Instacart Market Basket Analysis.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Instacart Market Basket Analysis\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\", message=\"numpy.dtype size changed\")\n",
    "warnings.filterwarnings(\"ignore\", message=\"numpy.ufunc size changed\")\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pylab as plt\n",
    "import seaborn as sns\n",
    "import os\n",
    "color = sns.color_palette()\n",
    "\n",
    "from datetime import datetime\n",
    "import lightgbm as lgb\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "priors = pd.read_csv('order_products__prior.csv', dtype={\n",
    "            'order_id': np.int32,\n",
    "            'product_id': np.int32,\n",
    "            'add_to_cart_order': np.int16,\n",
    "            'reordered': np.int8})\n",
    "\n",
    "train = pd.read_csv('order_products__train.csv', dtype={\n",
    "            'order_id': np.int32,\n",
    "            'product_id': np.int32,\n",
    "            'add_to_cart_order': np.int16,\n",
    "            'reordered': np.int8})\n",
    "\n",
    "orders = pd.read_csv('orders.csv', dtype={\n",
    "        'order_id': np.int32,\n",
    "        'user_id': np.int32,\n",
    "        'eval_set': 'category',\n",
    "        'order_number': np.int16,\n",
    "        'order_dow': np.int8,\n",
    "        'order_hour_of_day': np.int8,\n",
    "        'days_since_prior_order': np.float32})\n",
    "\n",
    "products = pd.read_csv('products.csv', dtype={\n",
    "        'product_id': np.int32,\n",
    "        'aisle_id': np.int16,\n",
    "        'department_id': np.int16},\n",
    "        usecols=['product_id', 'aisle_id', 'department_id'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>order_id</th>\n",
       "      <th>product_id</th>\n",
       "      <th>add_to_cart_order</th>\n",
       "      <th>reordered</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2</td>\n",
       "      <td>33120</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>28985</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>9327</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2</td>\n",
       "      <td>45918</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2</td>\n",
       "      <td>30035</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   order_id  product_id  add_to_cart_order  reordered\n",
       "0         2       33120                  1          1\n",
       "1         2       28985                  2          1\n",
       "2         2        9327                  3          0\n",
       "3         2       45918                  4          1\n",
       "4         2       30035                  5          0"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "priors.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>order_id</th>\n",
       "      <th>user_id</th>\n",
       "      <th>eval_set</th>\n",
       "      <th>order_number</th>\n",
       "      <th>order_dow</th>\n",
       "      <th>order_hour_of_day</th>\n",
       "      <th>days_since_prior_order</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2539329</td>\n",
       "      <td>1</td>\n",
       "      <td>prior</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>8</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2398795</td>\n",
       "      <td>1</td>\n",
       "      <td>prior</td>\n",
       "      <td>2</td>\n",
       "      <td>3</td>\n",
       "      <td>7</td>\n",
       "      <td>15.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>473747</td>\n",
       "      <td>1</td>\n",
       "      <td>prior</td>\n",
       "      <td>3</td>\n",
       "      <td>3</td>\n",
       "      <td>12</td>\n",
       "      <td>21.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2254736</td>\n",
       "      <td>1</td>\n",
       "      <td>prior</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>7</td>\n",
       "      <td>29.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>431534</td>\n",
       "      <td>1</td>\n",
       "      <td>prior</td>\n",
       "      <td>5</td>\n",
       "      <td>4</td>\n",
       "      <td>15</td>\n",
       "      <td>28.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   order_id  user_id eval_set  order_number  order_dow  order_hour_of_day  \\\n",
       "0   2539329        1    prior             1          2                  8   \n",
       "1   2398795        1    prior             2          3                  7   \n",
       "2    473747        1    prior             3          3                 12   \n",
       "3   2254736        1    prior             4          4                  7   \n",
       "4    431534        1    prior             5          4                 15   \n",
       "\n",
       "   days_since_prior_order  \n",
       "0                     NaN  \n",
       "1                    15.0  \n",
       "2                    21.0  \n",
       "3                    29.0  \n",
       "4                    28.0  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "orders.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>product_id</th>\n",
       "      <th>aisle_id</th>\n",
       "      <th>department_id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>61</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>104</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>94</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>38</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>5</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   product_id  aisle_id  department_id\n",
       "0           1        61             19\n",
       "1           2       104             13\n",
       "2           3        94              7\n",
       "3           4        38              1\n",
       "4           5         5             13"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "products.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Product features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>product_id</th>\n",
       "      <th>aisle_id</th>\n",
       "      <th>department_id</th>\n",
       "      <th>orders</th>\n",
       "      <th>reorders</th>\n",
       "      <th>reorder_rate</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>product_id</th>\n",
       "      <th></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>1</th>\n",
       "      <td>1</td>\n",
       "      <td>61</td>\n",
       "      <td>19</td>\n",
       "      <td>1852.0</td>\n",
       "      <td>1136.0</td>\n",
       "      <td>0.613391</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>104</td>\n",
       "      <td>13</td>\n",
       "      <td>90.0</td>\n",
       "      <td>12.0</td>\n",
       "      <td>0.133333</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>94</td>\n",
       "      <td>7</td>\n",
       "      <td>277.0</td>\n",
       "      <td>203.0</td>\n",
       "      <td>0.732852</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>38</td>\n",
       "      <td>1</td>\n",
       "      <td>329.0</td>\n",
       "      <td>147.0</td>\n",
       "      <td>0.446809</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5</td>\n",
       "      <td>5</td>\n",
       "      <td>13</td>\n",
       "      <td>15.0</td>\n",
       "      <td>9.0</td>\n",
       "      <td>0.600000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            product_id  aisle_id  department_id  orders  reorders  \\\n",
       "product_id                                                          \n",
       "1                    1        61             19  1852.0    1136.0   \n",
       "2                    2       104             13    90.0      12.0   \n",
       "3                    3        94              7   277.0     203.0   \n",
       "4                    4        38              1   329.0     147.0   \n",
       "5                    5         5             13    15.0       9.0   \n",
       "\n",
       "            reorder_rate  \n",
       "product_id                \n",
       "1               0.613391  \n",
       "2               0.133333  \n",
       "3               0.732852  \n",
       "4               0.446809  \n",
       "5               0.600000  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# product features\n",
    "p = pd.DataFrame()\n",
    "p['orders'] = priors.groupby('product_id').size().astype(np.int32)\n",
    "p['reorders'] = priors['reordered'].groupby(priors.product_id).sum().astype(np.float32)\n",
    "p['reorder_rate'] = (p.reorders/p.orders).astype(np.float32)\n",
    "\n",
    "products = products.join(p, on='product_id')\n",
    "products.set_index('product_id', drop=False, inplace=True)\n",
    "\n",
    "products.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "del p"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:4: FutureWarning: 'order_id' is both an index level and a column label.\n",
      "Defaulting to column, but this will raise an ambiguity error in a future version\n",
      "  after removing the cwd from sys.path.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>order_id</th>\n",
       "      <th>product_id</th>\n",
       "      <th>add_to_cart_order</th>\n",
       "      <th>reordered</th>\n",
       "      <th>user_id</th>\n",
       "      <th>eval_set</th>\n",
       "      <th>order_number</th>\n",
       "      <th>order_dow</th>\n",
       "      <th>order_hour_of_day</th>\n",
       "      <th>days_since_prior_order</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2</td>\n",
       "      <td>33120</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>202279</td>\n",
       "      <td>prior</td>\n",
       "      <td>3</td>\n",
       "      <td>5</td>\n",
       "      <td>9</td>\n",
       "      <td>8.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>28985</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>202279</td>\n",
       "      <td>prior</td>\n",
       "      <td>3</td>\n",
       "      <td>5</td>\n",
       "      <td>9</td>\n",
       "      <td>8.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>9327</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>202279</td>\n",
       "      <td>prior</td>\n",
       "      <td>3</td>\n",
       "      <td>5</td>\n",
       "      <td>9</td>\n",
       "      <td>8.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2</td>\n",
       "      <td>45918</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>202279</td>\n",
       "      <td>prior</td>\n",
       "      <td>3</td>\n",
       "      <td>5</td>\n",
       "      <td>9</td>\n",
       "      <td>8.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2</td>\n",
       "      <td>30035</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>202279</td>\n",
       "      <td>prior</td>\n",
       "      <td>3</td>\n",
       "      <td>5</td>\n",
       "      <td>9</td>\n",
       "      <td>8.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   order_id  product_id  add_to_cart_order  reordered  user_id eval_set  \\\n",
       "0         2       33120                  1          1   202279    prior   \n",
       "1         2       28985                  2          1   202279    prior   \n",
       "2         2        9327                  3          0   202279    prior   \n",
       "3         2       45918                  4          1   202279    prior   \n",
       "4         2       30035                  5          0   202279    prior   \n",
       "\n",
       "   order_number  order_dow  order_hour_of_day  days_since_prior_order  \n",
       "0             3          5                  9                     8.0  \n",
       "1             3          5                  9                     8.0  \n",
       "2             3          5                  9                     8.0  \n",
       "3             3          5                  9                     8.0  \n",
       "4             3          5                  9                     8.0  "
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# add info to priors\n",
    "orders_pr = orders[orders['eval_set'] == 'prior']\n",
    "orders_pr.set_index('order_id', inplace=True, drop=False)\n",
    "priors = priors.merge(orders_pr, on='order_id', how='left')\n",
    "\n",
    "priors.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### User features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>total_items</th>\n",
       "      <th>all_products</th>\n",
       "      <th>distinct_items</th>\n",
       "      <th>total_orders</th>\n",
       "      <th>avg_days_btw_orders</th>\n",
       "      <th>median_days_btw_orders</th>\n",
       "      <th>avg_basket</th>\n",
       "      <th>avg_distinct_basket</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>user_id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></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>1</th>\n",
       "      <td>59</td>\n",
       "      <td>{17122, 196, 26405, 46149, 14084, 13032, 26088...</td>\n",
       "      <td>18</td>\n",
       "      <td>10</td>\n",
       "      <td>19.555555</td>\n",
       "      <td>20.0</td>\n",
       "      <td>5.900000</td>\n",
       "      <td>1.800000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>195</td>\n",
       "      <td>{45066, 2573, 18961, 23, 32792, 1559, 22559, 1...</td>\n",
       "      <td>102</td>\n",
       "      <td>14</td>\n",
       "      <td>15.230769</td>\n",
       "      <td>13.0</td>\n",
       "      <td>13.928572</td>\n",
       "      <td>7.285714</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>88</td>\n",
       "      <td>{17668, 44683, 48523, 21903, 14992, 21137, 324...</td>\n",
       "      <td>33</td>\n",
       "      <td>12</td>\n",
       "      <td>12.090909</td>\n",
       "      <td>11.0</td>\n",
       "      <td>7.333333</td>\n",
       "      <td>2.750000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>18</td>\n",
       "      <td>{21573, 42329, 17769, 35469, 37646, 1200, 1905...</td>\n",
       "      <td>17</td>\n",
       "      <td>5</td>\n",
       "      <td>13.750000</td>\n",
       "      <td>17.0</td>\n",
       "      <td>3.600000</td>\n",
       "      <td>3.400000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>37</td>\n",
       "      <td>{11777, 40706, 28289, 48775, 20754, 6808, 1398...</td>\n",
       "      <td>23</td>\n",
       "      <td>4</td>\n",
       "      <td>13.333333</td>\n",
       "      <td>11.0</td>\n",
       "      <td>9.250000</td>\n",
       "      <td>5.750000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         total_items                                       all_products  \\\n",
       "user_id                                                                   \n",
       "1                 59  {17122, 196, 26405, 46149, 14084, 13032, 26088...   \n",
       "2                195  {45066, 2573, 18961, 23, 32792, 1559, 22559, 1...   \n",
       "3                 88  {17668, 44683, 48523, 21903, 14992, 21137, 324...   \n",
       "4                 18  {21573, 42329, 17769, 35469, 37646, 1200, 1905...   \n",
       "5                 37  {11777, 40706, 28289, 48775, 20754, 6808, 1398...   \n",
       "\n",
       "         distinct_items  total_orders  avg_days_btw_orders  \\\n",
       "user_id                                                      \n",
       "1                    18            10            19.555555   \n",
       "2                   102            14            15.230769   \n",
       "3                    33            12            12.090909   \n",
       "4                    17             5            13.750000   \n",
       "5                    23             4            13.333333   \n",
       "\n",
       "         median_days_btw_orders  avg_basket  avg_distinct_basket  \n",
       "user_id                                                           \n",
       "1                          20.0    5.900000             1.800000  \n",
       "2                          13.0   13.928572             7.285714  \n",
       "3                          11.0    7.333333             2.750000  \n",
       "4                          17.0    3.600000             3.400000  \n",
       "5                          11.0    9.250000             5.750000  "
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# user features\n",
    "u = pd.DataFrame()\n",
    "u['total_orders'] = orders_pr.groupby('user_id').size().astype(np.int16)\n",
    "u['avg_days_btw_orders'] = orders_pr.groupby('user_id')['days_since_prior_order'].mean().astype(np.float32)\n",
    "u['median_days_btw_orders'] = orders_pr.groupby('user_id')['days_since_prior_order'].median().astype(np.float32)\n",
    "\n",
    "u2 = pd.DataFrame()\n",
    "u2['total_items'] = priors.groupby('user_id').size().astype(np.int16)\n",
    "u2['all_products'] = priors.groupby('user_id')['product_id'].apply(set)\n",
    "u2['distinct_items'] = (u2.all_products.map(len)).astype(np.int16)\n",
    "\n",
    "users = u2.join(u)\n",
    "\n",
    "users['avg_basket'] = (users.total_items/users.total_orders).astype(np.float32)\n",
    "users['avg_distinct_basket'] = (users.distinct_items/users.total_orders).astype(np.float32)\n",
    "\n",
    "del u, u2\n",
    "users.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### UserXProduct features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>nb_orders</th>\n",
       "      <th>last_order_id</th>\n",
       "      <th>sum_pos_in_cart</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>20227933120</th>\n",
       "      <td>5</td>\n",
       "      <td>104690</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20227928985</th>\n",
       "      <td>5</td>\n",
       "      <td>132412</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20227909327</th>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20227945918</th>\n",
       "      <td>5</td>\n",
       "      <td>2382766</td>\n",
       "      <td>24</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20227930035</th>\n",
       "      <td>3</td>\n",
       "      <td>2382766</td>\n",
       "      <td>14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20227917794</th>\n",
       "      <td>7</td>\n",
       "      <td>2382766</td>\n",
       "      <td>25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20227940141</th>\n",
       "      <td>5</td>\n",
       "      <td>132412</td>\n",
       "      <td>29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20227901819</th>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20227943668</th>\n",
       "      <td>3</td>\n",
       "      <td>132412</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20597033754</th>\n",
       "      <td>17</td>\n",
       "      <td>368699</td>\n",
       "      <td>86</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             nb_orders  last_order_id  sum_pos_in_cart\n",
       "20227933120          5         104690               10\n",
       "20227928985          5         132412               16\n",
       "20227909327          1              2                3\n",
       "20227945918          5        2382766               24\n",
       "20227930035          3        2382766               14\n",
       "20227917794          7        2382766               25\n",
       "20227940141          5         132412               29\n",
       "20227901819          2              2               19\n",
       "20227943668          3         132412               20\n",
       "20597033754         17         368699               86"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# userXproduct features\n",
    "priors['user_product'] = priors.product_id + priors.user_id * 100000\n",
    "d = {}\n",
    "\n",
    "for row in priors.itertuples():\n",
    "    z = row.user_product\n",
    "    if z not in d:\n",
    "        d[z] = (1, \n",
    "                (row.order_number, row.order_id), \n",
    "                row.add_to_cart_order)\n",
    "    else:\n",
    "        d[z] = (d[z][0]+1,\n",
    "                max(d[z][1], (row.order_number, row.order_id)), \n",
    "                d[z][2]+row.add_to_cart_order)\n",
    "\n",
    "up = pd.DataFrame.from_dict(d, orient='index')\n",
    "up.columns = ['nb_orders', 'last_order_id', 'sum_pos_in_cart']\n",
    "\n",
    "del d\n",
    "\n",
    "up.nb_orders = up.nb_orders.astype(np.int16)\n",
    "up.last_order_id = up.last_order_id.map(lambda x: x[1]).astype(np.int32)\n",
    "up.sum_pos_in_cart = up.sum_pos_in_cart.astype(np.int16)\n",
    "\n",
    "up.head(10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Features based on recent 5 orders"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>order_id</th>\n",
       "      <th>product_id</th>\n",
       "      <th>add_to_cart_order</th>\n",
       "      <th>reordered</th>\n",
       "      <th>user_id</th>\n",
       "      <th>eval_set</th>\n",
       "      <th>order_number</th>\n",
       "      <th>order_dow</th>\n",
       "      <th>order_hour_of_day</th>\n",
       "      <th>days_since_prior_order</th>\n",
       "      <th>user_product</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>6</td>\n",
       "      <td>40462</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>22352</td>\n",
       "      <td>prior</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>12</td>\n",
       "      <td>30.0</td>\n",
       "      <td>2235240462</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6</td>\n",
       "      <td>15873</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>22352</td>\n",
       "      <td>prior</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>12</td>\n",
       "      <td>30.0</td>\n",
       "      <td>2235215873</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>6</td>\n",
       "      <td>41897</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>22352</td>\n",
       "      <td>prior</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>12</td>\n",
       "      <td>30.0</td>\n",
       "      <td>2235241897</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>7</td>\n",
       "      <td>34050</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>142903</td>\n",
       "      <td>prior</td>\n",
       "      <td>11</td>\n",
       "      <td>2</td>\n",
       "      <td>14</td>\n",
       "      <td>30.0</td>\n",
       "      <td>14290334050</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>7</td>\n",
       "      <td>46802</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>142903</td>\n",
       "      <td>prior</td>\n",
       "      <td>11</td>\n",
       "      <td>2</td>\n",
       "      <td>14</td>\n",
       "      <td>30.0</td>\n",
       "      <td>14290346802</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>12</td>\n",
       "      <td>30597</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>152610</td>\n",
       "      <td>prior</td>\n",
       "      <td>22</td>\n",
       "      <td>6</td>\n",
       "      <td>8</td>\n",
       "      <td>10.0</td>\n",
       "      <td>15261030597</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>12</td>\n",
       "      <td>15221</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>152610</td>\n",
       "      <td>prior</td>\n",
       "      <td>22</td>\n",
       "      <td>6</td>\n",
       "      <td>8</td>\n",
       "      <td>10.0</td>\n",
       "      <td>15261015221</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>12</td>\n",
       "      <td>43772</td>\n",
       "      <td>3</td>\n",
       "      <td>1</td>\n",
       "      <td>152610</td>\n",
       "      <td>prior</td>\n",
       "      <td>22</td>\n",
       "      <td>6</td>\n",
       "      <td>8</td>\n",
       "      <td>10.0</td>\n",
       "      <td>15261043772</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>12</td>\n",
       "      <td>37886</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>152610</td>\n",
       "      <td>prior</td>\n",
       "      <td>22</td>\n",
       "      <td>6</td>\n",
       "      <td>8</td>\n",
       "      <td>10.0</td>\n",
       "      <td>15261037886</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>12</td>\n",
       "      <td>37215</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>152610</td>\n",
       "      <td>prior</td>\n",
       "      <td>22</td>\n",
       "      <td>6</td>\n",
       "      <td>8</td>\n",
       "      <td>10.0</td>\n",
       "      <td>15261037215</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   order_id  product_id  add_to_cart_order  reordered  user_id eval_set  \\\n",
       "0         6       40462                  1          0    22352    prior   \n",
       "1         6       15873                  2          0    22352    prior   \n",
       "2         6       41897                  3          0    22352    prior   \n",
       "3         7       34050                  1          0   142903    prior   \n",
       "4         7       46802                  2          0   142903    prior   \n",
       "5        12       30597                  1          1   152610    prior   \n",
       "6        12       15221                  2          1   152610    prior   \n",
       "7        12       43772                  3          1   152610    prior   \n",
       "8        12       37886                  4          1   152610    prior   \n",
       "9        12       37215                  5          0   152610    prior   \n",
       "\n",
       "   order_number  order_dow  order_hour_of_day  days_since_prior_order  \\\n",
       "0             4          1                 12                    30.0   \n",
       "1             4          1                 12                    30.0   \n",
       "2             4          1                 12                    30.0   \n",
       "3            11          2                 14                    30.0   \n",
       "4            11          2                 14                    30.0   \n",
       "5            22          6                  8                    10.0   \n",
       "6            22          6                  8                    10.0   \n",
       "7            22          6                  8                    10.0   \n",
       "8            22          6                  8                    10.0   \n",
       "9            22          6                  8                    10.0   \n",
       "\n",
       "   user_product  \n",
       "0    2235240462  \n",
       "1    2235215873  \n",
       "2    2235241897  \n",
       "3   14290334050  \n",
       "4   14290346802  \n",
       "5   15261030597  \n",
       "6   15261015221  \n",
       "7   15261043772  \n",
       "8   15261037886  \n",
       "9   15261037215  "
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# take recent 5 orders of each user\n",
    "orders_n5 = orders_pr.groupby('user_id').order_number.nlargest(5).to_frame().reset_index()\n",
    "orders_n5.head(10)\n",
    "\n",
    "priors_n5 = priors.merge(orders_n5, \n",
    "                         on=['order_id','user_id','order_number'], how='inner')\n",
    "priors_n5.head(10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### User features of nearest 5 orders"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:3: FutureWarning: 'order_id' is both an index level and a column label.\n",
      "Defaulting to column, but this will raise an ambiguity error in a future version\n",
      "  This is separate from the ipykernel package so we can avoid doing imports until\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>total_items</th>\n",
       "      <th>distinct_items</th>\n",
       "      <th>total_orders</th>\n",
       "      <th>avg_days_btw_orders</th>\n",
       "      <th>median_days_btw_orders</th>\n",
       "      <th>avg_basket</th>\n",
       "      <th>avg_distinct_basket</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>user_id</th>\n",
       "      <th></th>\n",
       "      <th></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>1</th>\n",
       "      <td>30</td>\n",
       "      <td>10</td>\n",
       "      <td>5</td>\n",
       "      <td>16.600000</td>\n",
       "      <td>19.0</td>\n",
       "      <td>6.00</td>\n",
       "      <td>2.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>68</td>\n",
       "      <td>46</td>\n",
       "      <td>5</td>\n",
       "      <td>21.400000</td>\n",
       "      <td>28.0</td>\n",
       "      <td>13.60</td>\n",
       "      <td>9.20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>30</td>\n",
       "      <td>14</td>\n",
       "      <td>5</td>\n",
       "      <td>11.400000</td>\n",
       "      <td>11.0</td>\n",
       "      <td>6.00</td>\n",
       "      <td>2.80</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>18</td>\n",
       "      <td>17</td>\n",
       "      <td>5</td>\n",
       "      <td>13.750000</td>\n",
       "      <td>17.0</td>\n",
       "      <td>3.60</td>\n",
       "      <td>3.40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>37</td>\n",
       "      <td>23</td>\n",
       "      <td>4</td>\n",
       "      <td>13.333333</td>\n",
       "      <td>11.0</td>\n",
       "      <td>9.25</td>\n",
       "      <td>5.75</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         total_items  distinct_items  total_orders  avg_days_btw_orders  \\\n",
       "user_id                                                                   \n",
       "1                 30              10             5            16.600000   \n",
       "2                 68              46             5            21.400000   \n",
       "3                 30              14             5            11.400000   \n",
       "4                 18              17             5            13.750000   \n",
       "5                 37              23             4            13.333333   \n",
       "\n",
       "         median_days_btw_orders  avg_basket  avg_distinct_basket  \n",
       "user_id                                                           \n",
       "1                          19.0        6.00                 2.00  \n",
       "2                          28.0       13.60                 9.20  \n",
       "3                          11.0        6.00                 2.80  \n",
       "4                          17.0        3.60                 3.40  \n",
       "5                          11.0        9.25                 5.75  "
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# user features of nearest 5 orders\n",
    "orders_n5 = orders_n5.merge(orders_pr, \n",
    "                            on=['order_id','user_id','order_number'], how='inner')\n",
    "u = pd.DataFrame()\n",
    "u['total_orders'] = orders_n5.groupby('user_id').size().astype(np.int16)\n",
    "u['avg_days_btw_orders'] = orders_n5.groupby('user_id')['days_since_prior_order'].mean().astype(np.float32)\n",
    "u['median_days_btw_orders'] = orders_n5.groupby('user_id')['days_since_prior_order'].median().astype(np.float32)\n",
    "\n",
    "u2 = pd.DataFrame()\n",
    "u2['total_items'] = priors_n5.groupby('user_id').size().astype(np.int16)\n",
    "u2['all_products'] = priors_n5.groupby('user_id')['product_id'].apply(set)\n",
    "u2['distinct_items'] = (u2.all_products.map(len)).astype(np.int16)\n",
    "del u2['all_products']\n",
    "\n",
    "users_n5 = u2.join(u)\n",
    "users_n5['avg_basket'] = (users_n5.total_items/users_n5.total_orders).astype(np.float32)\n",
    "users_n5['avg_distinct_basket'] = (users_n5.distinct_items/users_n5.total_orders).astype(np.float32)\n",
    "\n",
    "del u, u2\n",
    "users_n5.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### UserXProduct features of nearest 5 orders"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>nb_orders</th>\n",
       "      <th>sum_pos_in_cart</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2235240462</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2235215873</th>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2235241897</th>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14290334050</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14290346802</th>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15261030597</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15261015221</th>\n",
       "      <td>2</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15261043772</th>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15261037886</th>\n",
       "      <td>1</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15261037215</th>\n",
       "      <td>1</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             nb_orders  sum_pos_in_cart\n",
       "2235240462           1                1\n",
       "2235215873           1                2\n",
       "2235241897           1                3\n",
       "14290334050          1                1\n",
       "14290346802          1                2\n",
       "15261030597          1                1\n",
       "15261015221          2                5\n",
       "15261043772          1                3\n",
       "15261037886          1                4\n",
       "15261037215          1                5"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# userXproduct features of nearest 5 orders\n",
    "d = {}\n",
    "\n",
    "for row in priors_n5.itertuples():\n",
    "    z = row.user_product\n",
    "    if z not in d:\n",
    "        d[z] = (1, \n",
    "                row.add_to_cart_order)\n",
    "    else:\n",
    "        d[z] = (d[z][0]+1,\n",
    "                d[z][1]+row.add_to_cart_order)\n",
    "\n",
    "up_n5 = pd.DataFrame.from_dict(d, orient='index')\n",
    "up_n5.columns = ['nb_orders', 'sum_pos_in_cart']\n",
    "\n",
    "del d\n",
    "\n",
    "up_n5.nb_orders = up_n5.nb_orders.astype(np.int16)\n",
    "up_n5.sum_pos_in_cart = up_n5.sum_pos_in_cart.astype(np.int16)\n",
    "\n",
    "up_n5.head(10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature engineering"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# train test orders\n",
    "train_orders = orders[orders.eval_set == 'train']\n",
    "test_orders = orders[orders.eval_set == 'test']\n",
    "\n",
    "train.set_index(['order_id', 'product_id'], inplace=True, drop=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def features(selected_orders, labels_given=False):\n",
    "    print('build candidate list')\n",
    "    order_list = []\n",
    "    product_list = []\n",
    "    labels = []\n",
    "    i=0\n",
    "    for row in selected_orders.itertuples():\n",
    "        i+=1\n",
    "        if i%10000 == 0: print('order row',i)\n",
    "        order_id = row.order_id\n",
    "        user_id = row.user_id\n",
    "        user_products = users.all_products[user_id]\n",
    "        product_list += user_products\n",
    "        order_list += [order_id] * len(user_products)\n",
    "        if labels_given:\n",
    "            labels += [(order_id, product) in train.index for product in user_products]\n",
    "        \n",
    "    df = pd.DataFrame({'order_id':order_list, 'product_id':product_list}, dtype=np.int32)\n",
    "    labels = np.array(labels, dtype=np.int8)\n",
    "    del order_list\n",
    "    del product_list\n",
    "    \n",
    "    print('user related features')\n",
    "    df['user_id'] = df.order_id.map(orders.user_id)\n",
    "    df['user_total_orders'] = df.user_id.map(users.total_orders)\n",
    "    df['user_total_items'] = df.user_id.map(users.total_items)\n",
    "    df['user_distinct_items'] = df.user_id.map(users.distinct_items)\n",
    "    df['user_avg_days_btw_orders'] = df.user_id.map(users.avg_days_btw_orders)\n",
    "    df['user_median_days_btw_orders'] = df.user_id.map(users.median_days_btw_orders)\n",
    "    df['user_avg_basket'] =  df.user_id.map(users.avg_basket)\n",
    "    \n",
    "    df['user_total_orders_n5'] = df.user_id.map(users_n5.total_orders)\n",
    "    df['user_total_items_n5'] = df.user_id.map(users_n5.total_items)\n",
    "    df['user_distinct_items_n5'] = df.user_id.map(users_n5.distinct_items)\n",
    "    df['user_avg_days_btw_orders_n5'] = df.user_id.map(users_n5.avg_days_btw_orders)\n",
    "    df['user_median_days_btw_orders_n5'] = df.user_id.map(users_n5.median_days_btw_orders)\n",
    "    df['user_avg_basket_n5'] =  df.user_id.map(users_n5.avg_basket)\n",
    "\n",
    "    print('order related features')\n",
    "    df['order_hour_of_day'] = df.order_id.map(orders.order_hour_of_day)\n",
    "    df['days_since_prior_order'] = df.order_id.map(orders.days_since_prior_order)\n",
    "    df['days_since_ratio'] = df.days_since_prior_order / df.user_avg_days_btw_orders\n",
    "    \n",
    "    df['days_since_ratio_n5'] = df.days_since_prior_order / df.user_avg_days_btw_orders_n5\n",
    "    \n",
    "    print('product related features')\n",
    "    df['aisle_id'] = df.product_id.map(products.aisle_id)\n",
    "    df['department_id'] = df.product_id.map(products.department_id)\n",
    "    df['product_orders'] = df.product_id.map(products.orders).astype(np.int32)\n",
    "    df['product_reorders'] = df.product_id.map(products.reorders)\n",
    "    df['product_reorder_rate'] = df.product_id.map(products.reorder_rate)\n",
    "\n",
    "    print('user_X_product related features')\n",
    "    df['z'] = df.user_id * 100000 + df.product_id\n",
    "    df.drop(['user_id'], axis=1, inplace=True)\n",
    "    df['UP_orders'] = df.z.map(up.nb_orders)\n",
    "    df['UP_last_order_id'] = df.z.map(up.last_order_id)\n",
    "    df['UP_avg_pos_in_cart'] = (df.z.map(up.sum_pos_in_cart) / df.UP_orders).astype(np.float32)\n",
    "    df['UP_reorder_rate'] = (df.UP_orders / df.user_total_orders).astype(np.float32)\n",
    "    df['UP_orders_since_last'] = df.user_total_orders - df.UP_last_order_id.map(orders.order_number)\n",
    "#    df['UP_delta_hour_vs_last'] = abs(df.order_hour_of_day - df.UP_last_order_id.map(orders.order_hour_of_day)).map(lambda x: min(x, 24-x)).astype(np.int8)\n",
    "    \n",
    "    df['UP_orders_n5'] = df.z.map(up_n5.nb_orders)\n",
    "    df['UP_avg_pos_in_cart_n5'] = (df.z.map(up_n5.sum_pos_in_cart) / df.UP_orders_n5).astype(np.float32)\n",
    "    df['UP_reorder_rate_n5'] = (df.UP_orders_n5 / df.user_total_orders_n5).astype(np.float32)\n",
    "    df.drop(['UP_last_order_id', 'z'], axis=1, inplace=True)\n",
    "    \n",
    "    print(df.dtypes)\n",
    "    print(df.memory_usage())\n",
    "    return (df, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, labels = features(train_orders, labels_given=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_train = pd.DataFrame(labels, columns=['label'])\n",
    "train = pd.concat([X_train, y_train], axis=1)\n",
    "train.to_pickle('train.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_test, _ = features(test_orders)\n",
    "\n",
    "X_test.to_pickle('test.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Overall behaviors vs recent 5 orders \n",
    "- Users' recent shopping behaviors (recent 5 orders) are correlated with their overall shopping behavior (0.7-0.9), but not exactly same.   \n",
    "- Especially with regard to product reorder rate, correlation between UP_reorder_rate and UP_reorder_rate_n5 is 0.7, lowest of all."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "UP_reorder_rate UP_reorder_rate_n5 0.7084104047544123\n",
      "user_avg_days_btw_orders user_avg_days_btw_orders_n5 0.8991115338270881\n",
      "user_avg_basket user_avg_basket_n5 0.9078683036790905\n",
      "days_since_ratio days_since_ratio_n5 0.8209298473885959\n",
      "UP_avg_pos_in_cart UP_avg_pos_in_cart_n5 0.8650148648311861\n"
     ]
    }
   ],
   "source": [
    "# all orders vs nearest 5 orders\n",
    "print('UP_reorder_rate', 'UP_reorder_rate_n5', corrmat['UP_reorder_rate']['UP_reorder_rate_n5'])\n",
    "print('user_avg_days_btw_orders', 'user_avg_days_btw_orders_n5', corrmat['user_avg_days_btw_orders']['user_avg_days_btw_orders_n5'])\n",
    "print('user_avg_basket', 'user_avg_basket_n5', corrmat['user_avg_basket']['user_avg_basket_n5'])\n",
    "print('days_since_ratio', 'days_since_ratio_n5', corrmat['days_since_ratio']['days_since_ratio_n5'])\n",
    "print('UP_avg_pos_in_cart', 'UP_avg_pos_in_cart_n5', corrmat['UP_avg_pos_in_cart']['UP_avg_pos_in_cart_n5'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Modeling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "x_tr, x_val, y_tr, y_val = train_test_split(X_train, labels, \n",
    "                                            test_size=0.20, random_state=23)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/lightgbm/basic.py:1042: UserWarning: categorical_feature in Dataset is overridden. New categorical_feature is ['aisle_id', 'department_id']\n",
      "  warnings.warn('categorical_feature in Dataset is overridden. New categorical_feature is {}'.format(sorted(list(categorical_feature))))\n",
      "/anaconda3/lib/python3.6/site-packages/lightgbm/basic.py:685: UserWarning: categorical_feature in param dict is overridden.\n",
      "  warnings.warn('categorical_feature in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[20]\tvalid_0's binary_logloss: 0.302669\n",
      "[40]\tvalid_0's binary_logloss: 0.302107\n",
      "[60]\tvalid_0's binary_logloss: 0.301872\n",
      "[80]\tvalid_0's binary_logloss: 0.301678\n",
      "[100]\tvalid_0's binary_logloss: 0.301521\n"
     ]
    }
   ],
   "source": [
    "feat = ['user_total_orders', 'user_total_items','user_distinct_items', \n",
    "        'user_avg_days_btw_orders','user_avg_basket', \n",
    "        'user_total_orders_n5', 'user_distinct_items_n5', \n",
    "        'user_avg_days_btw_orders_n5','user_avg_basket_n5', \n",
    "        'order_hour_of_day','days_since_prior_order', 'days_since_ratio', \n",
    "        'days_since_ratio_n5', 'aisle_id', 'department_id',\n",
    "        'product_orders', 'product_reorders', 'product_reorder_rate',\n",
    "        'UP_orders', 'UP_avg_pos_in_cart', 'UP_reorder_rate', 'UP_orders_since_last', \n",
    "        'UP_orders_n5', 'UP_reorder_rate_n5', 'UP_avg_pos_in_cart_n5']\n",
    "\n",
    "lgb12_params = {\n",
    "    'task': 'train',\n",
    "    'boosting_type': 'gbdt',\n",
    "    'objective': 'binary',\n",
    "    'metric': {'binary_logloss'},\n",
    "    'num_leaves': 112,\n",
    "    'max_depth': 10,\n",
    "    'feature_fraction': 0.9,\n",
    "    'bagging_fraction': 0.95,\n",
    "    'bagging_freq': 5}\n",
    "\n",
    "lgbtrain = lgb.Dataset(x_tr[feat], y_tr)\n",
    "lgbvalid = lgb.Dataset(x_val[feat], y_val)\n",
    "\n",
    "ROUNDS = 100\n",
    "#early_stopping = 100\n",
    "lgb12 = lgb.train(lgb12_params, \n",
    "                lgbtrain,\n",
    "                num_boost_round=ROUNDS,\n",
    "                valid_sets=[lgbvalid],\n",
    "                categorical_feature=['aisle_id', 'department_id'],\n",
    "#                early_stopping_rounds=early_stopping,\n",
    "                verbose_eval=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 203,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x1a80f09fd0>"
      ]
     },
     "execution_count": 203,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1b8ac289b0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "lgb.plot_importance(lgb12, figsize=(9,15), height=0.4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 199,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_preds['lgb12_test'] = lgb12.predict(X_test)\n",
    "\n",
    "THRESHOLD = 0.2\n",
    "\n",
    "d = {}\n",
    "for row in test_preds.itertuples():\n",
    "    if row.lgb12_test >= THRESHOLD:\n",
    "        try:\n",
    "            d[row.order_id] += ' ' + str(row.product_id)\n",
    "        except:\n",
    "            d[row.order_id] = str(row.product_id)\n",
    "        \n",
    "\n",
    "for order in test_orders.order_id:\n",
    "    if order not in d:\n",
    "        d[order] = 'None'   \n",
    "\n",
    "sub = pd.DataFrame.from_dict(d, orient='index')\n",
    "\n",
    "sub.reset_index(inplace=True)\n",
    "sub.columns = ['order_id', 'products']\n",
    "sub.to_csv('sub_lgb12.csv', index=False)\n",
    "\n",
    "# F1 score: 0.1900706"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## F1 score optimization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 137,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Author: @Faron\n",
    "\n",
    "class F1Optimizer():\n",
    "    def __init__(self):\n",
    "        pass\n",
    "\n",
    "    @staticmethod\n",
    "    def get_expectations(P, pNone=None):\n",
    "        expectations = []\n",
    "#        P = np.sort(P)[::-1]\n",
    "\n",
    "        n = np.array(P).shape[0]\n",
    "        DP_C = np.zeros((n + 2, n + 1))\n",
    "        if pNone is None:\n",
    "            pNone = (1.0 - P).prod()\n",
    "\n",
    "        DP_C[0][0] = 1.0\n",
    "        for j in range(1, n):\n",
    "            DP_C[0][j] = (1.0 - P[j - 1]) * DP_C[0, j - 1]\n",
    "\n",
    "        for i in range(1, n + 1):\n",
    "            DP_C[i, i] = DP_C[i - 1, i - 1] * P[i - 1]\n",
    "            for j in range(i + 1, n + 1):\n",
    "                DP_C[i, j] = P[j - 1] * DP_C[i - 1, j - 1] + (1.0 - P[j - 1]) * DP_C[i, j - 1]\n",
    "\n",
    "        DP_S = np.zeros((2 * n + 1,))\n",
    "        DP_SNone = np.zeros((2 * n + 1,))\n",
    "        for i in range(1, 2 * n + 1):\n",
    "            DP_S[i] = 1. / (1. * i)\n",
    "            DP_SNone[i] = 1. / (1. * i + 1)\n",
    "        for k in range(n + 1)[::-1]:\n",
    "            f1 = 0\n",
    "            f1None = 0\n",
    "            for k1 in range(n + 1):\n",
    "                f1 += 2 * k1 * DP_C[k1][k] * DP_S[k + k1]\n",
    "                f1None += 2 * k1 * DP_C[k1][k] * DP_SNone[k + k1]\n",
    "            for i in range(1, 2 * k - 1):\n",
    "                DP_S[i] = (1 - P[k - 1]) * DP_S[i] + P[k - 1] * DP_S[i + 1]\n",
    "                DP_SNone[i] = (1 - P[k - 1]) * DP_SNone[i] + P[k - 1] * DP_SNone[i + 1]\n",
    "            expectations.append([f1None + 2 * pNone / (2 + k), f1])\n",
    "\n",
    "        return np.array(expectations[::-1]).T\n",
    "\n",
    "    @staticmethod\n",
    "    def maximize_expectation(P, pNone=None):\n",
    "        expectations = F1Optimizer.get_expectations(P, pNone)\n",
    "\n",
    "        ix_max = np.unravel_index(expectations.argmax(), expectations.shape)\n",
    "        max_f1 = expectations[ix_max]\n",
    "\n",
    "        predNone = True if ix_max[0] == 0 else False\n",
    "        best_k = ix_max[1]\n",
    "\n",
    "        return best_k, predNone, max_f1\n",
    "\n",
    "    @staticmethod\n",
    "    def _F1(tp, fp, fn):\n",
    "        return 2 * tp / (2 * tp + fp + fn)\n",
    "\n",
    "    @staticmethod\n",
    "    def _Fbeta(tp, fp, fn, beta=1.0):\n",
    "        beta_squared = beta ** 2\n",
    "        return (1.0 + beta_squared) * tp / ((1.0 + beta_squared) * tp + fp + beta_squared * fn)\n",
    "    \n",
    "\n",
    "def get_best_prediction(P, L, pNone=None):\n",
    "#    print(\"Maximize F1-Expectation\")\n",
    "#    print(\"=\" * 23)\n",
    "#    P = np.sort(P)[::-1]\n",
    "    n = P.shape[0]\n",
    "#    L = ['L{}'.format(i + 1) for i in range(n)]\n",
    "\n",
    "    if pNone is None:\n",
    "#        print(\"Estimate p(None|x) as (1-p_1)*(1-p_2)*...*(1-p_n)\")\n",
    "        pNone = (1.0 - P).prod()\n",
    "\n",
    "#    PL = ['p({}|x)={}'.format(l, p) for l, p in zip(L, P)]\n",
    "#    print(\"Posteriors: {} (n={})\".format(PL, n))\n",
    "#    print(\"p(None|x)={}\".format(pNone))\n",
    "\n",
    "    opt = F1Optimizer.maximize_expectation(P, pNone)\n",
    "    best_prediction = []\n",
    "    if opt[1]: best_prediction.append('None')\n",
    "    for l in L[:opt[0]]: best_prediction.append(l)\n",
    "    f1_max = opt[2]\n",
    "    \n",
    "    return best_prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "metadata": {},
   "outputs": [],
   "source": [
    "def optimize_f1score(model_prediction):\n",
    "    tmp = X_test[['order_id', 'product_id']]\n",
    "    tmp['pred'] = model_prediction\n",
    "    tmp = tmp.sort_values(by=['order_id', 'pred'], ascending=False)\n",
    "    tmp['product_id'] = tmp['product_id'].astype(str)\n",
    "    l = tmp.groupby('order_id')['product_id'].apply(list).to_frame().reset_index()\n",
    "    p = tmp.groupby('order_id')['pred'].apply(np.array).to_frame().reset_index()\n",
    "    df = l.merge(p, on='order_id', how='left')\n",
    "\n",
    "    d={}\n",
    "    for row in df.itertuples():\n",
    "        P=row.pred\n",
    "        L=row.product_id\n",
    "        d[row.order_id] = \" \".join(get_best_prediction(P,L))\n",
    "    \n",
    "    sub = pd.DataFrame.from_dict(d, orient='index')\n",
    "    sub.reset_index(inplace=True)\n",
    "    sub.columns = ['order_id', 'products']\n",
    "    return sub"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 202,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:3: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  This is separate from the ipykernel package so we can avoid doing imports until\n"
     ]
    }
   ],
   "source": [
    "sub_lgb12_opt = optimize_f1score(test_preds['lgb12_test'])\n",
    "sub_lgb12_opt.to_csv('sub_lgb12_opt.csv', index=False)\n",
    "\n",
    "# F1 score: 0.2092306"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### To be done\n",
    "- LGBM model tuning\n",
    "- Try more classification models\n",
    "- CV"
   ]
  }
 ],
 "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}