DataPrep.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data prep for all training data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "What follows is a notebook I used to try and recreate the features Jeremy Howard teaches [here](https://github.com/fastai/courses/blob/master/deeplearning2/rossman.ipynb) on the [Corporación Favorita Grocery Sales Forecasting](https://www.kaggle.com/c/favorita-grocery-sales-forecasting/discussion/47185) Kaggle contest.\n",
    "\n",
    "I would highly discourage you from trying to copy and run this code. Although I did my best to systematically document and test it at every step, I'm sure I've made occasional errors. \n",
    "\n",
    "I learned a ton going through it - hopefully you will as well."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "env: CUDA_DEVICE_ORDER=PCI_BUS_ID\n",
      "env: CUDA_VISIBLE_DEVICES=3\n",
      "kevin loaded\n"
     ]
    }
   ],
   "source": [
    "gpu=3\n",
    "%env CUDA_DEVICE_ORDER=PCI_BUS_ID\n",
    "%env CUDA_VISIBLE_DEVICES=$gpu\n",
    "# libraries and utilities\n",
    "%matplotlib inline\n",
    "import importlib\n",
    "import kevin; importlib.reload(kevin)\n",
    "from kevin import *\n",
    "import gfunc; importlib.reload(gfunc)\n",
    "from gfunc import *\n",
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### memory saving notes - \n",
    "1. switch objects to categories earlier.\n",
    "2. save and delete older train objects when joined is created.\n",
    "3. clean up rolling promo function"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## paths"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#Store in new place since all data\n",
    "slowdata = data_path + 'slowdisk/'\n",
    "slowdata = slowdata + 'alldata/'\n",
    "results_path = results_path+'alldata/'\n",
    "\n",
    "def save_df(df, name):\n",
    "    if df.to_pickle(slowdata+name):\n",
    "        return True\n",
    "def load_df(name):\n",
    "    return pd.read_pickle(slowdata+name)\n",
    "def save_n_load_df(df, name):\n",
    "    if save_df(df, name): load_df(name)\n",
    "    return True\n",
    "#leaves duplicate field names on left and appends with _y on right\n",
    "def join_df(left, right, left_on, right_on=None):\n",
    "    if right_on is None: right_on=left_on\n",
    "    return left.merge(right, how='left', left_on=left_on, right_on=right_on, suffixes=(\"\", \"_y\"))\n",
    "def add_datepart(df):\n",
    "#     df.date = pd.to_datetime(df.date)\n",
    "    df[\"year\"] = df.date.dt.year.astype(np.int32)\n",
    "    df[\"month\"] = df.date.dt.month.astype(np.int8)\n",
    "    df[\"week\"] = df.date.dt.week.astype(np.int8)\n",
    "    df[\"day\"] = df.date.dt.day.astype(np.int8)\n",
    "    df[\"dayofweek\"] = df.date.dt.dayofweek.astype(np.int8)\n",
    "def fillna_with_zeros(df, column):\n",
    "    df[column].fillna(0, inplace=True)\n",
    "    df[column]=df[column].astype(np.int8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/home/kevin/nbs/kaggle/GrocerySales/results/alldata/'"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results_path"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Work plan\n",
    "### Data prep\n",
    "1. Load data, 20% of training. chunksize = 25_000_000\n",
    "2. Merge test/train datasets\n",
    "1. Remove null data\n",
    "3. Add columns for item, store, oil price, holiday\n",
    "1. Convert dates to year mon day weekday\n",
    "4. Convert labels to numbers. Change datatypes?\n",
    "### Set benchmark\n",
    "1. Create mean."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#From:\n",
    "# https://www.kaggle.com/ekami66/python-eda-with-few-interactive-graphs\n",
    "# https://www.kaggle.com/jagangupta/memory-optimization-and-eda-on-entire-dataset/notebook\n",
    "# https://www.kaggle.com/juanumusic/tidy-notebook-and-deep-analysis\n",
    "dtypes = {'id': np.uint32,\n",
    "          'store_nbr': np.uint8, \n",
    "          'item_nbr': np.uint32, \n",
    "          'unit_sales': np.float32,\n",
    "          'class': np.uint16,\n",
    "          'perishable': np.bool,\n",
    "          'dcoilwtico':np.float16,\n",
    "          'onpromotion': np.bool,\n",
    "          'transactions':np.uint16,\n",
    "          'cluster': np.uint32}\n",
    "chunksize = 5_000_000 #5% of training set\n",
    "#mid sample for data prep & checking\n",
    "mid = int(chunksize/2); sample = slice(mid, mid+5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/kevin/anaconda2/envs/ai/lib/python3.6/site-packages/IPython/core/interactiveshell.py:2718: DtypeWarning: Columns (5) have mixed types. Specify dtype option on import or set low_memory=False.\n",
      "  interactivity=interactivity, compiler=compiler, result=result)\n",
      "/home/kevin/anaconda2/envs/ai/lib/python3.6/site-packages/numpy/lib/arraysetops.py:463: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison\n",
      "  mask |= (ar1 == a)\n"
     ]
    }
   ],
   "source": [
    "#train_df\n",
    "train_df = pd.read_csv(slowdata + 'train.csv', index_col='id', parse_dates=['date'], dtype=dtypes)\n",
    "# all_train_df = pd.read_csv(slowdata + 'train.csv', index_col='id', parse_dates=['date'], dtype=dtypes)\n",
    "# train_df=all_train_df[-chunksize:]\n",
    "save_n_load_df(train_df, \"training_dev.pkl\")\n",
    "train_df.name = \"TRAINING\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/kevin/anaconda2/envs/ai/lib/python3.6/site-packages/numpy/lib/arraysetops.py:463: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison\n",
      "  mask |= (ar1 == a)\n"
     ]
    }
   ],
   "source": [
    "transactions_df = pd.read_csv(slowdata + 'transactions.csv', parse_dates=['date'], dtype=dtypes)\n",
    "transactions_df.name = \"TRANSACTIONS\"\n",
    "sample_sub_df = pd.read_csv(slowdata+'sample_submission.csv')\n",
    "sample_sub_df.name = \"SAMPLE\"\n",
    "test_df = pd.read_csv(slowdata+'test.csv', dtype=dtypes, index_col='id', parse_dates=['date'])\n",
    "test_df.name = \"TEST\"\n",
    "oil_df = pd.read_csv(slowdata + 'oil.csv', parse_dates=['date'], dtype=dtypes)\n",
    "oil_df.name = \"OIL\"\n",
    "holidays_df = pd.read_csv(slowdata + 'holidays_events.csv', parse_dates=['date'], dtype=dtypes)\n",
    "holidays_df.name = \"HOLIDAYS\"\n",
    "stores_df = pd.read_csv(slowdata + 'stores.csv', dtype=dtypes)\n",
    "stores_df.name = \"STORES\"\n",
    "items_df = pd.read_csv(slowdata + 'items.csv',dtype=dtypes)\n",
    "items_df.name = \"ITEMS\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 375,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# oil_df = pd.read_csv(slowdata + 'oil.csv', parse_dates=['date'], dtype=dtypes)\n",
    "# oil_df.name = \"OIL\"\n",
    "# stores_df = pd.read_csv(slowdata + 'stores.csv', dtype=dtypes)\n",
    "# stores_df.name = \"STORES\"\n",
    "# items_df = pd.read_csv(slowdata + 'items.csv',dtype=dtypes)\n",
    "# items_df.name = \"ITEMS\"\n",
    "# transactions_df = pd.read_csv(slowdata + 'transactions.csv', parse_dates=['date'], dtype=dtypes)\n",
    "# transactions_df.name = \"TRANSACTIONS\"\n",
    "# holidays_df = pd.read_csv(slowdata + 'holidays_events.csv', parse_dates=['date'], dtype=dtypes)\n",
    "# holidays_df.name = \"HOLIDAYS\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## View data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "tables = [train_df, transactions_df, sample_sub_df, test_df, oil_df, holidays_df, stores_df, items_df]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'TRAINING'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>store_nbr</th>\n",
       "      <th>item_nbr</th>\n",
       "      <th>unit_sales</th>\n",
       "      <th>onpromotion</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>120497040</th>\n",
       "      <td>2017-06-29</td>\n",
       "      <td>35</td>\n",
       "      <td>1162935</td>\n",
       "      <td>5.0</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>120497041</th>\n",
       "      <td>2017-06-29</td>\n",
       "      <td>35</td>\n",
       "      <td>1164290</td>\n",
       "      <td>1.0</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>120497042</th>\n",
       "      <td>2017-06-29</td>\n",
       "      <td>35</td>\n",
       "      <td>1165988</td>\n",
       "      <td>2.0</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>120497043</th>\n",
       "      <td>2017-06-29</td>\n",
       "      <td>35</td>\n",
       "      <td>1166474</td>\n",
       "      <td>50.0</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>120497044</th>\n",
       "      <td>2017-06-29</td>\n",
       "      <td>35</td>\n",
       "      <td>1166475</td>\n",
       "      <td>11.0</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                date  store_nbr  item_nbr  unit_sales onpromotion\n",
       "id                                                               \n",
       "120497040 2017-06-29         35   1162935         5.0       False\n",
       "120497041 2017-06-29         35   1164290         1.0       False\n",
       "120497042 2017-06-29         35   1165988         2.0       False\n",
       "120497043 2017-06-29         35   1166474        50.0       False\n",
       "120497044 2017-06-29         35   1166475        11.0       False"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>store_nbr</th>\n",
       "      <th>item_nbr</th>\n",
       "      <th>unit_sales</th>\n",
       "      <th>onpromotion</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>NaN</td>\n",
       "      <td>5e+06</td>\n",
       "      <td>5e+06</td>\n",
       "      <td>5e+06</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>NaN</td>\n",
       "      <td>28.4204</td>\n",
       "      <td>1.17055e+06</td>\n",
       "      <td>8.10904</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>NaN</td>\n",
       "      <td>16.3229</td>\n",
       "      <td>586921</td>\n",
       "      <td>23.7943</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>96995</td>\n",
       "      <td>-274</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>14</td>\n",
       "      <td>692105</td>\n",
       "      <td>2</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>29</td>\n",
       "      <td>1.20972e+06</td>\n",
       "      <td>4</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>44</td>\n",
       "      <td>1.57633e+06</td>\n",
       "      <td>8</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>NaN</td>\n",
       "      <td>54</td>\n",
       "      <td>2.12711e+06</td>\n",
       "      <td>7348</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>counts</th>\n",
       "      <td>5000000</td>\n",
       "      <td>5000000</td>\n",
       "      <td>5000000</td>\n",
       "      <td>5000000</td>\n",
       "      <td>5000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniques</th>\n",
       "      <td>48</td>\n",
       "      <td>54</td>\n",
       "      <td>3891</td>\n",
       "      <td>61385</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing</th>\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>missing_perc</th>\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>types</th>\n",
       "      <td>date</td>\n",
       "      <td>numeric</td>\n",
       "      <td>numeric</td>\n",
       "      <td>numeric</td>\n",
       "      <td>bool</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                 date store_nbr     item_nbr unit_sales onpromotion\n",
       "count             NaN     5e+06        5e+06      5e+06         NaN\n",
       "mean              NaN   28.4204  1.17055e+06    8.10904         NaN\n",
       "std               NaN   16.3229       586921    23.7943         NaN\n",
       "min               NaN         1        96995       -274         NaN\n",
       "25%               NaN        14       692105          2         NaN\n",
       "50%               NaN        29  1.20972e+06          4         NaN\n",
       "75%               NaN        44  1.57633e+06          8         NaN\n",
       "max               NaN        54  2.12711e+06       7348         NaN\n",
       "counts        5000000   5000000      5000000    5000000     5000000\n",
       "uniques            48        54         3891      61385           2\n",
       "missing             0         0            0          0           0\n",
       "missing_perc       0%        0%           0%         0%          0%\n",
       "types            date   numeric      numeric    numeric        bool"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'TRANSACTIONS'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>store_nbr</th>\n",
       "      <th>transactions</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2013-01-01</td>\n",
       "      <td>25</td>\n",
       "      <td>770</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2013-01-02</td>\n",
       "      <td>1</td>\n",
       "      <td>2111</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2013-01-02</td>\n",
       "      <td>2</td>\n",
       "      <td>2358</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2013-01-02</td>\n",
       "      <td>3</td>\n",
       "      <td>3487</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2013-01-02</td>\n",
       "      <td>4</td>\n",
       "      <td>1922</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date  store_nbr  transactions\n",
       "0  2013-01-01         25           770\n",
       "1  2013-01-02          1          2111\n",
       "2  2013-01-02          2          2358\n",
       "3  2013-01-02          3          3487\n",
       "4  2013-01-02          4          1922"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>store_nbr</th>\n",
       "      <th>transactions</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>NaN</td>\n",
       "      <td>83488</td>\n",
       "      <td>83488</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>NaN</td>\n",
       "      <td>26.9392</td>\n",
       "      <td>1694.6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>NaN</td>\n",
       "      <td>15.6082</td>\n",
       "      <td>963.287</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>13</td>\n",
       "      <td>1046</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>27</td>\n",
       "      <td>1393</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>40</td>\n",
       "      <td>2079</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>NaN</td>\n",
       "      <td>54</td>\n",
       "      <td>8359</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>counts</th>\n",
       "      <td>83488</td>\n",
       "      <td>83488</td>\n",
       "      <td>83488</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniques</th>\n",
       "      <td>1682</td>\n",
       "      <td>54</td>\n",
       "      <td>4993</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_perc</th>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>types</th>\n",
       "      <td>categorical</td>\n",
       "      <td>numeric</td>\n",
       "      <td>numeric</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     date store_nbr transactions\n",
       "count                 NaN     83488        83488\n",
       "mean                  NaN   26.9392       1694.6\n",
       "std                   NaN   15.6082      963.287\n",
       "min                   NaN         1            5\n",
       "25%                   NaN        13         1046\n",
       "50%                   NaN        27         1393\n",
       "75%                   NaN        40         2079\n",
       "max                   NaN        54         8359\n",
       "counts              83488     83488        83488\n",
       "uniques              1682        54         4993\n",
       "missing                 0         0            0\n",
       "missing_perc           0%        0%           0%\n",
       "types         categorical   numeric      numeric"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'SAMPLE'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>unit_sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>125497040</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>125497041</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>125497042</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>125497043</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>125497044</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          id  unit_sales\n",
       "0  125497040           0\n",
       "1  125497041           0\n",
       "2  125497042           0\n",
       "3  125497043           0\n",
       "4  125497044           0"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>unit_sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>3.37046e+06</td>\n",
       "      <td>3.37046e+06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>1.27182e+08</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>972969</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>1.25497e+08</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>1.2634e+08</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>1.27182e+08</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>1.28025e+08</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>1.28868e+08</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>counts</th>\n",
       "      <td>3370464</td>\n",
       "      <td>3370464</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniques</th>\n",
       "      <td>3370464</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_perc</th>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>types</th>\n",
       "      <td>numeric</td>\n",
       "      <td>constant</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                       id   unit_sales\n",
       "count         3.37046e+06  3.37046e+06\n",
       "mean          1.27182e+08            0\n",
       "std                972969            0\n",
       "min           1.25497e+08            0\n",
       "25%            1.2634e+08            0\n",
       "50%           1.27182e+08            0\n",
       "75%           1.28025e+08            0\n",
       "max           1.28868e+08            0\n",
       "counts            3370464      3370464\n",
       "uniques           3370464            1\n",
       "missing                 0            0\n",
       "missing_perc           0%           0%\n",
       "types             numeric     constant"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'TEST'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>store_nbr</th>\n",
       "      <th>item_nbr</th>\n",
       "      <th>onpromotion</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>125497040</th>\n",
       "      <td>2017-08-16</td>\n",
       "      <td>1</td>\n",
       "      <td>96995</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>125497041</th>\n",
       "      <td>2017-08-16</td>\n",
       "      <td>1</td>\n",
       "      <td>99197</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>125497042</th>\n",
       "      <td>2017-08-16</td>\n",
       "      <td>1</td>\n",
       "      <td>103501</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>125497043</th>\n",
       "      <td>2017-08-16</td>\n",
       "      <td>1</td>\n",
       "      <td>103520</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>125497044</th>\n",
       "      <td>2017-08-16</td>\n",
       "      <td>1</td>\n",
       "      <td>103665</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                date  store_nbr  item_nbr  onpromotion\n",
       "id                                                    \n",
       "125497040 2017-08-16          1     96995        False\n",
       "125497041 2017-08-16          1     99197        False\n",
       "125497042 2017-08-16          1    103501        False\n",
       "125497043 2017-08-16          1    103520        False\n",
       "125497044 2017-08-16          1    103665        False"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>store_nbr</th>\n",
       "      <th>item_nbr</th>\n",
       "      <th>onpromotion</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>NaN</td>\n",
       "      <td>3.37046e+06</td>\n",
       "      <td>3.37046e+06</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>NaN</td>\n",
       "      <td>27.5</td>\n",
       "      <td>1.2448e+06</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>NaN</td>\n",
       "      <td>15.5858</td>\n",
       "      <td>589836</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>96995</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>14</td>\n",
       "      <td>805321</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>27.5</td>\n",
       "      <td>1.29466e+06</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>41</td>\n",
       "      <td>1.73002e+06</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>NaN</td>\n",
       "      <td>54</td>\n",
       "      <td>2.13424e+06</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>counts</th>\n",
       "      <td>3370464</td>\n",
       "      <td>3370464</td>\n",
       "      <td>3370464</td>\n",
       "      <td>3370464</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniques</th>\n",
       "      <td>16</td>\n",
       "      <td>54</td>\n",
       "      <td>3901</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_perc</th>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>types</th>\n",
       "      <td>date</td>\n",
       "      <td>numeric</td>\n",
       "      <td>numeric</td>\n",
       "      <td>bool</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                 date    store_nbr     item_nbr onpromotion\n",
       "count             NaN  3.37046e+06  3.37046e+06         NaN\n",
       "mean              NaN         27.5   1.2448e+06         NaN\n",
       "std               NaN      15.5858       589836         NaN\n",
       "min               NaN            1        96995         NaN\n",
       "25%               NaN           14       805321         NaN\n",
       "50%               NaN         27.5  1.29466e+06         NaN\n",
       "75%               NaN           41  1.73002e+06         NaN\n",
       "max               NaN           54  2.13424e+06         NaN\n",
       "counts        3370464      3370464      3370464     3370464\n",
       "uniques            16           54         3901           2\n",
       "missing             0            0            0           0\n",
       "missing_perc       0%           0%           0%          0%\n",
       "types            date      numeric      numeric        bool"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'OIL'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>dcoilwtico</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2013-01-01</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2013-01-02</td>\n",
       "      <td>93.1250</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2013-01-03</td>\n",
       "      <td>93.0000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2013-01-04</td>\n",
       "      <td>93.1250</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2013-01-07</td>\n",
       "      <td>93.1875</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        date  dcoilwtico\n",
       "0 2013-01-01         NaN\n",
       "1 2013-01-02     93.1250\n",
       "2 2013-01-03     93.0000\n",
       "3 2013-01-04     93.1250\n",
       "4 2013-01-07     93.1875"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>dcoilwtico</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>NaN</td>\n",
       "      <td>1175</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>NaN</td>\n",
       "      <td>inf</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>NaN</td>\n",
       "      <td>25.625</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>NaN</td>\n",
       "      <td>26.1875</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>46.4062</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>53.1875</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>95.6562</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>NaN</td>\n",
       "      <td>110.625</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>counts</th>\n",
       "      <td>1218</td>\n",
       "      <td>1175</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniques</th>\n",
       "      <td>1218</td>\n",
       "      <td>704</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing</th>\n",
       "      <td>0</td>\n",
       "      <td>43</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_perc</th>\n",
       "      <td>0%</td>\n",
       "      <td>3.53%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>types</th>\n",
       "      <td>date</td>\n",
       "      <td>numeric</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              date dcoilwtico\n",
       "count          NaN       1175\n",
       "mean           NaN        inf\n",
       "std            NaN     25.625\n",
       "min            NaN    26.1875\n",
       "25%            NaN    46.4062\n",
       "50%            NaN    53.1875\n",
       "75%            NaN    95.6562\n",
       "max            NaN    110.625\n",
       "counts        1218       1175\n",
       "uniques       1218        704\n",
       "missing          0         43\n",
       "missing_perc    0%      3.53%\n",
       "types         date    numeric"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'HOLIDAYS'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>type</th>\n",
       "      <th>locale</th>\n",
       "      <th>locale_name</th>\n",
       "      <th>description</th>\n",
       "      <th>transferred</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2012-03-02</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>Local</td>\n",
       "      <td>Manta</td>\n",
       "      <td>Fundacion de Manta</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2012-04-01</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>Regional</td>\n",
       "      <td>Cotopaxi</td>\n",
       "      <td>Provincializacion de Cotopaxi</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2012-04-12</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>Local</td>\n",
       "      <td>Cuenca</td>\n",
       "      <td>Fundacion de Cuenca</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2012-04-14</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>Local</td>\n",
       "      <td>Libertad</td>\n",
       "      <td>Cantonizacion de Libertad</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2012-04-21</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>Local</td>\n",
       "      <td>Riobamba</td>\n",
       "      <td>Cantonizacion de Riobamba</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        date     type    locale locale_name                    description  \\\n",
       "0 2012-03-02  Holiday     Local       Manta             Fundacion de Manta   \n",
       "1 2012-04-01  Holiday  Regional    Cotopaxi  Provincializacion de Cotopaxi   \n",
       "2 2012-04-12  Holiday     Local      Cuenca            Fundacion de Cuenca   \n",
       "3 2012-04-14  Holiday     Local    Libertad      Cantonizacion de Libertad   \n",
       "4 2012-04-21  Holiday     Local    Riobamba      Cantonizacion de Riobamba   \n",
       "\n",
       "   transferred  \n",
       "0        False  \n",
       "1        False  \n",
       "2        False  \n",
       "3        False  \n",
       "4        False  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>type</th>\n",
       "      <th>locale</th>\n",
       "      <th>locale_name</th>\n",
       "      <th>description</th>\n",
       "      <th>transferred</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>unique</th>\n",
       "      <td>312</td>\n",
       "      <td>6</td>\n",
       "      <td>3</td>\n",
       "      <td>24</td>\n",
       "      <td>103</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>top</th>\n",
       "      <td>2014-06-25 00:00:00</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>National</td>\n",
       "      <td>Ecuador</td>\n",
       "      <td>Carnaval</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>freq</th>\n",
       "      <td>4</td>\n",
       "      <td>221</td>\n",
       "      <td>174</td>\n",
       "      <td>174</td>\n",
       "      <td>10</td>\n",
       "      <td>338</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>first</th>\n",
       "      <td>2012-03-02 00:00:00</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>last</th>\n",
       "      <td>2017-12-26 00:00:00</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>counts</th>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "      <td>350</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniques</th>\n",
       "      <td>312</td>\n",
       "      <td>6</td>\n",
       "      <td>3</td>\n",
       "      <td>24</td>\n",
       "      <td>103</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_perc</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>types</th>\n",
       "      <td>date</td>\n",
       "      <td>categorical</td>\n",
       "      <td>categorical</td>\n",
       "      <td>categorical</td>\n",
       "      <td>categorical</td>\n",
       "      <td>bool</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                             date         type       locale  locale_name  \\\n",
       "count                         350          350          350          350   \n",
       "unique                        312            6            3           24   \n",
       "top           2014-06-25 00:00:00      Holiday     National      Ecuador   \n",
       "freq                            4          221          174          174   \n",
       "first         2012-03-02 00:00:00          NaN          NaN          NaN   \n",
       "last          2017-12-26 00:00:00          NaN          NaN          NaN   \n",
       "counts                        350          350          350          350   \n",
       "uniques                       312            6            3           24   \n",
       "missing                         0            0            0            0   \n",
       "missing_perc                   0%           0%           0%           0%   \n",
       "types                        date  categorical  categorical  categorical   \n",
       "\n",
       "              description transferred  \n",
       "count                 350         350  \n",
       "unique                103           2  \n",
       "top              Carnaval       False  \n",
       "freq                   10         338  \n",
       "first                 NaN         NaN  \n",
       "last                  NaN         NaN  \n",
       "counts                350         350  \n",
       "uniques               103           2  \n",
       "missing                 0           0  \n",
       "missing_perc           0%          0%  \n",
       "types         categorical        bool  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'STORES'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>city</th>\n",
       "      <th>state</th>\n",
       "      <th>type</th>\n",
       "      <th>cluster</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>store_nbr</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Quito</td>\n",
       "      <td>Pichincha</td>\n",
       "      <td>D</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Quito</td>\n",
       "      <td>Pichincha</td>\n",
       "      <td>D</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Quito</td>\n",
       "      <td>Pichincha</td>\n",
       "      <td>D</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Quito</td>\n",
       "      <td>Pichincha</td>\n",
       "      <td>D</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>Santo Domingo</td>\n",
       "      <td>Santo Domingo de los Tsachilas</td>\n",
       "      <td>D</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                    city                           state type  cluster\n",
       "store_nbr                                                             \n",
       "1                  Quito                       Pichincha    D       13\n",
       "2                  Quito                       Pichincha    D       13\n",
       "3                  Quito                       Pichincha    D        8\n",
       "4                  Quito                       Pichincha    D        9\n",
       "5          Santo Domingo  Santo Domingo de los Tsachilas    D        4"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>city</th>\n",
       "      <th>state</th>\n",
       "      <th>type</th>\n",
       "      <th>cluster</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>54</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>8.48148</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>4.69339</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>8.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>17</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>counts</th>\n",
       "      <td>54</td>\n",
       "      <td>54</td>\n",
       "      <td>54</td>\n",
       "      <td>54</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniques</th>\n",
       "      <td>22</td>\n",
       "      <td>16</td>\n",
       "      <td>5</td>\n",
       "      <td>17</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_perc</th>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>types</th>\n",
       "      <td>categorical</td>\n",
       "      <td>categorical</td>\n",
       "      <td>categorical</td>\n",
       "      <td>numeric</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     city        state         type  cluster\n",
       "count                 NaN          NaN          NaN       54\n",
       "mean                  NaN          NaN          NaN  8.48148\n",
       "std                   NaN          NaN          NaN  4.69339\n",
       "min                   NaN          NaN          NaN        1\n",
       "25%                   NaN          NaN          NaN        4\n",
       "50%                   NaN          NaN          NaN      8.5\n",
       "75%                   NaN          NaN          NaN       13\n",
       "max                   NaN          NaN          NaN       17\n",
       "counts                 54           54           54       54\n",
       "uniques                22           16            5       17\n",
       "missing                 0            0            0        0\n",
       "missing_perc           0%           0%           0%       0%\n",
       "types         categorical  categorical  categorical  numeric"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'ITEMS'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>family</th>\n",
       "      <th>class</th>\n",
       "      <th>perishable</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>item_nbr</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>96995</th>\n",
       "      <td>GROCERY I</td>\n",
       "      <td>1093</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>99197</th>\n",
       "      <td>GROCERY I</td>\n",
       "      <td>1067</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>103501</th>\n",
       "      <td>CLEANING</td>\n",
       "      <td>3008</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>103520</th>\n",
       "      <td>GROCERY I</td>\n",
       "      <td>1028</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>103665</th>\n",
       "      <td>BREAD/BAKERY</td>\n",
       "      <td>2712</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                family  class  perishable\n",
       "item_nbr                                 \n",
       "96995        GROCERY I   1093       False\n",
       "99197        GROCERY I   1067       False\n",
       "103501        CLEANING   3008       False\n",
       "103520       GROCERY I   1028       False\n",
       "103665    BREAD/BAKERY   2712        True"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>family</th>\n",
       "      <th>class</th>\n",
       "      <th>perishable</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>NaN</td>\n",
       "      <td>4100</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>NaN</td>\n",
       "      <td>2169.65</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>NaN</td>\n",
       "      <td>1484.91</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>NaN</td>\n",
       "      <td>1002</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>1068</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>2004</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>NaN</td>\n",
       "      <td>2990.5</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>NaN</td>\n",
       "      <td>7780</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>counts</th>\n",
       "      <td>4100</td>\n",
       "      <td>4100</td>\n",
       "      <td>4100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniques</th>\n",
       "      <td>33</td>\n",
       "      <td>337</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_perc</th>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "      <td>0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>types</th>\n",
       "      <td>categorical</td>\n",
       "      <td>numeric</td>\n",
       "      <td>bool</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   family    class perishable\n",
       "count                 NaN     4100        NaN\n",
       "mean                  NaN  2169.65        NaN\n",
       "std                   NaN  1484.91        NaN\n",
       "min                   NaN     1002        NaN\n",
       "25%                   NaN     1068        NaN\n",
       "50%                   NaN     2004        NaN\n",
       "75%                   NaN   2990.5        NaN\n",
       "max                   NaN     7780        NaN\n",
       "counts               4100     4100       4100\n",
       "uniques                33      337          2\n",
       "missing                 0        0          0\n",
       "missing_perc           0%       0%         0%\n",
       "types         categorical  numeric       bool"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for t in tables: display(t.name, t.head(), DataFrameSummary(t).summary())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Merget test/train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#prob don't need\n",
    "# last_train = train_df.iloc[-1].name\n",
    "all_df = train_df.append(test_df)\n",
    "# last_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_df_name='new_full_df.pkl'\n",
    "save_n_load_df(all_df, all_df_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "del(train_df); del(test_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Remove null data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "date           False\n",
       "item_nbr       False\n",
       "onpromotion     True\n",
       "store_nbr      False\n",
       "unit_sales      True\n",
       "dtype: bool"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_df.isnull().any()\n",
    "# of course ignore unit_sales since missing from test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "all_df.onpromotion.fillna(False, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "all_df['onpromotion']=all_df['onpromotion'].astype(np.int8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(all_df, all_df_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data clean / feature eng\n",
    "From https://github.com/fastai/courses/blob/master/deeplearning2/rossman.ipynb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### add dateparts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "add_datepart(all_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(all_df, all_df_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Joins"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Oil"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(all_df, oil_df, 'date')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined['dcoilwtico'].interpolate(inplace=True, limit_direction='both')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#make ints to simplify\n",
    "joined['dcoilwtico']=joined['dcoilwtico'].astype(np.int8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'oil_joined.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Stores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(joined, stores_df, 'store_nbr')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'joined_stores.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Items"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(joined, items_df, 'item_nbr')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'joined_items.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Transactions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined = join_df(joined, transactions_df, ['date', 'store_nbr'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'joined_transactions.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### HOliday"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### todo\n",
    "\n",
    "1. Add boolean loc_hol, reg_hol, nat_hol columns to holidays_df\n",
    "1. Add city, state to holiday_df based on locale_name\n",
    "1. merge local_hol with joined on date, city\n",
    "1. merge regional_hol with joined on date, state\n",
    "1. merge nat_hol with joined on date\n",
    "\n",
    "##### todo\n",
    "1. go back and check the holiday rules to see if these are actually correct"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#Local non-Holiday types are all holiday_like. Allows us to set cities['local_hol']=1 below\n",
    "ph.loc[(ph['locale']=='Local') & (ph['type']!='Holiday')]['type'].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#All regional are type Holiday. Allows us to set cities['local_hol']=1 below\n",
    "ph.loc[(ph['locale']=='Regional')]['type'].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Generate mergable holidays table"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#Using the locale_name to identify cities and states. Then we can merge with joined table.\n",
    "#Effectively ignoring locale Local and Regional since they don't align perfectly with either.\n",
    "#from https://www.kaggle.com/c/favorita-grocery-sales-forecasting/data\n",
    "holiday_like = ['Holiday', 'Additional', 'Transfer', 'Bridge']\n",
    "#copy of holiday df for data wrangling\n",
    "ph=holidays_df.copy()\n",
    "#cities, Local\n",
    "cities = stores_df[['city']].drop_duplicates() \n",
    "cities['locale_name'] = cities['city']\n",
    "cities['city_hol']=1\n",
    "ph=join_df(ph,cities, 'locale_name')\n",
    "#states, Regional\n",
    "states = stores_df[['state']].drop_duplicates() \n",
    "states['locale_name'] = states['state']\n",
    "states['state_hol']=1\n",
    "ph=join_df(ph,states, 'locale_name')\n",
    "#cleanup\n",
    "for p in ['city_hol', 'state_hol']:\n",
    "    ph[p].fillna(0, inplace=True)\n",
    "    ph[p]=ph[p].astype(np.int8)\n",
    "#National holiday and events. More complex, must check for holiday_like\n",
    "ph['nat_hol']=ph.apply(lambda row: row['locale']=='National' and row['type'] in holiday_like, axis=1).astype(np.int8)\n",
    "ph['nat_event']=ph.apply(lambda row:row['type']=='Event', axis=1).astype(np.int8)\n",
    "#TODO - Consider adding descriptions of holidays"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(ph, 'mergeable_holidays.pkl')\n",
    "# ph = load_df('mergeable_holidays.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "display(DataFrameSummary(ph).summary(), ph.info(), ph.head())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "city_holidays_df = ph.loc[ph['city_hol']==1, ['date', 'city', 'city_hol']].copy().reset_index(drop=True)\n",
    "state_holidays_df = ph.loc[ph['state_hol']==1, ['date', 'state', 'state_hol']].copy().reset_index(drop=True)\n",
    "nat_holidays_df = ph.loc[ph['nat_hol']==1, ['date', 'nat_hol']].copy().reset_index(drop=True)\n",
    "nat_events_df = ph.loc[ph['nat_event']==1, ['date', 'nat_event']].copy().reset_index(drop=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Merge with joined"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined = load_df('joined_transactions.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### city holidays"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>city</th>\n",
       "      <th>city_hol</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2012-03-02</td>\n",
       "      <td>Manta</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2012-04-12</td>\n",
       "      <td>Cuenca</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2012-04-14</td>\n",
       "      <td>Libertad</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2012-04-21</td>\n",
       "      <td>Riobamba</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2012-05-12</td>\n",
       "      <td>Puyo</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        date      city  city_hol\n",
       "0 2012-03-02     Manta         1\n",
       "1 2012-04-12    Cuenca         1\n",
       "2 2012-04-14  Libertad         1\n",
       "3 2012-04-21  Riobamba         1\n",
       "4 2012-05-12      Puyo         1"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "city_holidays_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(joined, city_holidays_df, ['date', 'city'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "fillna_with_zeros(joined, 'city_hol')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "chk = joined.loc[joined['city_hol']==1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>type</th>\n",
       "      <th>locale</th>\n",
       "      <th>locale_name</th>\n",
       "      <th>description</th>\n",
       "      <th>transferred</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>47</th>\n",
       "      <td>2013-04-01</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>Regional</td>\n",
       "      <td>Cotopaxi</td>\n",
       "      <td>Provincializacion de Cotopaxi</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date     type    locale locale_name                    description  \\\n",
       "47 2013-04-01  Holiday  Regional    Cotopaxi  Provincializacion de Cotopaxi   \n",
       "\n",
       "    transferred  \n",
       "47        False  "
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "holidays_df.loc[holidays_df['date']==chk.head()['date'].iloc[0]].head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### State holidays"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(joined, state_holidays_df, ['date', 'state'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "fillna_with_zeros(joined, 'state_hol')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>type</th>\n",
       "      <th>locale</th>\n",
       "      <th>locale_name</th>\n",
       "      <th>description</th>\n",
       "      <th>transferred</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>47</th>\n",
       "      <td>2013-04-01</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>Regional</td>\n",
       "      <td>Cotopaxi</td>\n",
       "      <td>Provincializacion de Cotopaxi</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date     type    locale locale_name                    description  \\\n",
       "47 2013-04-01  Holiday  Regional    Cotopaxi  Provincializacion de Cotopaxi   \n",
       "\n",
       "    transferred  \n",
       "47        False  "
      ]
     },
     "execution_count": 74,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "chk = joined.loc[joined['state_hol']==1]\n",
    "holidays_df.loc[holidays_df['date']==chk.head()['date'].iloc[0]].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>state</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Pichincha</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Santo Domingo de los Tsachilas</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>Cotopaxi</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>Chimborazo</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>Imbabura</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>Bolivar</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>Pastaza</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>Tungurahua</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>Guayas</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>Santa Elena</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>Los Rios</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>36</th>\n",
       "      <td>Azuay</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>37</th>\n",
       "      <td>Loja</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>39</th>\n",
       "      <td>El Oro</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>42</th>\n",
       "      <td>Esmeraldas</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>51</th>\n",
       "      <td>Manabi</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                             state\n",
       "0                        Pichincha\n",
       "4   Santo Domingo de los Tsachilas\n",
       "11                        Cotopaxi\n",
       "13                      Chimborazo\n",
       "14                        Imbabura\n",
       "18                         Bolivar\n",
       "21                         Pastaza\n",
       "22                      Tungurahua\n",
       "23                          Guayas\n",
       "24                     Santa Elena\n",
       "30                        Los Rios\n",
       "36                           Azuay\n",
       "37                            Loja\n",
       "39                          El Oro\n",
       "42                      Esmeraldas\n",
       "51                          Manabi"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "stores_df[['state']].drop_duplicates()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Nat holidays and events"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(joined, nat_holidays_df, 'date')\n",
    "joined=join_df(joined, nat_events_df, 'date')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "fillna_with_zeros(joined, 'nat_hol')\n",
    "fillna_with_zeros(joined, 'nat_event')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>item_nbr</th>\n",
       "      <th>onpromotion</th>\n",
       "      <th>store_nbr</th>\n",
       "      <th>unit_sales</th>\n",
       "      <th>year</th>\n",
       "      <th>month</th>\n",
       "      <th>week</th>\n",
       "      <th>day</th>\n",
       "      <th>dayofweek</th>\n",
       "      <th>...</th>\n",
       "      <th>type</th>\n",
       "      <th>cluster</th>\n",
       "      <th>family</th>\n",
       "      <th>class</th>\n",
       "      <th>perishable</th>\n",
       "      <th>transactions</th>\n",
       "      <th>city_hol</th>\n",
       "      <th>state_hol</th>\n",
       "      <th>nat_hol</th>\n",
       "      <th>nat_event</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2013-01-01</td>\n",
       "      <td>103665</td>\n",
       "      <td>False</td>\n",
       "      <td>25</td>\n",
       "      <td>7.0</td>\n",
       "      <td>2013</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>...</td>\n",
       "      <td>D</td>\n",
       "      <td>1</td>\n",
       "      <td>BREAD/BAKERY</td>\n",
       "      <td>2712</td>\n",
       "      <td>True</td>\n",
       "      <td>770.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2013-01-01</td>\n",
       "      <td>105574</td>\n",
       "      <td>False</td>\n",
       "      <td>25</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2013</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>...</td>\n",
       "      <td>D</td>\n",
       "      <td>1</td>\n",
       "      <td>GROCERY I</td>\n",
       "      <td>1045</td>\n",
       "      <td>False</td>\n",
       "      <td>770.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2013-01-01</td>\n",
       "      <td>105575</td>\n",
       "      <td>False</td>\n",
       "      <td>25</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2013</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>...</td>\n",
       "      <td>D</td>\n",
       "      <td>1</td>\n",
       "      <td>GROCERY I</td>\n",
       "      <td>1045</td>\n",
       "      <td>False</td>\n",
       "      <td>770.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2013-01-01</td>\n",
       "      <td>108079</td>\n",
       "      <td>False</td>\n",
       "      <td>25</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2013</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>...</td>\n",
       "      <td>D</td>\n",
       "      <td>1</td>\n",
       "      <td>GROCERY I</td>\n",
       "      <td>1030</td>\n",
       "      <td>False</td>\n",
       "      <td>770.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2013-01-01</td>\n",
       "      <td>108701</td>\n",
       "      <td>False</td>\n",
       "      <td>25</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2013</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>...</td>\n",
       "      <td>D</td>\n",
       "      <td>1</td>\n",
       "      <td>DELI</td>\n",
       "      <td>2644</td>\n",
       "      <td>True</td>\n",
       "      <td>770.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 23 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        date  item_nbr  onpromotion  store_nbr  unit_sales  year  month  week  \\\n",
       "0 2013-01-01    103665        False         25         7.0  2013      1     1   \n",
       "1 2013-01-01    105574        False         25         1.0  2013      1     1   \n",
       "2 2013-01-01    105575        False         25         2.0  2013      1     1   \n",
       "3 2013-01-01    108079        False         25         1.0  2013      1     1   \n",
       "4 2013-01-01    108701        False         25         1.0  2013      1     1   \n",
       "\n",
       "   day  dayofweek    ...      type cluster        family class  perishable  \\\n",
       "0    1          1    ...         D       1  BREAD/BAKERY  2712        True   \n",
       "1    1          1    ...         D       1     GROCERY I  1045       False   \n",
       "2    1          1    ...         D       1     GROCERY I  1045       False   \n",
       "3    1          1    ...         D       1     GROCERY I  1030       False   \n",
       "4    1          1    ...         D       1          DELI  2644        True   \n",
       "\n",
       "  transactions  city_hol  state_hol  nat_hol  nat_event  \n",
       "0        770.0         0          0        1          0  \n",
       "1        770.0         0          0        1          0  \n",
       "2        770.0         0          0        1          0  \n",
       "3        770.0         0          0        1          0  \n",
       "4        770.0         0          0        1          0  \n",
       "\n",
       "[5 rows x 23 columns]"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "chk = joined.loc[joined['nat_hol']==1].head(); chk"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>type</th>\n",
       "      <th>locale</th>\n",
       "      <th>locale_name</th>\n",
       "      <th>description</th>\n",
       "      <th>transferred</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>41</th>\n",
       "      <td>2013-01-01</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>National</td>\n",
       "      <td>Ecuador</td>\n",
       "      <td>Primer dia del ano</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date     type    locale locale_name         description  transferred\n",
       "41 2013-01-01  Holiday  National     Ecuador  Primer dia del ano        False"
      ]
     },
     "execution_count": 81,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "holidays_df.loc[holidays_df['date']==chk.head()['date'].iloc[0]].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date</th>\n",
       "      <th>type</th>\n",
       "      <th>locale</th>\n",
       "      <th>locale_name</th>\n",
       "      <th>description</th>\n",
       "      <th>transferred</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>54</th>\n",
       "      <td>2013-05-12</td>\n",
       "      <td>Holiday</td>\n",
       "      <td>Local</td>\n",
       "      <td>Puyo</td>\n",
       "      <td>Cantonizacion del Puyo</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>55</th>\n",
       "      <td>2013-05-12</td>\n",
       "      <td>Event</td>\n",
       "      <td>National</td>\n",
       "      <td>Ecuador</td>\n",
       "      <td>Dia de la Madre</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date     type    locale locale_name             description  \\\n",
       "54 2013-05-12  Holiday     Local        Puyo  Cantonizacion del Puyo   \n",
       "55 2013-05-12    Event  National     Ecuador         Dia de la Madre   \n",
       "\n",
       "    transferred  \n",
       "54        False  \n",
       "55        False  "
      ]
     },
     "execution_count": 82,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Have to run later, no national events in training set\n",
    "chk = joined.loc[joined['nat_event']==1].head()\n",
    "holidays_df.loc[holidays_df['date']==chk.head()['date'].iloc[0]].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 83,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'joined_holidays.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Calculate time durations"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 30-day percentage oil change"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Timestamp('2013-01-01 00:00:00')"
      ]
     },
     "execution_count": 84,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dates=joined['date'].drop_duplicates(); dates[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f6f9436f0b8>"
      ]
     },
     "execution_count": 87,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX8AAAD9CAYAAABUS3cAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n/yRJkiHkH/vP08yK9B4EAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f6fa3c395c0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "oil_prices = joined[['date', 'dcoilwtico']].copy().drop_duplicates(['date']).reset_index(drop=True)\n",
    "oil_prices['oil_30_day_change'] = oil_prices['dcoilwtico'].pct_change(periods=30).fillna(method='bfill')\n",
    "oil_prices[['date', 'oil_30_day_change']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "oil_prices.drop(labels='dcoilwtico', axis=1, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(joined, oil_prices, 'date')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'joined_30_day_oil.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Elapsed since events"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# joined = load_df('joined_30_day_oil.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "columns=['date', 'store_nbr', 'state_hol', 'city_hol', 'nat_hol', 'nat_event']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#clean up, using last_store and store_nbr\n",
    "class elapsed(object):\n",
    "    def __init__(self, fld):\n",
    "        self.fld = fld\n",
    "        self.last = pd.to_datetime(np.nan)\n",
    "        self.last_store = 0\n",
    "        \n",
    "    def get(self, row):\n",
    "        if row.store_nbr != self.last_store:\n",
    "            self.last = pd.to_datetime(np.nan)\n",
    "            self.last_store = row.store_nbr\n",
    "        if (row[self.fld]): self.last = row.date\n",
    "        return row.date-self.last\n",
    "\n",
    "def add_elapsed(fld, prefix):\n",
    "    sh_el = elapsed(fld)\n",
    "    df[prefix+fld] = df.apply(sh_el.get, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df=joined[columns].copy()\n",
    "df.drop_duplicates(inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Begin samp test to see if this code works"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "def add_elapsed_samp(fld, prefix):\n",
    "    sh_el = elapsed(fld)\n",
    "    samp[prefix+fld] = samp.apply(sh_el.get, axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "samp=pd.concat([df[0:5]]*5).reset_index(drop=True)\n",
    "samp.iloc[5:10]['date'] = samp.iloc[0].date + pd.DateOffset(1)\n",
    "samp.iloc[10:15]['date'] = samp.iloc[0].date + pd.DateOffset(2)\n",
    "samp.iloc[15:20]['date'] = samp.iloc[0].date + pd.DateOffset(3)\n",
    "samp.iloc[20:25]['date'] = samp.iloc[0].date + pd.DateOffset(4)\n",
    "samp=samp.sort_values(['store_nbr', 'date'])\n",
    "samp.set_value(10, 'state_hol', 1)\n",
    "samp.set_value(12, 'city_hol', 1)\n",
    "samp.set_value(17, 'city_hol', 1)\n",
    "samp.set_value(22, 'city_hol', 1)\n",
    "samp.set_value(6, 'nat_hol', 1)\n",
    "samp.set_value(13, 'nat_event', 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "for fld in holidays:\n",
    "    add_elapsed_samp(fld, 'after_')\n",
    "samp = samp.sort_values(['store_nbr', 'date'], ascending=[True, False])\n",
    "for fld in holidays:\n",
    "    add_elapsed_samp(fld, 'before_')\n",
    "hol_dur = [o+p for o in ['before_', 'after_'] for p in holidays]\n",
    "for a in hol_dur:\n",
    "    samp[a]=samp[a].fillna(pd.Timedelta(0)).dt.days\n",
    "samp"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "samp = samp.set_index('date')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "bwd = samp[['store_nbr']+ holidays].sort_index().groupby('store_nbr').rolling(7, min_periods=1).sum()\n",
    "fwd = samp[['store_nbr']+ holidays].sort_index(ascending=False).groupby('store_nbr').rolling(7, min_periods=1).sum()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "bwd"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "bwd.drop('store_nbr', 1, inplace=True)\n",
    "bwd.reset_index(inplace=True)\n",
    "fwd.drop('store_nbr', 1, inplace=True)\n",
    "fwd.reset_index(inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "samp.reset_index(inplace=True)\n",
    "samp=samp.merge(bwd, 'left', ['date', 'store_nbr'], suffixes=['', '_bw'])\n",
    "samp=samp.merge(fwd, 'left', ['date', 'store_nbr'], suffixes=['', '_fw'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "samp[['date','store_nbr', 'city_hol', 'city_hol_fw']]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## End samp test, seems fine"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "fld='state_hol'\n",
    "df=df.sort_values(['store_nbr', 'date'])\n",
    "add_elapsed(fld, 'after')\n",
    "#df=df.sort_values(['store_nbr', 'date'], ascending=[True, False])\n",
    "#add_elapsed(fld, 'before')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "holidays=['state_hol', 'city_hol', 'nat_hol', 'nat_event']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df=df.sort_values(['store_nbr', 'date'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4/4 [00:42<00:00, 10.79s/it]\n"
     ]
    }
   ],
   "source": [
    "for fld in tqdm(holidays):\n",
    "    add_elapsed(fld, 'after_')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df=df.sort_values(['store_nbr', 'date'], ascending=[True, False])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4/4 [00:49<00:00, 12.34s/it]\n"
     ]
    }
   ],
   "source": [
    "for fld in tqdm(holidays):\n",
    "    add_elapsed(fld, 'before_')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 101,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(df, 'elapsed_events_df.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 327,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# df=load_df('elapsed_events_df.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "hol_dur = [o+p for o in ['before_', 'after_'] for p in holidays]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 329,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#since there are no nat_event\n",
    "#don't run on full dataframe later!!!!\n",
    "# df['after_nat_event']=pd.Series([0 for x in range(len(df.index))], index=df.index)\n",
    "# df['before_nat_event']=pd.Series([0 for x in range(len(df.index))], index=df.index)\n",
    "# hol_dur.remove('after_nat_event')\n",
    "# hol_dur.remove('before_nat_event')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 8/8 [00:04<00:00,  1.61it/s]\n"
     ]
    }
   ],
   "source": [
    "for a in tqdm(hol_dur):\n",
    "    df[a]=df[a].fillna(pd.Timedelta(0)).dt.days"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 104,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(df, 'filled_elapsed_events_df.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Now rolling windows"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 333,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# df = load_df('filled_elapsed_events_df.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df=df.set_index('date')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "bwd = df[['store_nbr']+ holidays].sort_index().groupby('store_nbr').rolling(7, min_periods=1).sum()\n",
    "fwd = df[['store_nbr']+ holidays].sort_index(ascending=False).groupby('store_nbr').rolling(7, min_periods=1).sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "bwd.drop('store_nbr', 1, inplace=True)\n",
    "bwd.reset_index(inplace=True)\n",
    "fwd.drop('store_nbr', 1, inplace=True)\n",
    "fwd.reset_index(inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df.reset_index(inplace=True)\n",
    "df=df.merge(bwd, 'left', ['date', 'store_nbr'], suffixes=['', '_bw'])\n",
    "df=df.merge(fwd, 'left', ['date', 'store_nbr'], suffixes=['', '_fw'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df.drop(holidays, 1, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 110,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(df, 'rolled_filled_elapsed_events_df.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 84470 entries, 0 to 84469\n",
      "Data columns (total 18 columns):\n",
      "date                84470 non-null datetime64[ns]\n",
      "store_nbr           84470 non-null uint8\n",
      "after_state_hol     84470 non-null int64\n",
      "after_city_hol      84470 non-null int64\n",
      "after_nat_hol       84470 non-null int64\n",
      "after_nat_event     84470 non-null int64\n",
      "before_state_hol    84470 non-null int64\n",
      "before_city_hol     84470 non-null int64\n",
      "before_nat_hol      84470 non-null int64\n",
      "before_nat_event    84470 non-null int64\n",
      "state_hol_bw        84470 non-null float64\n",
      "city_hol_bw         84470 non-null float64\n",
      "nat_hol_bw          84470 non-null float64\n",
      "nat_event_bw        84470 non-null float64\n",
      "state_hol_fw        84470 non-null float64\n",
      "city_hol_fw         84470 non-null float64\n",
      "nat_hol_fw          84470 non-null float64\n",
      "nat_event_fw        84470 non-null float64\n",
      "dtypes: datetime64[ns](1), float64(8), int64(8), uint8(1)\n",
      "memory usage: 11.7 MB\n"
     ]
    }
   ],
   "source": [
    "# df=load_df('rolled_filled_elapsed_events_df.pkl')\n",
    "df.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Cleanup to save memory on merge"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cols=df.columns.tolist()\n",
    "cols.remove('date')\n",
    "cols.remove('store_nbr')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 84470 entries, 0 to 84469\n",
      "Data columns (total 18 columns):\n",
      "date                84470 non-null datetime64[ns]\n",
      "store_nbr           84470 non-null uint8\n",
      "after_state_hol     84470 non-null int16\n",
      "after_city_hol      84470 non-null int16\n",
      "after_nat_hol       84470 non-null int16\n",
      "after_nat_event     84470 non-null int16\n",
      "before_state_hol    84470 non-null int16\n",
      "before_city_hol     84470 non-null int16\n",
      "before_nat_hol      84470 non-null int16\n",
      "before_nat_event    84470 non-null int16\n",
      "state_hol_bw        84470 non-null int16\n",
      "city_hol_bw         84470 non-null int16\n",
      "nat_hol_bw          84470 non-null int16\n",
      "nat_event_bw        84470 non-null int16\n",
      "state_hol_fw        84470 non-null int16\n",
      "city_hol_fw         84470 non-null int16\n",
      "nat_hol_fw          84470 non-null int16\n",
      "nat_event_fw        84470 non-null int16\n",
      "dtypes: datetime64[ns](1), int16(16), uint8(1)\n",
      "memory usage: 3.9 MB\n"
     ]
    }
   ],
   "source": [
    "for c in cols: df[c]=df[c].astype(np.int16)\n",
    "df.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(joined, df, ['date', 'store_nbr'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 115,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'joined_elapsed_events.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "False    121027360\n",
       "True       8026704\n",
       "Name: onpromotion, dtype: int64"
      ]
     },
     "execution_count": 117,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joined['onpromotion'].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Promotion durations\n",
    "Promotions are unique to date, item and store"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=load_df('joined_elapsed_events.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "promo_df = joined[['date', 'item_nbr', 'store_nbr', 'onpromotion']].copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#future - look at ways to speed up or free up memory by deleting objects.\n",
    "class itm_elapsed(object):\n",
    "    def __init__(self, fld, total):\n",
    "        self.fld = fld\n",
    "        self.last = pd.to_datetime(np.nan)\n",
    "        self.last_store_nbr = 0\n",
    "        self.last_item_nbr = 0\n",
    "        self.count=0\n",
    "        self.total=total\n",
    "        \n",
    "    def get(self, row):\n",
    "        self.count+=1\n",
    "        if (not self.count % 1e6):\n",
    "            print('Another 1 million out of {}'.format(self.total))\n",
    "        if (row.item_nbr != self.last_item_nbr) or (row.store_nbr != self.last_store_nbr):\n",
    "            self.last = pd.to_datetime(np.nan)\n",
    "            self.last_item_nbr = row.item_nbr\n",
    "            self.last_store_nbr = row.store_nbr\n",
    "        if (row[self.fld]): self.last = row.date\n",
    "        return row.date-self.last\n",
    "\n",
    "def itm_add_elapsed(fld, prefix):\n",
    "    sh_el = itm_elapsed(fld, total=len(promo_df)//1e6)\n",
    "    promo_df[prefix+fld] = promo_df.apply(sh_el.get, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'joined' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-7-ae83d0694327>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0;32mdel\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mjoined\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m: name 'joined' is not defined"
     ]
    }
   ],
   "source": [
    "del(joined)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "promo_df['onpromotion']=promo_df['onpromotion'].astype(np.int8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(promo_df, 'promo_df1.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "promo_df.sort_values(['item_nbr','store_nbr', 'date'], inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(promo_df, 'promo_df2.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n"
     ]
    }
   ],
   "source": [
    "itm_add_elapsed('onpromotion', 'after_')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(promo_df, 'promo_df3.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "promo_df.sort_values(['item_nbr','store_nbr', 'date'], ascending=[True, True, False], inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(promo_df, 'promo_df4.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n",
      "Another 1 million out of 129.0\n"
     ]
    }
   ],
   "source": [
    "itm_add_elapsed('onpromotion', 'before_')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(promo_df, 'promo_df5.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# promo_df = load_df('promo_df_elapsed.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "promo_df['after_onpromotion']=promo_df['after_onpromotion'].fillna(pd.Timedelta(0)).dt.days\n",
    "promo_df['before_onpromotion']=promo_df['before_onpromotion'].fillna(pd.Timedelta(0)).dt.days"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0     6215995\n",
       "1      234142\n",
       "2      201832\n",
       "3      192847\n",
       "4      188255\n",
       "5      183151\n",
       "6      179268\n",
       "8       48578\n",
       "10      46766\n",
       "9       46486\n",
       "11      46133\n",
       "12      45193\n",
       "13      43937\n",
       "7       43304\n",
       "15      34913\n",
       "14      34053\n",
       "16      23446\n",
       "17      23149\n",
       "18      23030\n",
       "19      22483\n",
       "20      22337\n",
       "21      20539\n",
       "22      20436\n",
       "24      19822\n",
       "25      19760\n",
       "23      19734\n",
       "26      19354\n",
       "27      19260\n",
       "28      18386\n",
       "29      18189\n",
       "       ...   \n",
       "34      14332\n",
       "35      13899\n",
       "36      13742\n",
       "37      11951\n",
       "38      11593\n",
       "39      11558\n",
       "40      11279\n",
       "41      11049\n",
       "42      10931\n",
       "43      10406\n",
       "44      10264\n",
       "45      10140\n",
       "47      10106\n",
       "46      10106\n",
       "48       9836\n",
       "49       9611\n",
       "50       9410\n",
       "51       6705\n",
       "52       6182\n",
       "53       5780\n",
       "54       4854\n",
       "55       4560\n",
       "56       4272\n",
       "57       4045\n",
       "58       2000\n",
       "59       1489\n",
       "60       1212\n",
       "61        995\n",
       "62        753\n",
       "63        142\n",
       "Name: after_onpromotion, Length: 64, dtype: int64"
      ]
     },
     "execution_count": 145,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "promo_df['after_onpromotion'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "promo_df['after_onpromotion'] = promo_df['after_onpromotion'].astype(np.int16)\n",
    "promo_df['before_onpromotion'] = promo_df['before_onpromotion'].astype(np.int16)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(promo_df, 'promo_df6.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "#### Future - add rolling for onpromotion"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "promo_df.drop('onpromotion', 1, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=load_df('joined_elapsed_events.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=join_df(joined, promo_df, ['date', 'store_nbr', 'item_nbr'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'promo_df' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-34-7e1e7c3aa6a9>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0;32mdel\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpromo_df\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m: name 'promo_df' is not defined"
     ]
    }
   ],
   "source": [
    "del(promo_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'joined_promo_bef_af.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined['onpromotion']=joined['onpromotion'].astype(np.int8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined['oil_30_day_change']=joined['oil_30_day_change'].astype(np.float32)\n",
    "joined['transactions']=joined['transactions'].astype(np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save_n_load_df(joined, 'joined_promo_bef_af2.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Rolling transactions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined=load_df('joined_promo_bef_af2.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "trans = joined[['date', 'store_nbr','transactions']].drop_duplicates()\n",
    "trans.set_index('date', inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "roll_17=trans.sort_index().groupby('store_nbr').rolling(17,\n",
    "    min_periods=1).mean().rename(columns={'transactions': 'tr_roll_17'}).astype(np.int16)\n",
    "roll_90=trans.sort_index().groupby('store_nbr').rolling(90,\n",
    "    min_periods=1).mean().rename(columns={'transactions': 'tr_roll_90'}).astype(np.int16)\n",
    "roll_360=trans.sort_index().groupby('store_nbr').rolling(360,\n",
    "    min_periods=1).mean().rename(columns={'transactions': 'tr_roll_360'}).astype(np.int16)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for dfr in [roll_17, roll_90, roll_360]:\n",
    "    dfr.drop('store_nbr', 1, inplace=True)\n",
    "    dfr.reset_index(inplace=True) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "tr_roll = roll_17.merge(roll_90, 'left', ['date', 'store_nbr']).merge(roll_360, 'left', ['date', 'store_nbr'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined = joined.merge(tr_roll, 'left', ['date', 'store_nbr'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "if save_n_load_df(joined, 'joined_roll_tran.pkl'): del(tr_roll)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create features for NN"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Future - not sure how to deal with transactions. Perhaps predict both sales and transactions in model?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cls = joined.columns.tolist()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "['date',\n",
    " 'item_nbr',\n",
    " 'onpromotion',\n",
    " 'store_nbr',\n",
    " 'unit_sales',\n",
    " 'year',\n",
    " 'month',\n",
    " 'week',\n",
    " 'day',\n",
    " 'dayofweek',\n",
    " 'dcoilwtico',\n",
    " 'city',\n",
    " 'state',\n",
    " 'type',\n",
    " 'cluster',\n",
    " 'family',\n",
    " 'class',\n",
    " 'perishable',\n",
    " 'transactions',\n",
    " 'state_hol',\n",
    " 'city_hol',\n",
    " 'nat_hol',\n",
    " 'nat_event',\n",
    " 'oil_30_day_change',\n",
    " 'after_state_hol',\n",
    " 'after_city_hol',\n",
    " 'after_nat_hol',\n",
    " 'after_nat_event',\n",
    " 'before_state_hol',\n",
    " 'before_city_hol',\n",
    " 'before_nat_hol',\n",
    " 'before_nat_event',\n",
    " 'state_hol_bw',\n",
    " 'city_hol_bw',\n",
    " 'nat_hol_bw',\n",
    " 'nat_event_bw',\n",
    " 'state_hol_fw',\n",
    " 'city_hol_fw',\n",
    " 'nat_hol_fw',\n",
    " 'nat_event_fw',\n",
    " 'after_onpromotion',\n",
    " 'before_onpromotion']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cat_var_dict = {'item_nbr': 50,\n",
    "                'onpromotion': 3,\n",
    "                'store_nbr': 25, \n",
    "                'year': 3,\n",
    "                'month': 6,\n",
    "                'week': 6,\n",
    "                'day': 10,\n",
    "                'dayofweek': 10,\n",
    "                'city': 10,\n",
    "                'state': 10,\n",
    "                'type': 5,\n",
    "                'cluster': 10,\n",
    "                'family': 15,\n",
    "                'class': 20,\n",
    "                'perishable': 5,\n",
    "                'state_hol': 3,\n",
    "                'city_hol': 3,\n",
    "                'nat_hol': 3,\n",
    "                'nat_event': 3,\n",
    "                'state_hol_bw': 2,\n",
    "                'city_hol_bw': 2,\n",
    "                'nat_hol_bw': 2,\n",
    "                'nat_event_bw': 2,\n",
    "                'state_hol_fw': 2,\n",
    "                'city_hol_fw': 2,\n",
    "                'nat_hol_fw': 2,\n",
    "                'nat_event_fw': 2}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pickle.dump(cat_var_dict, open(slowdata + 'cat_var_dict.pkl', 'wb'))\n",
    "cat_var_dict = pickle.load(open(slowdata + 'cat_var_dict.pkl', 'rb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cat_vars = [o[0] for o in \n",
    "            sorted(cat_var_dict.items(), key=lambda x: x[1], reverse=True)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "contin_vars = ['dcoilwtico',\n",
    "               'oil_30_day_change',\n",
    "               'after_state_hol',\n",
    "               'after_city_hol',\n",
    "               'after_nat_hol',\n",
    "               'after_nat_event',\n",
    "               'before_state_hol',\n",
    "               'before_city_hol',\n",
    "               'before_nat_hol',\n",
    "               'before_nat_event',\n",
    "               'after_onpromotion',\n",
    "               'before_onpromotion',\n",
    "               'tr_roll_17',\n",
    "              'tr_roll_90',\n",
    "              'tr_roll_360']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#check, should miss date, unit_sales, transactions\n",
    "len(cls) ==  len(cat_var_dict) + len(contin_vars) + 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cat_maps = [(o, LabelEncoder()) for o in cat_vars]\n",
    "contin_maps = [([o], StandardScaler()) for o in contin_vars]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "27"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cat_mapper = DataFrameMapper(cat_maps)m\n",
    "cat_cols = len(cat_map_fit.features); cat_cols"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/kevin/anaconda2/envs/ai/lib/python3.6/site-packages/sklearn/utils/validation.py:444: DataConversionWarning: Data with input dtype int8 was converted to float64 by StandardScaler.\n",
      "  warnings.warn(msg, DataConversionWarning)\n",
      "/home/kevin/anaconda2/envs/ai/lib/python3.6/site-packages/sklearn/utils/validation.py:444: DataConversionWarning: Data with input dtype int16 was converted to float64 by StandardScaler.\n",
      "  warnings.warn(msg, DataConversionWarning)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "15"
      ]
     },
     "execution_count": 101,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "contin_mapper = DataFrameMapper(contin_maps)\n",
    "contin_map_fit = contin_mapper.fit(joined)\n",
    "contin_cols = len(contin_map_fit.features); contin_cols"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pickle.dump(contin_map_fit, open(slowdata + 'contin_maps.pickle', 'wb'))\n",
    "pickle.dump(cat_map_fit, open(slowdata + 'cat_maps.pickle', 'wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "contin_map_fit = pickle.load(open(slowdata + 'contin_maps.pickle', 'rb'))\n",
    "cat_map_fit = pickle.load(open(slowdata + 'cat_maps.pickle', 'rb'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Split training/test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "129427184"
      ]
     },
     "execution_count": 103,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(joined)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined_train_df = joined[joined.unit_sales.notnull()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "126056720"
      ]
     },
     "execution_count": 105,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(joined_train_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined_test_df = joined[joined.unit_sales.isnull()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 107,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(joined_train_df)+len(joined_test_df)==len(joined)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined_train_df.to_pickle(slowdata+'joined_train_df.pkl')\n",
    "joined_test_df.to_pickle(slowdata+'joined_test_df.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joined_train_df = pd.read_pickle(slowdata+'joined_train_df.pkl')\n",
    "joined_test_df = pd.read_pickle(slowdata+'joined_test_df.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load data code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "slowdata = data_path + 'slowdisk/'\n",
    "slowdata = slowdata + 'alldata/'\n",
    "results_path = results_path+'alldata/'\n",
    "joined_train_df = pd.read_pickle(slowdata+'joined_train_df.pkl')\n",
    "joined_test_df = pd.read_pickle(slowdata+'joined_test_df.pkl')\n",
    "contin_map_fit = pickle.load(open(slowdata + 'contin_maps.pickle', 'rb'))\n",
    "cat_map_fit = pickle.load(open(slowdata + 'cat_maps.pickle', 'rb'))\n",
    "cat_var_dict = pickle.load(open(slowdata + 'cat_var_dict.pkl', 'rb'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Split train/valid"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_ratio = 0.95\n",
    "train_size = int(train_ratio*len(joined_train_df))\n",
    "joined_train = joined_train_df[:train_size]\n",
    "joined_valid = joined_train_df[train_size:]\n",
    "len(joined_train)+len(joined_valid)==len(joined_train_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Sample Weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#from rules, perishable weighted as 1.25\n",
    "sample_weight = np.array([1 if not p else 1.25 for p in joined_train.perishable])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pickle.dump(sample_weight, open(slowdata + 'sample_weight.pkl', 'wb'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preprocess"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def cat_preproc(dat):\n",
    "    return cat_map_fit.transform(dat).astype(np.int16)\n",
    "def contin_preproc(dat):\n",
    "    return contin_map_fit.transform(dat).astype(np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cat_map_train = cat_preproc(joined_train)\n",
    "cat_map_valid = cat_preproc(joined_valid)\n",
    "cat_map_test = cat_preproc(joined_test_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "contin_map_train = contin_preproc(joined_train)\n",
    "contin_map_valid = contin_preproc(joined_valid)\n",
    "contin_map_test = contin_preproc(joined_test_df)\n",
    "contin_cols=contin_map_train.shape[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pickle.dump(cat_map_train, open(slowdata + 'cat_map_train.pkl', 'wb'), protocol=4)\n",
    "pickle.dump(cat_map_valid, open(slowdata + 'cat_map_valid.pkl', 'wb'))\n",
    "pickle.dump(contin_map_train, open(slowdata + 'contin_map_train.pkl', 'wb'), protocol=4)\n",
    "pickle.dump(contin_map_valid, open(slowdata + 'contin_map_valid.pkl', 'wb'))\n",
    "pickle.dump(cat_map_test, open(slowdata + 'cat_map_test.pkl', 'wb'))\n",
    "pickle.dump(contin_map_test, open(slowdata + 'contin_map_test.pkl', 'wb'))\n",
    "pickle.dump(contin_cols, open(slowdata + 'contin_cols.pkl', 'wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "save_df(joined_train, 'join_train_all.pkl')\n",
    "save_df(joined_valid, 'joined_valid_all.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cat_map_train = pickle.load(open(slowdata + 'cat_map_train.pkl', 'rb'))\n",
    "cat_map_valid = pickle.load(open(slowdata + 'cat_map_valid.pkl', 'rb'))\n",
    "contin_map_train = pickle.load(open(slowdata + 'contin_map_train.pkl', 'rb'))\n",
    "contin_map_valid = pickle.load(open(slowdata + 'contin_map_valid.pkl', 'rb'))\n",
    "sample_weight = pickle.load(open(slowdata + 'sample_weight.pkl', 'rb'))\n",
    "cat_map_fit = pickle.load(open(slowdata + 'cat_maps.pickle', 'rb'))\n",
    "contin_map_fit = pickle.load(open(slowdata + 'contin_maps.pickle', 'rb'))\n",
    "cat_var_dict = pickle.load(open(slowdata + 'cat_var_dict.pkl', 'rb'))\n",
    "cat_map_test = pickle.load(open(slowdata + 'cat_map_test.pkl', 'rb'))\n",
    "contin_map_test = pickle.load(open(slowdata + 'contin_map_test.pkl', 'rb'))\n",
    "contin_cols = pickle.load(open(slowdata + 'contin_cols.pkl', 'rb'))\n",
    "joined_train = load_df('join_train_all.pkl')\n",
    "joined_valid = load_df('joined_valid_all.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create model data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "map_train = split_cols(cat_map_train) + [contin_map_train]\n",
    "map_valid = split_cols(cat_map_valid) + [contin_map_valid]\n",
    "map_test = split_cols(cat_map_test) + [contin_map_test]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "meany_train, stdevy_train, y_train = normy(joined_train.as_matrix(columns = ['unit_sales']))\n",
    "meany_valid, stdevy_valid, y_valid = normy(joined_valid.as_matrix(columns = ['unit_sales']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#remove normalization\n",
    "y_train = joined_train.as_matrix(columns = ['unit_sales'])\n",
    "y_valid = joined_valid.as_matrix(columns = ['unit_sales'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Save model data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pickle.dump(map_train, open(slowdata + 'map_train.pkl', 'wb'), protocol=4)\n",
    "pickle.dump(map_valid, open(slowdata + 'map_valid.pkl', 'wb'))\n",
    "pickle.dump(map_test, open(slowdata + 'map_test.pkl', 'wb'))\n",
    "\n",
    "pickle.dump(meany_train, open(slowdata + 'meany_train.pkl', 'wb'))\n",
    "pickle.dump(stdevy_train, open(slowdata + 'stdevy_train.pkl', 'wb'))\n",
    "pickle.dump(y_train, open(slowdata + 'y_train.pkl', 'wb'))\n",
    "pickle.dump(meany_valid, open(slowdata + 'meany_valid.pkl', 'wb'))\n",
    "pickle.dump(stdevy_valid, open(slowdata + 'stdevy_valid.pkl', 'wb'))\n",
    "pickle.dump(y_valid, open(slowdata + 'y_valid.pkl', 'wb'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load model data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cat_var_dict = pickle.load(open(slowdata + 'cat_var_dict.pkl', 'rb'))\n",
    "contin_cols = pickle.load(open(slowdata + 'contin_cols.pkl', 'rb'))\n",
    "cat_map_fit = pickle.load(open(slowdata + 'cat_maps.pickle', 'rb'))\n",
    "\n",
    "map_train = pickle.load(open(slowdata + 'map_train.pkl', 'rb'))\n",
    "map_valid = pickle.load(open(slowdata + 'map_valid.pkl', 'rb'))\n",
    "map_test = pickle.load(open(slowdata + 'map_test.pkl', 'rb'))\n",
    "\n",
    "sample_weight = pickle.load(open(slowdata + 'sample_weight.pkl', 'rb'))\n",
    "\n",
    "meany_train = pickle.load(open(slowdata + 'meany_train.pkl', 'rb'))\n",
    "stdevy_train = pickle.load(open(slowdata + 'stdevy_train.pkl', 'rb'))\n",
    "y_train = pickle.load(open(slowdata + 'y_train.pkl', 'rb'))\n",
    "meany_valid = pickle.load(open(slowdata + 'meany_valid.pkl', 'rb'))\n",
    "stdevy_valid = pickle.load(open(slowdata + 'stdevy_valid.pkl', 'rb'))\n",
    "y_valid = pickle.load(open(slowdata + 'y_valid.pkl', 'rb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "rec_mem()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}