True)\n",
      "  (3): Sigmoid()\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "# Define network dimensions\n",
    "n_input_dim = train_X .shape[1]\n",
    "# Layer size\n",
    "n_hidden = 4 # Number of hidden nodes\n",
    "n_output = 1 # Number of output nodes = for binary classifier\n",
    "\n",
    "# Build your network\n",
    "net = nn.Sequential(\n",
    "    nn.Linear(n_input_dim, n_hidden),\n",
    "    nn.ELU(),\n",
    "    nn.Linear(n_hidden, n_output),\n",
    "    nn.Sigmoid())\n",
    "    \n",
    "print(net)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Define the lost function \n",
    "loss_func = nn.BCELoss()\n",
    "learning_rate = 0.2\n",
    "#Set the optimize, we choose Adam\n",
    "optimizer = torch.optim.Adam(net.parameters(), lr=learning_rate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/nick/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:15: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number\n",
      "  from ipykernel import kernelapp as app\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "number of epoch 0 loss tensor(0.5034)\n",
      "number of epoch 10 loss tensor(0.5868)\n",
      "number of epoch 20 loss tensor(0.5119)\n",
      "number of epoch 30 loss tensor(0.5192)\n",
      "number of epoch 40 loss tensor(0.5051)\n",
      "number of epoch 50 loss tensor(0.5061)\n",
      "number of epoch 60 loss tensor(0.5036)\n",
      "number of epoch 70 loss tensor(0.5031)\n",
      "number of epoch 80 loss tensor(0.5028)\n",
      "number of epoch 90 loss tensor(0.5026)\n",
      "number of epoch 100 loss tensor(0.5024)\n",
      "number of epoch 110 loss tensor(0.5022)\n",
      "number of epoch 120 loss tensor(0.5021)\n",
      "number of epoch 130 loss tensor(0.5020)\n",
      "number of epoch 140 loss tensor(0.5018)\n",
      "number of epoch 150 loss tensor(0.5017)\n",
      "number of epoch 160 loss tensor(0.5016)\n",
      "number of epoch 170 loss tensor(0.5014)\n",
      "number of epoch 180 loss tensor(0.5013)\n",
      "number of epoch 190 loss tensor(0.5012)\n",
      "number of epoch 200 loss tensor(0.5011)\n",
      "number of epoch 210 loss tensor(0.5009)\n",
      "number of epoch 220 loss tensor(0.5008)\n",
      "number of epoch 230 loss tensor(0.5007)\n",
      "number of epoch 240 loss tensor(0.5005)\n",
      "Finished training!\n"
     ]
    }
   ],
   "source": [
    "#We run the train \n",
    "train_loss = []\n",
    "epoch = 250\n",
    "Y_train_t = torch.FloatTensor(train_y).reshape(-1, 1)\n",
    "for epoch in range(epoch):\n",
    "    net.cpu()\n",
    "    X_train_t = torch.FloatTensor(train_X).to(device='cpu')\n",
    "    y_hat = net(X_train_t).to(device='cpu')\n",
    "    loss = loss_func(y_hat, Y_train_t).to(device='cpu')\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "    optimizer.zero_grad()\n",
    "    train_loss.append(loss.item())\n",
    "    if epoch % 10 == 0:\n",
    "        print ('number of epoch', epoch, 'loss', loss.data[0])\n",
    "print(\"Finished training!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[0.8511],\n",
       "        [0.6902],\n",
       "        [0.2966],\n",
       "        ...,\n",
       "        [0.5100],\n",
       "        [0.8115],\n",
       "        [0.3953]], grad_fn=<SigmoidBackward>)"
      ]
     },
     "execution_count": 127,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_hat"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[1.],\n",
       "        [1.],\n",
       "        [0.],\n",
       "        ...,\n",
       "        [1.],\n",
       "        [1.],\n",
       "        [0.]], grad_fn=<RoundBackward>)"
      ]
     },
     "execution_count": 128,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#We convert the sigmoid result to a corresponding class\n",
    "y_hat=torch.round(y_hat)\n",
    "y_hat"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 130,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "prediction accuracy 0.7555719227629463\n",
      "macro precision 0.7879921313377632\n"
     ]
    }
   ],
   "source": [
    "print ('prediction accuracy', accuracy_score(train_y, y_hat.detach().numpy()))\n",
    "print ('macro precision', precision_score(train_y, y_hat.detach().numpy()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "# Plot training & validation accuracy values\n",
    "plt.plot(train_loss,'-')\n",
    "plt.title('Model lost')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.xlabel('Epoch')\n",
    "plt.grid(True)\n",
    "plt.savefig(\"lost.png\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our network trained over the 1,000 iterations and pushed its classification accuracy towards 75%. Let’s  visualize this to see how well it fares on our test set. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_test_t = torch.FloatTensor(test_X)\n",
    "y_hat_test = net(X_test_t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Create the class, we could also use torch.round()\n",
    "y_hat_test=np.where(y_hat_test<0.5,0,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test accuracy 0.7474075151750281\n",
      "macro precision 0.7433905050611125\n",
      "micro precision 0.7474075151750281\n",
      "macro recall 0.7396644879878598\n",
      "micro recall 0.7474075151750281\n"
     ]
    }
   ],
   "source": [
    "print ('Test accuracy', accuracy_score(test_y , y_hat_test))\n",
    "print ('macro precision', precision_score(test_y.data, y_hat_test.data, average='macro'))\n",
    "print ('micro precision', precision_score(test_y.data, y_hat_test.data, average='micro'))\n",
    "print ('macro recall', recall_score(test_y.data, y_hat_test.data, average='macro'))\n",
    "print ('micro recall', recall_score(test_y.data, y_hat_test.data, average='micro'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We could improve the network by using, dropout, or regularization and see the result. To avoid overfitting in the begining of the training"
   ]
  }
 ],
 "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": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}