digit recognizer kaggle.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### get data from https://www.kaggle.com/c/digit-recognizer/data?select=train.csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 335,
   "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>label</th>\n",
       "      <th>pixel0</th>\n",
       "      <th>pixel1</th>\n",
       "      <th>pixel2</th>\n",
       "      <th>pixel3</th>\n",
       "      <th>pixel4</th>\n",
       "      <th>pixel5</th>\n",
       "      <th>pixel6</th>\n",
       "      <th>pixel7</th>\n",
       "      <th>pixel8</th>\n",
       "      <th>...</th>\n",
       "      <th>pixel774</th>\n",
       "      <th>pixel775</th>\n",
       "      <th>pixel776</th>\n",
       "      <th>pixel777</th>\n",
       "      <th>pixel778</th>\n",
       "      <th>pixel779</th>\n",
       "      <th>pixel780</th>\n",
       "      <th>pixel781</th>\n",
       "      <th>pixel782</th>\n",
       "      <th>pixel783</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 785 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   label  pixel0  pixel1  pixel2  pixel3  pixel4  pixel5  pixel6  pixel7  \\\n",
       "0      1       0       0       0       0       0       0       0       0   \n",
       "1      0       0       0       0       0       0       0       0       0   \n",
       "2      1       0       0       0       0       0       0       0       0   \n",
       "3      4       0       0       0       0       0       0       0       0   \n",
       "4      0       0       0       0       0       0       0       0       0   \n",
       "\n",
       "   pixel8  ...  pixel774  pixel775  pixel776  pixel777  pixel778  pixel779  \\\n",
       "0       0  ...         0         0         0         0         0         0   \n",
       "1       0  ...         0         0         0         0         0         0   \n",
       "2       0  ...         0         0         0         0         0         0   \n",
       "3       0  ...         0         0         0         0         0         0   \n",
       "4       0  ...         0         0         0         0         0         0   \n",
       "\n",
       "   pixel780  pixel781  pixel782  pixel783  \n",
       "0         0         0         0         0  \n",
       "1         0         0         0         0  \n",
       "2         0         0         0         0  \n",
       "3         0         0         0         0  \n",
       "4         0         0         0         0  \n",
       "\n",
       "[5 rows x 785 columns]"
      ]
     },
     "execution_count": 335,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "train = pd.read_csv('digit-recognizer/train.csv')\n",
    "train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 336,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(42000, 785)"
      ]
     },
     "execution_count": 336,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 337,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "28.0"
      ]
     },
     "execution_count": 337,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "784**0.5 # 28 x 28 image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 338,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 339,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "label\n",
       "0    4132\n",
       "1    4684\n",
       "2    4177\n",
       "3    4351\n",
       "4    4072\n",
       "5    3795\n",
       "6    4137\n",
       "7    4401\n",
       "8    4063\n",
       "9    4188\n",
       "Name: label, dtype: int64"
      ]
     },
     "execution_count": 339,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.groupby(['label'])['label'].count()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### randomly display 9 samples along with class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 340,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "index = np.random.randint(0, train.shape[0], size=9)\n",
    "fig, axis = plt.subplots(3, 3, figsize=(6,6))\n",
    "for i in range(3):\n",
    "    for j in range(3):\n",
    "        idx = index[3*i + j]\n",
    "        label = train.label[idx]\n",
    "        data = train.iloc[idx,1:]\n",
    "        img = np.array(data).reshape((28,28))\n",
    "        axis[i][j].set_xticks([])\n",
    "        axis[i][j].set_yticks([])\n",
    "        axis[i][j].set_title(f'class = {label}')\n",
    "        axis[i][j].imshow(img)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "training and test splits"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 341,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2.380952380952381"
      ]
     },
     "execution_count": 341,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# approx 4000 images from each classes\n",
    "(1000/train.shape[0]) * 100 # 11 % data for testing >> 100 samples from each classes for testing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 342,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([1, 0, 1, ..., 7, 6, 9], dtype=int64)"
      ]
     },
     "execution_count": 342,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels = np.array(train.label)\n",
    "labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 343,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.random.seed(110)\n",
    "test_samples_from_each_class = 100\n",
    "train_idxs=[]\n",
    "test_idxs=[]\n",
    "for digit in range(10):\n",
    "    arr = np.where(labels == digit)[0]\n",
    "    np.random.shuffle(arr)\n",
    "    test_idxs.extend(list(arr)[:test_samples_from_each_class])\n",
    "    train_idxs.extend(list(arr)[test_samples_from_each_class:])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 344,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 344,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(set(train_idxs + test_idxs)) == train.shape[0] # sanity checks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 345,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1000"
      ]
     },
     "execution_count": 345,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(test_idxs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 346,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((41000, 785), (1000, 785))"
      ]
     },
     "execution_count": 346,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data = train.iloc[train_idxs,:]\n",
    "test_data = train.iloc[test_idxs,:]\n",
    "train_data.shape, test_data.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 347,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp = train_x, train_y, test_x, test_y = train_data.iloc[:,1:], train_data.iloc[:,0], test_data.iloc[:,1:], test_data.iloc[:,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 348,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(41000, 784), (41000,), (1000, 784), (1000,)]"
      ]
     },
     "execution_count": 348,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "[i.shape for i in temp]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 349,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_x, test_x = train_x.values, test_x.values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 350,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([[0, 0, 0, ..., 0, 0, 0],\n",
       "        [0, 0, 0, ..., 0, 0, 0],\n",
       "        [0, 0, 0, ..., 0, 0, 0],\n",
       "        ...,\n",
       "        [0, 0, 0, ..., 0, 0, 0],\n",
       "        [0, 0, 0, ..., 0, 0, 0],\n",
       "        [0, 0, 0, ..., 0, 0, 0]], dtype=int64),\n",
       " array([[0, 0, 0, ..., 0, 0, 0],\n",
       "        [0, 0, 0, ..., 0, 0, 0],\n",
       "        [0, 0, 0, ..., 0, 0, 0],\n",
       "        ...,\n",
       "        [0, 0, 0, ..., 0, 0, 0],\n",
       "        [0, 0, 0, ..., 0, 0, 0],\n",
       "        [0, 0, 0, ..., 0, 0, 0]], dtype=int64),\n",
       " (41000, 784),\n",
       " (1000, 784))"
      ]
     },
     "execution_count": 350,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_x, test_x, train_x.shape, test_x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 351,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([0, 0, 0, ..., 9, 9, 9], dtype=int64),\n",
       " array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
       "        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
       "        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
       "        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
       "        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
       "        1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
       "        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
       "        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
       "        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
       "        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3,\n",
       "        3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,\n",
       "        3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,\n",
       "        3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,\n",
       "        3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,\n",
       "        3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,\n",
       "        4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,\n",
       "        4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,\n",
       "        4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,\n",
       "        4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5,\n",
       "        5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,\n",
       "        5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,\n",
       "        5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,\n",
       "        5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,\n",
       "        5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,\n",
       "        6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,\n",
       "        6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,\n",
       "        6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,\n",
       "        6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7,\n",
       "        7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,\n",
       "        7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,\n",
       "        7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,\n",
       "        7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,\n",
       "        7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,\n",
       "        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,\n",
       "        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,\n",
       "        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,\n",
       "        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9,\n",
       "        9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,\n",
       "        9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,\n",
       "        9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,\n",
       "        9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,\n",
       "        9, 9, 9, 9, 9, 9, 9, 9, 9, 9], dtype=int64),\n",
       " (41000,),\n",
       " (1000,))"
      ]
     },
     "execution_count": 351,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_y, test_y = train_y.values, test_y.values\n",
    "train_y, test_y, train_y.shape, test_y.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 352,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 353,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_y = tf.keras.utils.to_categorical(train_y)\n",
    "test_y = tf.keras.utils.to_categorical(test_y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 354,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([[1., 0., 0., ..., 0., 0., 0.],\n",
       "        [1., 0., 0., ..., 0., 0., 0.],\n",
       "        [1., 0., 0., ..., 0., 0., 0.],\n",
       "        ...,\n",
       "        [0., 0., 0., ..., 0., 0., 1.],\n",
       "        [0., 0., 0., ..., 0., 0., 1.],\n",
       "        [0., 0., 0., ..., 0., 0., 1.]], dtype=float32),\n",
       " array([[1., 0., 0., ..., 0., 0., 0.],\n",
       "        [1., 0., 0., ..., 0., 0., 0.],\n",
       "        [1., 0., 0., ..., 0., 0., 0.],\n",
       "        ...,\n",
       "        [0., 0., 0., ..., 0., 0., 1.],\n",
       "        [0., 0., 0., ..., 0., 0., 1.],\n",
       "        [0., 0., 0., ..., 0., 0., 1.]], dtype=float32),\n",
       " (41000, 10),\n",
       " (1000, 10))"
      ]
     },
     "execution_count": 354,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_y, test_y, train_y.shape, test_y.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### data preparation complete\n",
    "### ('_') Hurray"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### data normalization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 355,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([[0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        ...,\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.]]),\n",
       " array([[0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        ...,\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.]]),\n",
       " (41000, 784),\n",
       " (1000, 784))"
      ]
     },
     "execution_count": 355,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_x, test_x = train_x/255, test_x/255\n",
    "train_x, test_x, train_x.shape, test_x.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### create neural network architecture"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 369,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow import keras\n",
    "from tensorflow.keras.layers import Dense\n",
    "from tensorflow.keras import Sequential\n",
    "from tensorflow.keras.losses import CategoricalCrossentropy\n",
    "from tensorflow.keras.optimizers import Adam, RMSprop\n",
    "from tensorflow.keras.regularizers import l2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 378,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential([\n",
    "    Dense(16, activation='relu', input_shape=(784,), kernel_regularizer=l2(0.001)),\n",
    "    Dense(16, activation='relu', kernel_regularizer=l2(0.001)),\n",
    "    Dense(10)\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 379,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(optimizer=Adam(learning_rate=0.0015),\n",
    "              loss=CategoricalCrossentropy(from_logits=True),\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 380,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "1282/1282 [==============================] - 2s 2ms/step - loss: 0.5414 - accuracy: 0.8588 - val_loss: 0.3979 - val_accuracy: 0.9040\n",
      "Epoch 2/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.3289 - accuracy: 0.9252 - val_loss: 0.3723 - val_accuracy: 0.9190\n",
      "Epoch 3/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.3011 - accuracy: 0.9344 - val_loss: 0.3394 - val_accuracy: 0.9240\n",
      "Epoch 4/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2817 - accuracy: 0.9397 - val_loss: 0.3314 - val_accuracy: 0.9280\n",
      "Epoch 5/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2674 - accuracy: 0.9421 - val_loss: 0.3182 - val_accuracy: 0.9400\n",
      "Epoch 6/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2580 - accuracy: 0.9451 - val_loss: 0.3280 - val_accuracy: 0.9330\n",
      "Epoch 7/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2524 - accuracy: 0.9470 - val_loss: 0.3184 - val_accuracy: 0.9350\n",
      "Epoch 8/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2460 - accuracy: 0.9488 - val_loss: 0.3066 - val_accuracy: 0.9300\n",
      "Epoch 9/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2403 - accuracy: 0.9501 - val_loss: 0.2899 - val_accuracy: 0.9380\n",
      "Epoch 10/20\n",
      "1282/1282 [==============================] - 2s 1ms/step - loss: 0.2328 - accuracy: 0.9531 - val_loss: 0.2993 - val_accuracy: 0.9350\n",
      "Epoch 11/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2290 - accuracy: 0.9524 - val_loss: 0.3109 - val_accuracy: 0.9330\n",
      "Epoch 12/20\n",
      "1282/1282 [==============================] - 2s 1ms/step - loss: 0.2258 - accuracy: 0.9531 - val_loss: 0.2794 - val_accuracy: 0.9420\n",
      "Epoch 13/20\n",
      "1282/1282 [==============================] - 2s 1ms/step - loss: 0.2214 - accuracy: 0.9545 - val_loss: 0.2701 - val_accuracy: 0.9410\n",
      "Epoch 14/20\n",
      "1282/1282 [==============================] - 2s 1ms/step - loss: 0.2150 - accuracy: 0.9564 - val_loss: 0.2769 - val_accuracy: 0.9440\n",
      "Epoch 15/20\n",
      "1282/1282 [==============================] - 2s 1ms/step - loss: 0.2135 - accuracy: 0.9567 - val_loss: 0.2694 - val_accuracy: 0.9490\n",
      "Epoch 16/20\n",
      "1282/1282 [==============================] - 2s 1ms/step - loss: 0.2100 - accuracy: 0.9570 - val_loss: 0.2724 - val_accuracy: 0.9390\n",
      "Epoch 17/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2108 - accuracy: 0.9575 - val_loss: 0.2869 - val_accuracy: 0.9380\n",
      "Epoch 18/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2080 - accuracy: 0.9579 - val_loss: 0.2779 - val_accuracy: 0.9400\n",
      "Epoch 19/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2053 - accuracy: 0.9579 - val_loss: 0.2641 - val_accuracy: 0.9410\n",
      "Epoch 20/20\n",
      "1282/1282 [==============================] - 1s 1ms/step - loss: 0.2063 - accuracy: 0.9576 - val_loss: 0.2663 - val_accuracy: 0.9450\n"
     ]
    }
   ],
   "source": [
    "hstory = model.fit(train_x, train_y, batch_size=None, epochs=20, validation_data=(test_x, test_y))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 239,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy'])"
      ]
     },
     "execution_count": 239,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "hstory.history.keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 381,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x26119910100>"
      ]
     },
     "execution_count": 381,
     "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": [
    "train_acc = hstory.history['accuracy']\n",
    "test_acc = hstory.history['val_accuracy']\n",
    "plt.plot(train_acc, label='training accuracy')\n",
    "plt.plot(test_acc, label='test accuracy')\n",
    "plt.legend(bbox_to_anchor=(1.05, 1), loc='upper left')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### plot failed classification (failures from validation set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 382,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "Y_hat = model.predict(test_x)\n",
    "Y_hat = np.argmax(Y_hat, axis=1)\n",
    "Y = np.argmax(test_y, axis=1)\n",
    "failed = np.where(Y_hat != Y)[0]\n",
    "np.random.shuffle(failed)\n",
    "\n",
    "fig, axis = plt.subplots(3, 3, figsize=(6,6))\n",
    "for k, i in enumerate(failed[:9]):\n",
    "    actual = Y_hat[i]\n",
    "    expected = Y[i]\n",
    "    data = test_x[i,:]\n",
    "    img = np.array(data).reshape((28,28))\n",
    "    j = k % 3\n",
    "    i = k // 3\n",
    "    axis[i][j].set_xticks([])\n",
    "    axis[i][j].set_yticks([])\n",
    "    axis[i][j].set_title(f'Expected {expected} but {actual}')\n",
    "    axis[i][j].imshow(img)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### plot failed classification (failures from train set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 383,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "Y_hat = model.predict(train_x)\n",
    "Y_hat = np.argmax(Y_hat, axis=1)\n",
    "Y = np.argmax(train_y, axis=1)\n",
    "failed = np.where(Y_hat != Y)[0]\n",
    "np.random.shuffle(failed)\n",
    "\n",
    "fig, axis = plt.subplots(3, 3, figsize=(6,6))\n",
    "for k, i in enumerate(failed[:9]):\n",
    "    actual = Y_hat[i]\n",
    "    expected = Y[i]\n",
    "    data = train_x[i,:]\n",
    "    img = np.array(data).reshape((28,28))\n",
    "    j = k % 3\n",
    "    i = k // 3\n",
    "    axis[i][j].set_xticks([])\n",
    "    axis[i][j].set_yticks([])\n",
    "    axis[i][j].set_title(f'Expected {expected} but {actual}')\n",
    "    axis[i][j].imshow(img)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### do the prediction and submit to kaggle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 384,
   "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>pixel0</th>\n",
       "      <th>pixel1</th>\n",
       "      <th>pixel2</th>\n",
       "      <th>pixel3</th>\n",
       "      <th>pixel4</th>\n",
       "      <th>pixel5</th>\n",
       "      <th>pixel6</th>\n",
       "      <th>pixel7</th>\n",
       "      <th>pixel8</th>\n",
       "      <th>pixel9</th>\n",
       "      <th>...</th>\n",
       "      <th>pixel774</th>\n",
       "      <th>pixel775</th>\n",
       "      <th>pixel776</th>\n",
       "      <th>pixel777</th>\n",
       "      <th>pixel778</th>\n",
       "      <th>pixel779</th>\n",
       "      <th>pixel780</th>\n",
       "      <th>pixel781</th>\n",
       "      <th>pixel782</th>\n",
       "      <th>pixel783</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 784 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   pixel0  pixel1  pixel2  pixel3  pixel4  pixel5  pixel6  pixel7  pixel8  \\\n",
       "0       0       0       0       0       0       0       0       0       0   \n",
       "1       0       0       0       0       0       0       0       0       0   \n",
       "2       0       0       0       0       0       0       0       0       0   \n",
       "3       0       0       0       0       0       0       0       0       0   \n",
       "4       0       0       0       0       0       0       0       0       0   \n",
       "\n",
       "   pixel9  ...  pixel774  pixel775  pixel776  pixel777  pixel778  pixel779  \\\n",
       "0       0  ...         0         0         0         0         0         0   \n",
       "1       0  ...         0         0         0         0         0         0   \n",
       "2       0  ...         0         0         0         0         0         0   \n",
       "3       0  ...         0         0         0         0         0         0   \n",
       "4       0  ...         0         0         0         0         0         0   \n",
       "\n",
       "   pixel780  pixel781  pixel782  pixel783  \n",
       "0         0         0         0         0  \n",
       "1         0         0         0         0  \n",
       "2         0         0         0         0  \n",
       "3         0         0         0         0  \n",
       "4         0         0         0         0  \n",
       "\n",
       "[5 rows x 784 columns]"
      ]
     },
     "execution_count": 384,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test = pd.read_csv('digit-recognizer/test.csv')\n",
    "test.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 385,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_x = test.values/255"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 386,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([[0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        ...,\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.],\n",
       "        [0., 0., 0., ..., 0., 0., 0.]]),\n",
       " (28000, 784))"
      ]
     },
     "execution_count": 386,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "input_x, input_x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 387,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ -1.5165982 ,  -8.509129  ,   9.149618  , ...,  -3.5331945 ,\n",
       "         -6.2718625 ,  -8.347802  ],\n",
       "       [  8.521107  , -10.739945  ,  -0.8408563 , ...,  -2.3259387 ,\n",
       "         -6.2840095 ,  -5.8270555 ],\n",
       "       [ -6.4342127 ,  -4.0320992 ,  -4.206473  , ...,  -3.0456643 ,\n",
       "         -0.76932687,   2.036527  ],\n",
       "       ...,\n",
       "       [-13.477259  ,  -6.103618  ,  -3.3563764 , ...,  -5.1315684 ,\n",
       "          0.34396422,  -2.3755665 ],\n",
       "       [ -9.098355  , -12.51534   , -10.028203  , ...,  -5.900145  ,\n",
       "         -6.7969112 ,   4.3441534 ],\n",
       "       [ -1.0857413 ,  -9.10067   ,   8.57851   , ...,  -5.465493  ,\n",
       "         -8.378959  ,  -7.4467764 ]], dtype=float32)"
      ]
     },
     "execution_count": 387,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions = model.predict(input_x)\n",
    "predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 388,
   "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>ImageId</th>\n",
       "      <th>Label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   ImageId  Label\n",
       "0        1      2\n",
       "1        2      0\n",
       "2        3      9\n",
       "3        4      9\n",
       "4        5      3"
      ]
     },
     "execution_count": 388,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ImageId = list(range(1, predictions.shape[0]+1))\n",
    "Label = list(np.argmax(predictions, axis=1))\n",
    "df_out = pd.DataFrame({'ImageId':ImageId, 'Label':Label})\n",
    "df_out.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 389,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_out.to_csv('digit-recognizer/submissions/submit1.csv', index=False)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "DLVenv",
   "language": "python",
   "name": "dlvenv"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}