tensorflow/Reuters News Classification using LSTM (long-short tem memory) in Tensorflow.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Reuters News Classification"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We are going to implement text classifcation on Reuters News (available through sklearn) based on LSTM method "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.datasets import reuters\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, LSTM, Embedding\n",
    "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint\n",
    "from tensorflow.keras.models import load_model\n",
    "\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 1. Load Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/reuters.npz\n",
      "2113536/2110848 [==============================] - 0s 0us/step\n"
     ]
    }
   ],
   "source": [
    "# num_words: 1000 to use the words up to 1000th frequently appearing. Put 0 to use all words\n",
    "(X_train, y_train), (X_test, y_test) = reuters.load_data(num_words=1000, test_split=0.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of Training News : 8982\n",
      "Number of Test News : 2246\n",
      "Number of Predictor Classes : 46\n"
     ]
    }
   ],
   "source": [
    "print('Number of Training News : {}'.format(len(X_train)))\n",
    "print('Number of Test News : {}'.format(len(X_test)))\n",
    "\n",
    "num_classes = max(y_train) + 1\n",
    "print('Number of Predictor Classes : {}'.format(num_classes))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1, 2, 2, 8, 43, 10, 447, 5, 25, 207, 270, 5, 2, 111, 16, 369, 186, 90, 67, 7, 89, 5, 19, 102, 6, 19, 124, 15, 90, 67, 84, 22, 482, 26, 7, 48, 4, 49, 8, 864, 39, 209, 154, 6, 151, 6, 83, 11, 15, 22, 155, 11, 15, 7, 48, 9, 2, 2, 504, 6, 258, 6, 272, 11, 15, 22, 134, 44, 11, 15, 16, 8, 197, 2, 90, 67, 52, 29, 209, 30, 32, 132, 6, 109, 15, 17, 12]\n"
     ]
    }
   ],
   "source": [
    "# Examine the very first news (Already converted in integers)\n",
    "print(X_train[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 2. EDA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Max length of news :2376\n",
      "Avg length of news :145.5398574927633\n"
     ]
    },
    {
     "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": [
    "# We also want to check the maximum and average length of news\n",
    "print('Max length of news :{}'.format(max(len(a) for a in X_train)))\n",
    "print('Avg length of news :{}'.format(sum(map(len, X_train))/len(X_train)))\n",
    "\n",
    "plt.hist([len(l) for l in X_train], bins=50)\n",
    "plt.xlabel('length of samples')\n",
    "plt.ylabel('number of samples')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x22ed29dbb88>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAt0AAAEyCAYAAAAmzWw4AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzt3Xu8ZFV14PHfkpfiC5AGsSFpNK0jJhGdDmJMFCHhpaEBIYH4IIiDURA0OgbiTEAZJj6CjBglQUFAUUSeLbYBQjAmE3k0yqOhA7RIpAHpNiiY8AlO65o/zm4sbledOufeu7v7Xn7fz6c+99SuvWrvqlq3atWpXaciM5EkSZJUz1PW9wQkSZKk2c6iW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqbOP1PYEatt5665w3b976noYkSZJmuRtvvPGHmTlnXL9ZWXTPmzePJUuWrO9pSJIkaZaLiH/t0s/lJZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRuv7wnoF+7/1Hs7933eUadUnIkkSZKmk3u6JUmSpMosuiVJkqTKLLolSZKkyiy6JUmSpMosuiVJkqTKLLolSZKkyiy6JUmSpMosuiVJkqTKLLolSZKkyqoV3RHx1Ii4PiJujojbIuKDpX3HiLguIu6KiC9HxKalfbNyfnm5fN7AdR1f2u+IiL1qzVmSJEmqoeae7seA3TPzpcDOwN4RsSvwEeDUzJwP/Ag4ovQ/AvhRZv4KcGrpR0TsBBwCvATYG/h0RGxUcd6SJEnStKpWdGfj38vZTcopgd2BC0v7OcD+ZXthOU+5fI+IiNJ+fmY+lpnfA5YDu9SatyRJkjTdqq7pjoiNIuImYCVwFfBd4MeZubp0WQHMLdtzgXsByuUPA88ZbB8SMzjWkRGxJCKWrFq1qsbNkSRJkialatGdmT/LzJ2B7Wn2Tr94WLfyN0ZcNqp94lhnZOaCzFwwZ86cyU5ZkiRJmnbr5Oglmflj4BvArsAWEbFxuWh74P6yvQLYAaBc/mzgocH2ITGSJEnSBq/m0UvmRMQWZftpwO8Ay4BrgINKt8OAy8r2onKecvnfZ2aW9kPK0U12BOYD19eatyRJkjTdNh7fZdK2A84pRxp5CnBBZl4eEbcD50fE/wK+A5xZ+p8JfD4iltPs4T4EIDNvi4gLgNuB1cBRmfmzivOWJEmSplW1ojszbwFeNqT9boYcfSQz/xM4eMR1nQycPN1zlCRJktYFf5FSkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSaqsWtEdETtExDURsSwibouIY0v7iRFxX0TcVE77DsQcHxHLI+KOiNhroH3v0rY8Io6rNWdJkiSpho0rXvdq4L2Z+e2IeCZwY0RcVS47NTP/crBzROwEHAK8BHge8HcR8cJy8aeA3wVWADdExKLMvL3i3CVJkqRpU63ozswHgAfK9k8iYhkwtyVkIXB+Zj4GfC8ilgO7lMuWZ+bdABFxfulr0S1JkqQZYZ2s6Y6IecDLgOtK09ERcUtEnBURW5a2ucC9A2ErStuo9oljHBkRSyJiyapVq6b5FkiSJEmTV73ojohnABcB787MR4DTgRcAO9PsCT9lTdch4dnS/sSGzDMyc0FmLpgzZ860zF2SJEmaDjXXdBMRm9AU3Odl5sUAmfngwOWfAS4vZ1cAOwyEbw/cX7ZHtUuSJEkbvJpHLwngTGBZZn58oH27gW4HAEvL9iLgkIjYLCJ2BOYD1wM3APMjYseI2JTmy5aLas1bkiRJmm4193S/CngzcGtE3FTa/gw4NCJ2plkicg/wdoDMvC0iLqD5guRq4KjM/BlARBwNXAFsBJyVmbdVnLckSZI0rWoeveSfGL4ee3FLzMnAyUPaF7fFSZIkSRsyf5FSkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSarMoluSJEmqzKJbkiRJqsyiW5IkSaqsWtEdETtExDURsSwibouIY0v7VhFxVUTcVf5uWdojIk6LiOURcUtEvHzgug4r/e+KiMNqzVmSJEmqoeae7tXAezPzxcCuwFERsRNwHHB1Zs4Hri7nAfYB5pfTkcDp0BTpwAnAK4BdgBPWFOqSJEnSTFCt6M7MBzLz22X7J8AyYC6wEDindDsH2L9sLwTOzca1wBYRsR2wF3BVZj6UmT8CrgL2rjVvSZIkabqtkzXdETEPeBlwHbBtZj4ATWEObFO6zQXuHQhbUdpGtU8c48iIWBIRS1atWjXdN0GSJEmatOpFd0Q8A7gIeHdmPtLWdUhbtrQ/sSHzjMxckJkL5syZM7nJSpIkSRVULbojYhOagvu8zLy4ND9Ylo1Q/q4s7SuAHQbCtwfub2mXJEmSZoSaRy8J4ExgWWZ+fOCiRcCaI5AcBlw20P6WchSTXYGHy/KTK4A9I2LL8gXKPUubJEmSNCNsXPG6XwW8Gbg1Im4qbX8GfBi4ICKOAL4PHFwuWwzsCywHHgUOB8jMhyLiJOCG0u9DmflQxXnPOPectv/4TgPmHXNppZlIkiRpmGpFd2b+E8PXYwPsMaR/AkeNuK6zgLOmb3aSJEnSuuMvUkqSJEmVWXRLkiRJlVl0S5IkSZVZdEuSJEmVWXRLkiRJlVl0S5IkSZVZdEuSJEmVWXRLkiRJlVl0S5IkSZVZdEuSJEmVWXRLkiRJlXUquiPi6i5tkiRJkta2cduFEfFUYHNg64jYEohy0bOA51WemyRJkjQrtBbdwNuBd9MU2Dfyi6L7EeBTFeclSZIkzRqtRXdmfgL4RES8KzM/uY7mJEmSJM0q4/Z0A5CZn4yI3wTmDcZk5rmV5iVJkiTNGp2K7oj4PPAC4CbgZ6U5AYtuSZIkaYxORTewANgpM7PmZCRJkqTZqOtxupcCz605EUmSJGm26rqne2vg9oi4HnhsTWNm7ldlVpIkSdIs0rXoPrHmJCRJkqTZrOvRS/6h9kQkSZKk2arr0Ut+QnO0EoBNgU2A/8jMZ9WamCRJkjRbdN3T/czB8xGxP7BLlRlJkiRJs0zXo5c8QWZeCuw+zXORJEmSZqWuy0sOHDj7FJrjdnvMbkmSJKmDrkcv+b2B7dXAPcDCaZ+NJEmSNAt1XdN9eO2JSJIkSbNVpzXdEbF9RFwSESsj4sGIuCgitq89OUmSJGk26PpFys8Bi4DnAXOBr5Y2SZIkSWN0LbrnZObnMnN1OZ0NzKk4L0mSJGnW6Fp0/zAi3hQRG5XTm4B/qzkxSZIkabboWnS/Ffh94AfAA8BBQOuXKyPirLIGfOlA24kRcV9E3FRO+w5cdnxELI+IOyJir4H2vUvb8og4rs+NkyRJkjYEXYvuk4DDMnNOZm5DU4SfOCbmbGDvIe2nZubO5bQYICJ2Ag4BXlJiPr1mrzrwKWAfYCfg0NJXkiRJmjG6Ft2/npk/WnMmMx8CXtYWkJnfBB7qeP0LgfMz87HM/B6wnOZn5ncBlmfm3Zn5U+B8PD64JEmSZpiuRfdTImLLNWciYiu6/7DOREdHxC1l+cma65wL3DvQZ0VpG9W+log4MiKWRMSSVatWTXJqkiRJ0vTrWnSfAvxzRJwUER8C/hn46CTGOx14AbAzzdrwU0p7DOmbLe1rN2aekZkLMnPBnDkeWEWSJEkbjq6/SHluRCwBdqcphA/MzNv7DpaZD67ZjojPAJeXsyuAHQa6bg/cX7ZHtUuSJEkzQuclIqXI7l1oD4qI7TLzgXL2AGDNkU0WAV+MiI/T/ADPfOB6mgJ/fkTsCNxH82XLP5zKHCRJkqR1bbLrsseKiC8BuwFbR8QK4ARgt4jYmWaJyD3A2wEy87aIuICmqF8NHJWZPyvXczRwBbARcFZm3lZrzpIkSVIN1YruzDx0SPOZLf1PBk4e0r4YWDyNU5MkSZLWqa5fpJQkSZI0SRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmUW3ZIkSVJlFt2SJElSZRbdkiRJUmXViu6IOCsiVkbE0oG2rSLiqoi4q/zdsrRHRJwWEcsj4paIePlAzGGl/10RcVit+UqSJEm11NzTfTaw94S244CrM3M+cHU5D7APML+cjgROh6ZIB04AXgHsApywplCXJEmSZopqRXdmfhN4aELzQuCcsn0OsP9A+7nZuBbYIiK2A/YCrsrMhzLzR8BVrF3IS5IkSRu0db2me9vMfACg/N2mtM8F7h3ot6K0jWpfS0QcGRFLImLJqlWrpn3ikiRJ0mRtKF+kjCFt2dK+dmPmGZm5IDMXzJkzZ1onJ0mSJE3Fui66HyzLRih/V5b2FcAOA/22B+5vaZckSZJmjHVddC8C1hyB5DDgsoH2t5SjmOwKPFyWn1wB7BkRW5YvUO5Z2iRJkqQZY+NaVxwRXwJ2A7aOiBU0RyH5MHBBRBwBfB84uHRfDOwLLAceBQ4HyMyHIuIk4IbS70OZOfHLmZIkSdIGrVrRnZmHjrhojyF9EzhqxPWcBZw1jVOTJEmS1qkN5YuUkiRJ0qxl0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFW28fqegCTNBPte8tFe/Rcf8P5KM5EkzUTrZU93RNwTEbdGxE0RsaS0bRURV0XEXeXvlqU9IuK0iFgeEbdExMvXx5wlSZKkyVqfy0tem5k7Z+aCcv444OrMnA9cXc4D7APML6cjgdPX+UwlSZKkKdiQ1nQvBM4p2+cA+w+0n5uNa4EtImK79TFBSZIkaTLWV9GdwJURcWNEHFnats3MBwDK321K+1zg3oHYFaXtCSLiyIhYEhFLVq1aVXHqkiRJUj/r64uUr8rM+yNiG+CqiPiXlr4xpC3Xasg8AzgDYMGCBWtdLkmSJK0v62VPd2beX/6uBC4BdgEeXLNspPxdWbqvAHYYCN8euH/dzVaSJEmamnVedEfE0yPimWu2gT2BpcAi4LDS7TDgsrK9CHhLOYrJrsDDa5ahSJIkSTPB+lhesi1wSUSsGf+Lmfm3EXEDcEFEHAF8Hzi49F8M7AssBx4FDl/3U5ZUwz6L9uvc9+v7Lao4E0mS6lrnRXdm3g28dEj7vwF7DGlP4Kh1MLXHrfrrv+nVf84fv73STCRJkjQbbEiHDJQkSZJmJYtuSZIkqTKLbkmSJKmy9XWcbmnWO/PcPXv1P+ItV1aaiSRJWt/c0y1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFXmL1JKUmWvu/i0Xv2/duAxlWYiSVpf3NMtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVWbRLUmSJFVm0S1JkiRVZtEtSZIkVeZxuqVZ5KNf2qtX//cfekWlmUiSpEHu6ZYkSZIqc0+3Zoyvn7lvr/77HLG40kwkSZL6cU+3JEmSVJl7uqUW553db430G//INdKSJGlt7umWJEmSKnNPt54ULv7c3r36H3j431aaidTP6y76m859v/aGt1ecSV2/d+Flnft+9aCFFWciSXW4p1uSJEmqzD3d0+zB0z/aq/+273h/pZlIGmbfS07o1X/xAR+sNBNJ0pPJrC66V53+hV7957zjTZVmIkmz3+sv/Eqv/pcfdHClmUjShmfGFN0RsTfwCWAj4LOZ+eH1PKUntev+5vW9+r/i7ZdXmsns9Fdf6H7UlKPf5BFTNL1ef+F5nfteftAbK85kw3TgRd/q1f/iN7xyWsb9g4vv7tz3ywc+f1rGXB8u+8oPO/ddePDW0zLmt85Z1av/Kw+bMy3j6sllRhTdEbER8Cngd4EVwA0RsSgzb1+/M5O0Puxz6TG9+n99/9MqzUTa8B1/yX29+v/FAXMf3/7EJT/oHHfsAc/tNY7WvR+cclev/s997/zHtx889ebOcdu+56W9xnmymBFFN7ALsDwz7waIiPOBhYBF9xTd+un9Ovf9tXcumpYxr/ns6zr3fe3bvjYtY2q8/3lB9yO8nPT7vzi6yzsv7ndkmE8f6JFhNH0WXtgvny47qF++TreDLupeuABc+AaLlw3ZzZ9Z2av/S//bNo9vL//kg71if+Vd2z6+/cBHHugct92fbtdrnA3Jg6d9o3PfbY/ZbVrGXPmpS3r13+aoAzr3nSlF91zg3oHzK4BXrKe5SNIG7/UXnd2r/+Vv+KMq89iQ7X/RNb36X/qG11aayezz+Yv7Ldd484FTX67xd1/sN+bv/KFLRNaFBz9xba/+2x6765THXPlXX+/Vf5uj95nymF1EZq6TgaYiIg4G9srMt5XzbwZ2ycx3DfQ5EjiynH0RcEfLVW4NdF80NvW4mTbmVGIdc3aNOZVYx5xdY04l1jFn15hTiXXM2TXmVGJn05i/nJnj38Vl5gZ/Al4JXDFw/njg+Clc35J1GTfTxpxp83XMDTPWMWfXmDNtvo65YcY65uwac6bNd33dR2tOM+XHcW4A5kfEjhGxKXAIMD0LjCVJkqTKZsSa7sxcHRFHA1fQHDLwrMy8bT1PS5IkSepkRhTdAJm5GFg8TVd3xjqOm2ljTiXWMWfXmFOJdczZNeZUYh1zdo05lVjHnF1jTiX2yTLm42bEFyklSZKkmWymrOmWJEmSZiyLbkmSJKmyJ1XRHRF7R8QdEbE8Io7rEXdWRKyMiKU9x9shIq6JiGURcVtEHNsj9qkRcX1E3FxiP9hz7I0i4jsRcXnPuHsi4taIuCkilvSM3SIiLoyIfym3+ZUdYl5UxlpzeiQi3t1jzPeU+2dpRHwpIp7aMe7YEnPbuPGGPf4RsVVEXBURd5W/W/aIPbiM+/OIWNAj7mPlvr0lIi6JiC16xJ5U4m6KiCsj4nld4gYue19EZERs3WPMEyPivoHHdt+uY0bEu8r/6m0R8dEeY355YLx7IuKmHrE7R8S1a3I/InbpGPfSiPhW+b/5akQ8a0jc0OeCLnnUEtuaRy1xY/OoJbZLHrU+743KpZYxu+TRyDHbcqllzLF51BLbmkctcV3yaOjrQjRH9rqu5NGXoznKV5e4o6N5PWz73x4Ve165X5dG83+xSY/YM0vbLdG8ZjyjS9zA5Z+MiH/vOd+zI+J7A4/rzh3jIiJOjog7y2N2TI8x/3FgvPsj4tKOcXtExLdL3D9FxK/0GHP3Ers0Is6JiKHf3YsJNcK4HBoTOzaPRsSNzaGW2NYcGhU30D4yh1rGbM2hTqZ6zMGZcqI56sl3gecDmwI3Azt1jH018HJgac8xtwNeXrafCdzZY8wAnlG2NwGuA3btMfafAF8ELu8553uArSd5H58DvK1sbwpsMYnH6Ac0B5nv0n8u8D3gaeX8BcAfdYj7VWApsDnNl4n/Dpjf5/EHPgocV7aPAz7SI/bFND/g9A1gQY+4PYGNy/ZHeo75rIHtY4C/7prnwA40Rw7611G5MWLME4H3jXkshsW9tjwmm5Xz23SNnXD5KcCf9xj3SmCfsr0v8I2OcTcArynbbwVOGhI39LmgSx61xLbmUUvc2Dxqie2SRyOf99pyqWXMLnk0KrY1l9rmOi6PWsZszaOWuC55NPR1gea575DS/tfAOzrGvQyYR8vzfkvsvuWyAL40ccwxsYN59HHK/8C4uHJ+AfB54N97zvds4KCWHBoVdzhwLvCUYTk0br4DfS4C3tJxzDuBF5f2dwJndxzzN2l+vfuFpf1DwBEjbu8TaoRxOTQmdmwejYgbm0Mtsa05NCquSw61jNmaQ11OT6Y93bsAyzPz7sz8KXA+sLBLYGZ+E3io74CZ+UBmfrts/wRYRlModonNzFzzLmyTcur0rdeI2B54HfDZvnOerGj2yrwaOBMgM3+amT/ueTV7AN/NzH/tEbMx8LTybn5z4P4OMS8Grs3MRzNzNfAPwAGjOo94/BfSvMmg/N2/a2xmLsvMtl9MHRV3ZZkvwLXA9j1iHxk4+3SG5FJLnp8KvH9YTIfYViPi3gF8ODMfK31W9h0zIgL4fZon8q6xCazZu/hshuTSiLgXAd8s21cBbxgSN+q5YGwejYodl0ctcWPzqCW2Sx61Pe+NzKUpPl+Oim3NpXFjtuVRS2xrHrXEdcmjUa8LuwMXlva18mhUXGZ+JzPvmThOx9jF5bIErmd4Ho2KfQQev3+fxoR8GBUXERsBH6PJoV7zbbuNY+LeAXwoM39e+q31fDRuzIh4Js1jdGnHuC7PRcNifwY8lpl3lvaheTSxRiiPQ2sOjYotcxmbRyPixuZQS2xrDo2K65JDo2Knw5Op6J5L8w5wjRV0fEKfDhExj+bd4HU9YjaK5mPNlcBVmdk19v/QJNTPe04TmsS9MiJujIgje8Q9H1gFfK58HPPZiHh6z7EPYUSRNHSimfcBfwl8H3gAeDgzr+wQuhR4dUQ8JyI2p3m3vUPPuW6bmQ+UeTwAbNMzfqreCny9T0D5iPRe4I3An3eM2Q+4LzNv7j9FAI4uH/+dFSOW4AzxQuC3y0ed/xARvzGJcX8beDAz7+oR827gY+U++kuaX77tYimwX9k+mDG5NOG5oFceTeZ5ZEzc2DyaGNsnjwZj++TSkPl2zqMJsZ1zacR91CmPJsR2zqMJcZ3yaOLrAs0nuD8eeCM19LVtCq8nrbFlScCbgb/tExsRn6P5ZPO/AJ/sGHc0sGjN/8wk5ntyyaNTI2KzjnEvAP4gmqVCX4+I+X3vI5qdOldPeNPaFvc2YHFErKC5bz/cZUyawnWT+MVys4MYnkcTa4Tn0CGHRsR2NTJuXA6Nih2XQyPiOuVQy3xbc2icJ1PRHUPaOu05nvLAzVqji4B3D/unGyUzf5aZO9O8+9slIn61w1ivB1Zm5o2TnO6rMvPlwD7AURHx6o5xG9N87H56Zr4M+A+aj8s7iWb92H7AV3rEbEmzp3BH4HnA0yPiTePiMnMZzcfqV9H8k98MrG4N2oBExAdo5nten7jM/EBm7lDiju4wzubAB+hYoA9xOs0L1s40b4pO6Ri3MbAlzces/x24oOzN6ONQeryBK94BvKfcR++hfGrTwVtp/ldupFku8NNRHSf7XDCV2FFxXfJoWGzXPBqMLeN0yqUhY3bOoyGxnXKp5b4dm0dDYjvl0ZC4Tnk08XWB5pO7tbqNi+vyetIx9tPANzPzH/vEZubhNM/by4A/6BD3apo3I8OKqy5jHk9TnP0GsBXwpx3jNgP+MzMXAJ8BzupzO4uReTQi7j3Avpm5PfA5muUTY2OBl9DsvDo1Iq4HfsKE17YRNUKn+miy9UWHuJE51BbblkPD4qL5DsrYHGoZc2wOjZVTWJsyk07AK4ErBs4fDxzfI34ePdd0l7hNaNYw/skU538CY9Y1ln5/QfMu9R6ad4CPAl+Y5Jgndhmz9H0ucM/A+d8GvtZjrIXAlT3ndzBw5sD5twCfnsTt/N/AO/s8/sAdwHZlezvgjr65Q8ua7lFxwGHAt4DN+8x3wmW/3HLZ43HAr9HsQbmnnFbTfKrw3EmM2fkymjdCuw2c/y4wp8d9tDHwILB9z8f0YXj8twsCeGQSt+WFwPUjLlvruaBrHg2L7ZJHo+K65FHbmB3y6AmxXXOpw5ht9/2w+3dsLrXcR2PzaMSYY/Oow+0cmUcT+p1A82bih/xinf4TXuta4t43cP4eOn6XZzC2bF9KWe/cJ3ag7TWM+e5RiTuB5jVtTQ79nGbJ6GTG3K3jmO8D/gWYN/B4PtzzPnoO8G/AU3s8nt8daPsl4PZJ3s49gQsmtA2rEc7rkkMjYr8wcPnQPGqLG5dD48YclUMj4n7UJYc6jjk2h4adnkx7um8A5kfzDd1Nad4NLqo5YNmjciawLDOHvlNtiZ0T5agCEfE04Hdo/vlbZebxmbl9Zs6juY1/n5lj9/6WcZ4ezdozytKQPWk+8hwrM38A3BsRLypNewC3d4ktJrNn8vvArhGxebmv96B5xztWRGxT/v4ScOAkxl5EU7hQ/l7WM763iNib5p31fpn5aM/YwY9E96NbLt2amdtk5rySTytovgD2g45jbjdw9gA65hLNE/Du5TpeSPOl3B92jIXyv5KZK3rEQLNu8jVle3eg09KUgVx6CvA/aL6ENLHPqOeCsXk02eeRUXERkBC7AAACoklEQVRd8qgldmweDYvtkkstY47No5b7qDWXxty3rXnUEtuaRy23s0seDXtdWAZcQ7OUAIbk0WRfT9piI+JtwF7AoVnWO3eMvSPK0TjKffF7E+cyIu7GzHzuQA49mpnDjuoxar7bDYy5PxPyqOU+ejyHaB7XO5lgzP17ME1x9p8d45YBzy75CvC7DHlda7mda/JoM5r/8yfk0Yga4Y2MyaGW2C6fLg+N65JDw2KBN4/LoRFjbtklh1rm25pDnfSt0mfyiWbt7p00ezs+0CPuSzQfa/4/mheLod8GHhL3WzQf0dwC3FRO+3aM/XXgOyV2KSOOwjDmOnajxzsxmnXZN5fTbX3uoxK/M7CkzPlSYMuOcZvT7Al49iRu4wdp/tmW0nwbebOOcf9I86bgZmCPvo8/zd6Lq2leUK8GtuoRe0DZfoxmT9qwPQrD4pbTfC9hTS6tdeSIltiLyn10C/BVmi/F9cpz2o9wMGzMzwO3ljEXUfbodojbFPhCme+3gd37/F/SfMP8jyfxmP4WcGPJieuA/9ox7lia55U7adZexpC4oc8FXfKoJbY1j1rixuZRS2yXPBr7vDcsl1rG7JJHo2Jbc6ltruPyqGXM1jxqieuSR0NfF2ieu68vj+1XmPA82BJ3TMmh1TRvFj7bY8zVNK+la27DsCO8rBVLs6z1/5bHdCnNXtZndRlzQp9RRy8ZNd+/HxjzC5Qjf3SI2wL4Won9FvDSrmOWy74B7N1zrgeU8W4u8c/vEfsxmiL9DpqlS23Pg7vxiyNztObQmNixeTQibmwODYvtkkOjxuySQy3zbc2hLid/Bl6SJEmq7Mm0vESSJElaLyy6JUmSpMosuiVJkqTKLLolSZKkyiy6JUmSpMosuiVJkqTKLLolSZKkyv4/BXNg3yMKwLAAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 864x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Check how many each classes have\n",
    "fig, axe = plt.subplots(ncols=1)\n",
    "fig.set_size_inches(12,5)\n",
    "sns.countplot(y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see the class 3 and 4 has the most number of news"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 3. PreProcessing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The length of the news are all different. We want to make the length equal for all news. \n",
    "we gave 100 to max_len in pad_sequences(), which means that we want to make the length of all articles to 100 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_len = 100\n",
    "X_train = pad_sequences(X_train, maxlen=max_len) # Training set padding\n",
    "X_test = pad_sequences(X_test, maxlen=max_len) # Test set padding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_train = to_categorical(y_train) # Lables in Training set One-hot encoding\n",
    "y_test = to_categorical(y_test) # Lables in Testing set One-hot encoding"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 4. Modeling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We first create 'Embedding()' layer which receives 2 inputs: first the size of the word-cluster, and second the dimension of vector. So, below codes generated 1,000 vector that has 120 dimension. LSTM layer gets the hidden layer's size \n",
    "\n",
    "The last output layers has to get the num_classes of predictor. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(Embedding(1000, 120))\n",
    "model.add(LSTM(120))\n",
    "model.add(Dense(num_classes, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create earlystopping and checkpoint on saving best model (only when val_acc improves)\n",
    "es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=4) \n",
    "mc = ModelCheckpoint('best_model.h5', monitor='val_acc', mode='max', verbose=1, save_best_only=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Use 'categorical_crossentropy' on multiclass classification problem\n",
    "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['acc'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 2.6242 - acc: 0.3377\n",
      "Epoch 00001: val_acc improved from -inf to 0.36198, saving model to best_model.h5\n",
      "71/71 [==============================] - 10s 136ms/step - loss: 2.6242 - acc: 0.3377 - val_loss: 2.3547 - val_acc: 0.3620\n",
      "Epoch 2/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 2.1471 - acc: 0.4565\n",
      "Epoch 00002: val_acc improved from 0.36198 to 0.50490, saving model to best_model.h5\n",
      "71/71 [==============================] - 11s 160ms/step - loss: 2.1471 - acc: 0.4565 - val_loss: 2.0005 - val_acc: 0.5049\n",
      "Epoch 3/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.8976 - acc: 0.5106- ETA: 3s - loss: 1.9025 \n",
      "Epoch 00003: val_acc improved from 0.50490 to 0.51959, saving model to best_model.h5\n",
      "71/71 [==============================] - 18s 256ms/step - loss: 1.8976 - acc: 0.5106 - val_loss: 1.8414 - val_acc: 0.5196\n",
      "Epoch 4/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.7766 - acc: 0.5340\n",
      "Epoch 00004: val_acc improved from 0.51959 to 0.54853, saving model to best_model.h5\n",
      "71/71 [==============================] - 13s 180ms/step - loss: 1.7766 - acc: 0.5340 - val_loss: 1.7564 - val_acc: 0.5485\n",
      "Epoch 5/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.6879 - acc: 0.5650\n",
      "Epoch 00005: val_acc improved from 0.54853 to 0.56411, saving model to best_model.h5\n",
      "71/71 [==============================] - 12s 162ms/step - loss: 1.6879 - acc: 0.5650 - val_loss: 1.6917 - val_acc: 0.5641\n",
      "Epoch 6/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.5844 - acc: 0.5986\n",
      "Epoch 00006: val_acc improved from 0.56411 to 0.59439, saving model to best_model.h5\n",
      "71/71 [==============================] - 11s 153ms/step - loss: 1.5844 - acc: 0.5986 - val_loss: 1.6126 - val_acc: 0.5944\n",
      "Epoch 7/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.4732 - acc: 0.6284\n",
      "Epoch 00007: val_acc improved from 0.59439 to 0.63713, saving model to best_model.h5\n",
      "71/71 [==============================] - 18s 247ms/step - loss: 1.4732 - acc: 0.6284 - val_loss: 1.5107 - val_acc: 0.6371\n",
      "Epoch 8/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.3861 - acc: 0.6523\n",
      "Epoch 00008: val_acc did not improve from 0.63713\n",
      "71/71 [==============================] - 14s 193ms/step - loss: 1.3861 - acc: 0.6523 - val_loss: 1.4858 - val_acc: 0.6220\n",
      "Epoch 9/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.3003 - acc: 0.6749\n",
      "Epoch 00009: val_acc improved from 0.63713 to 0.65583, saving model to best_model.h5\n",
      "71/71 [==============================] - 15s 205ms/step - loss: 1.3003 - acc: 0.6749 - val_loss: 1.3819 - val_acc: 0.6558\n",
      "Epoch 10/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.2200 - acc: 0.6921- ETA: 5\n",
      "Epoch 00010: val_acc improved from 0.65583 to 0.65806, saving model to best_model.h5\n",
      "71/71 [==============================] - 14s 199ms/step - loss: 1.2200 - acc: 0.6921 - val_loss: 1.3777 - val_acc: 0.6581\n",
      "Epoch 11/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.1707 - acc: 0.7031\n",
      "Epoch 00011: val_acc improved from 0.65806 to 0.66162, saving model to best_model.h5\n",
      "71/71 [==============================] - 12s 173ms/step - loss: 1.1707 - acc: 0.7031 - val_loss: 1.3295 - val_acc: 0.6616\n",
      "Epoch 12/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.1188 - acc: 0.7141\n",
      "Epoch 00012: val_acc improved from 0.66162 to 0.67097, saving model to best_model.h5\n",
      "71/71 [==============================] - 19s 268ms/step - loss: 1.1188 - acc: 0.7141 - val_loss: 1.2802 - val_acc: 0.6710\n",
      "Epoch 13/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.0826 - acc: 0.7227\n",
      "Epoch 00013: val_acc improved from 0.67097 to 0.68388, saving model to best_model.h5\n",
      "71/71 [==============================] - 18s 249ms/step - loss: 1.0826 - acc: 0.7227 - val_loss: 1.2551 - val_acc: 0.6839\n",
      "Epoch 14/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 1.0180 - acc: 0.7413\n",
      "Epoch 00014: val_acc improved from 0.68388 to 0.69457, saving model to best_model.h5\n",
      "71/71 [==============================] - 17s 236ms/step - loss: 1.0180 - acc: 0.7413 - val_loss: 1.2380 - val_acc: 0.6946\n",
      "Epoch 15/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.9920 - acc: 0.7470\n",
      "Epoch 00015: val_acc improved from 0.69457 to 0.70303, saving model to best_model.h5\n",
      "71/71 [==============================] - 15s 218ms/step - loss: 0.9920 - acc: 0.7470 - val_loss: 1.2167 - val_acc: 0.7030\n",
      "Epoch 16/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.9418 - acc: 0.7614\n",
      "Epoch 00016: val_acc did not improve from 0.70303\n",
      "71/71 [==============================] - 16s 231ms/step - loss: 0.9418 - acc: 0.7614 - val_loss: 1.2207 - val_acc: 0.6950\n",
      "Epoch 17/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.8886 - acc: 0.7720\n",
      "Epoch 00017: val_acc did not improve from 0.70303\n",
      "71/71 [==============================] - 16s 222ms/step - loss: 0.8886 - acc: 0.7720 - val_loss: 1.1922 - val_acc: 0.7008\n",
      "Epoch 18/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.8490 - acc: 0.7819\n",
      "Epoch 00018: val_acc improved from 0.70303 to 0.70793, saving model to best_model.h5\n",
      "71/71 [==============================] - 16s 226ms/step - loss: 0.8490 - acc: 0.7819 - val_loss: 1.1966 - val_acc: 0.7079\n",
      "Epoch 19/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.8103 - acc: 0.7934\n",
      "Epoch 00019: val_acc improved from 0.70793 to 0.71327, saving model to best_model.h5\n",
      "71/71 [==============================] - 16s 226ms/step - loss: 0.8103 - acc: 0.7934 - val_loss: 1.2016 - val_acc: 0.7133\n",
      "Epoch 20/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.7702 - acc: 0.8055\n",
      "Epoch 00020: val_acc did not improve from 0.71327\n",
      "71/71 [==============================] - 16s 229ms/step - loss: 0.7702 - acc: 0.8055 - val_loss: 1.2377 - val_acc: 0.7075\n",
      "Epoch 21/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.7525 - acc: 0.8073\n",
      "Epoch 00021: val_acc improved from 0.71327 to 0.71460, saving model to best_model.h5\n",
      "71/71 [==============================] - 17s 238ms/step - loss: 0.7525 - acc: 0.8073 - val_loss: 1.1885 - val_acc: 0.7146\n",
      "Epoch 22/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.7122 - acc: 0.8223\n",
      "Epoch 00022: val_acc improved from 0.71460 to 0.72885, saving model to best_model.h5\n",
      "71/71 [==============================] - 18s 252ms/step - loss: 0.7122 - acc: 0.8223 - val_loss: 1.2124 - val_acc: 0.7289\n",
      "Epoch 23/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.6706 - acc: 0.8310\n",
      "Epoch 00023: val_acc improved from 0.72885 to 0.72974, saving model to best_model.h5\n",
      "71/71 [==============================] - 20s 278ms/step - loss: 0.6706 - acc: 0.8310 - val_loss: 1.1866 - val_acc: 0.7297\n",
      "Epoch 24/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.6435 - acc: 0.8376\n",
      "Epoch 00024: val_acc improved from 0.72974 to 0.73108, saving model to best_model.h5\n",
      "71/71 [==============================] - 17s 246ms/step - loss: 0.6435 - acc: 0.8376 - val_loss: 1.2118 - val_acc: 0.7311\n",
      "Epoch 25/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.6256 - acc: 0.8424- ETA: 2s - loss: 0.6237 - a\n",
      "Epoch 00025: val_acc did not improve from 0.73108\n",
      "71/71 [==============================] - 16s 221ms/step - loss: 0.6256 - acc: 0.8424 - val_loss: 1.2430 - val_acc: 0.7195\n",
      "Epoch 26/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.6025 - acc: 0.8514\n",
      "Epoch 00026: val_acc did not improve from 0.73108\n",
      "71/71 [==============================] - 15s 218ms/step - loss: 0.6025 - acc: 0.8514 - val_loss: 1.2196 - val_acc: 0.7231\n",
      "Epoch 27/30\n",
      "71/71 [==============================] - ETA: 0s - loss: 0.5636 - acc: 0.8582\n",
      "Epoch 00027: val_acc did not improve from 0.73108\n",
      "71/71 [==============================] - 17s 237ms/step - loss: 0.5636 - acc: 0.8582 - val_loss: 1.2589 - val_acc: 0.7204\n",
      "Epoch 00027: early stopping\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(X_train, y_train, batch_size=128, epochs=30, callbacks=[es, mc], validation_data=(X_test, y_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "71/71 [==============================] - 1s 16ms/step - loss: 1.2118 - acc: 0.7311\n",
      "\n",
      " Test Accuracy: 0.7311\n"
     ]
    }
   ],
   "source": [
    "loaded_model = load_model('best_model.h5')\n",
    "\n",
    "# Check the Testset Accuracy\n",
    "print(\"\\n Test Accuracy: %.4f\" % (loaded_model.evaluate(X_test, y_test)[1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We get the 73% accuracy on test set with the very basic model that has one Embedding and one LSTM layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x22ed0ef1d48>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "# Check the loss graph\n",
    "epochs = range(1, len(history.history['acc']) + 1)\n",
    "plt.plot(epochs, history.history['loss'])\n",
    "plt.plot(epochs, history.history['val_loss'])\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Overfitting seems to begin at aronud 5th epoch, which we can improve by adding normalization layers and revising hyperparameters"
   ]
  }
 ],
 "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}