LightGBM_for_region.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 138,
   "id": "28dc652c",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd          \n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from datetime import datetime    \n",
    "from pandas import Series \n",
    "from sklearn.metrics import mean_squared_error\n",
    "from math import sqrt\n",
    "from statsmodels.tsa.seasonal import seasonal_decompose\n",
    "import statsmodels\n",
    "import statsmodels.api as sm\n",
    "from statsmodels.tsa.arima_model import ARIMA\n",
    "%matplotlib inline\n",
    "plt.rcParams['font.sans-serif'] = [\"SimHei\"]\n",
    "plt.rcParams[\"axes.unicode_minus\"] = False\n",
    "import seaborn as sns\n",
    "sns.set(style=\"darkgrid\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "id": "a024efdf",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 一些全局变量 \n",
    "is_need_log = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "id": "d7a77895",
   "metadata": {},
   "outputs": [],
   "source": [
    "sales_data = pd.read_csv(\"./示例数据/order_train2.csv\")\n",
    "test  = pd.read_csv(\"./示例数据/predict_sku1.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "id": "97522bcb",
   "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_date</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>sales_chan_name</th>\n",
       "      <th>item_price</th>\n",
       "      <th>ord_qty</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2015/9/1</td>\n",
       "      <td>104</td>\n",
       "      <td>22069</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>offline</td>\n",
       "      <td>1114.0</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2015/9/1</td>\n",
       "      <td>104</td>\n",
       "      <td>20028</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "      <td>offline</td>\n",
       "      <td>1012.0</td>\n",
       "      <td>12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2015/9/2</td>\n",
       "      <td>104</td>\n",
       "      <td>21183</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>online</td>\n",
       "      <td>428.0</td>\n",
       "      <td>109</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2015/9/2</td>\n",
       "      <td>104</td>\n",
       "      <td>20448</td>\n",
       "      <td>308</td>\n",
       "      <td>404</td>\n",
       "      <td>online</td>\n",
       "      <td>962.0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2015/9/2</td>\n",
       "      <td>104</td>\n",
       "      <td>21565</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>offline</td>\n",
       "      <td>1400.0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  order_date  sales_region_code  item_code  first_cate_code  second_cate_code  \\\n",
       "0   2015/9/1                104      22069              307               403   \n",
       "1   2015/9/1                104      20028              301               405   \n",
       "2   2015/9/2                104      21183              307               403   \n",
       "3   2015/9/2                104      20448              308               404   \n",
       "4   2015/9/2                104      21565              307               403   \n",
       "\n",
       "  sales_chan_name  item_price  ord_qty  \n",
       "0         offline      1114.0       19  \n",
       "1         offline      1012.0       12  \n",
       "2          online       428.0      109  \n",
       "3          online       962.0        3  \n",
       "4         offline      1400.0        3  "
      ]
     },
     "execution_count": 141,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "id": "c9c4a9d0",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 597694 entries, 0 to 597693\n",
      "Data columns (total 8 columns):\n",
      " #   Column             Non-Null Count   Dtype  \n",
      "---  ------             --------------   -----  \n",
      " 0   order_date         597694 non-null  object \n",
      " 1   sales_region_code  597694 non-null  int64  \n",
      " 2   item_code          597694 non-null  int64  \n",
      " 3   first_cate_code    597694 non-null  int64  \n",
      " 4   second_cate_code   597694 non-null  int64  \n",
      " 5   sales_chan_name    597694 non-null  object \n",
      " 6   item_price         597694 non-null  float64\n",
      " 7   ord_qty            597694 non-null  int64  \n",
      "dtypes: float64(1), int64(5), object(2)\n",
      "memory usage: 36.5+ MB\n"
     ]
    }
   ],
   "source": [
    "sales_data.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "id": "89449d5e",
   "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>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>item_price</th>\n",
       "      <th>ord_qty</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>597694.000000</td>\n",
       "      <td>597694.000000</td>\n",
       "      <td>597694.000000</td>\n",
       "      <td>597694.000000</td>\n",
       "      <td>597694.000000</td>\n",
       "      <td>597694.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>102.903949</td>\n",
       "      <td>21008.015983</td>\n",
       "      <td>304.932094</td>\n",
       "      <td>406.457174</td>\n",
       "      <td>1076.241578</td>\n",
       "      <td>91.650505</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>1.504880</td>\n",
       "      <td>590.896347</td>\n",
       "      <td>2.004195</td>\n",
       "      <td>3.107329</td>\n",
       "      <td>1167.511104</td>\n",
       "      <td>199.843336</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>101.000000</td>\n",
       "      <td>20001.000000</td>\n",
       "      <td>301.000000</td>\n",
       "      <td>401.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>102.000000</td>\n",
       "      <td>20518.000000</td>\n",
       "      <td>303.000000</td>\n",
       "      <td>404.000000</td>\n",
       "      <td>598.000000</td>\n",
       "      <td>10.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>103.000000</td>\n",
       "      <td>20976.000000</td>\n",
       "      <td>306.000000</td>\n",
       "      <td>407.000000</td>\n",
       "      <td>883.000000</td>\n",
       "      <td>29.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>105.000000</td>\n",
       "      <td>21521.000000</td>\n",
       "      <td>306.000000</td>\n",
       "      <td>408.000000</td>\n",
       "      <td>1291.000000</td>\n",
       "      <td>101.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>105.000000</td>\n",
       "      <td>22084.000000</td>\n",
       "      <td>308.000000</td>\n",
       "      <td>412.000000</td>\n",
       "      <td>260014.000000</td>\n",
       "      <td>16308.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       sales_region_code      item_code  first_cate_code  second_cate_code  \\\n",
       "count      597694.000000  597694.000000    597694.000000     597694.000000   \n",
       "mean          102.903949   21008.015983       304.932094        406.457174   \n",
       "std             1.504880     590.896347         2.004195          3.107329   \n",
       "min           101.000000   20001.000000       301.000000        401.000000   \n",
       "25%           102.000000   20518.000000       303.000000        404.000000   \n",
       "50%           103.000000   20976.000000       306.000000        407.000000   \n",
       "75%           105.000000   21521.000000       306.000000        408.000000   \n",
       "max           105.000000   22084.000000       308.000000        412.000000   \n",
       "\n",
       "          item_price        ord_qty  \n",
       "count  597694.000000  597694.000000  \n",
       "mean     1076.241578      91.650505  \n",
       "std      1167.511104     199.843336  \n",
       "min         1.000000       1.000000  \n",
       "25%       598.000000      10.000000  \n",
       "50%       883.000000      29.000000  \n",
       "75%      1291.000000     101.000000  \n",
       "max    260014.000000   16308.000000  "
      ]
     },
     "execution_count": 143,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_data.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c7bfdd6d",
   "metadata": {},
   "source": [
    "# 1. Data cleaning"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d71fa102",
   "metadata": {},
   "source": [
    "## Remove outliers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "id": "1423d1df",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot: xlabel='item_price'>"
      ]
     },
     "execution_count": 144,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 销售量\n",
    "plt.figure(figsize=(10,4))\n",
    "plt.xlim(-100, 20000)\n",
    "flierprops = dict(marker='o', markerfacecolor='purple', markersize=6,\n",
    "                  linestyle='none', markeredgecolor='black')\n",
    "sns.boxplot(x=sales_data.ord_qty, flierprops=flierprops)\n",
    "\n",
    "# 价格\n",
    "plt.figure(figsize=(10,4))\n",
    "plt.xlim(sales_data.item_price.min(), sales_data.item_price.max()*1.1)\n",
    "sns.boxplot(x=sales_data.item_price, flierprops=flierprops)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "id": "f3ba4beb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "260014.0"
      ]
     },
     "execution_count": 145,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_data.item_price.max()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c0b7cdb7",
   "metadata": {},
   "source": [
    "- 去除销量大于4000的以及价格大于30000的\n",
    "    - 其实这里还需要验证一下去除的那几条记录，是不是好的，因为可能有些商品就是卖这么多或者怎样\n",
    "    - 验证一下test集里有没有出现"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 146,
   "id": "2b74a57d",
   "metadata": {},
   "outputs": [],
   "source": [
    "sales_data = sales_data[(sales_data.item_price < 30000 )& (sales_data.ord_qty < 6000)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "id": "1ed7582e",
   "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_date</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>sales_chan_name</th>\n",
       "      <th>item_price</th>\n",
       "      <th>ord_qty</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2015/9/1</td>\n",
       "      <td>104</td>\n",
       "      <td>22069</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>offline</td>\n",
       "      <td>1114.0</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2015/9/1</td>\n",
       "      <td>104</td>\n",
       "      <td>20028</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "      <td>offline</td>\n",
       "      <td>1012.0</td>\n",
       "      <td>12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2015/9/2</td>\n",
       "      <td>104</td>\n",
       "      <td>21183</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>online</td>\n",
       "      <td>428.0</td>\n",
       "      <td>109</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2015/9/2</td>\n",
       "      <td>104</td>\n",
       "      <td>20448</td>\n",
       "      <td>308</td>\n",
       "      <td>404</td>\n",
       "      <td>online</td>\n",
       "      <td>962.0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2015/9/2</td>\n",
       "      <td>104</td>\n",
       "      <td>21565</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>offline</td>\n",
       "      <td>1400.0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  order_date  sales_region_code  item_code  first_cate_code  second_cate_code  \\\n",
       "0   2015/9/1                104      22069              307               403   \n",
       "1   2015/9/1                104      20028              301               405   \n",
       "2   2015/9/2                104      21183              307               403   \n",
       "3   2015/9/2                104      20448              308               404   \n",
       "4   2015/9/2                104      21565              307               403   \n",
       "\n",
       "  sales_chan_name  item_price  ord_qty  \n",
       "0         offline      1114.0       19  \n",
       "1         offline      1012.0       12  \n",
       "2          online       428.0      109  \n",
       "3          online       962.0        3  \n",
       "4         offline      1400.0        3  "
      ]
     },
     "execution_count": 147,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3360481b",
   "metadata": {},
   "source": [
    "## 数据类型处理"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 148,
   "id": "c36924e1",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 时间 object->datetime格式\n",
    "sales_data[\"order_date\"]=sales_data[\"order_date\"].apply(pd.to_datetime,format='%Y-%m-%d')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 149,
   "id": "d4ecf444",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 线上0 线下1\n",
    "dic = {\"offline\":1,\"online\":0}\n",
    "sales_data['sales_chan_name'] = sales_data['sales_chan_name'].map(dic)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "id": "a1e602d3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([20001, 20002, 20003, ..., 22082, 22083, 22084]),\n",
       " array([  13,  694, 2767, ...,  159,  311,  468]))"
      ]
     },
     "execution_count": 150,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 单独的item代码，以及其对应的记录条数\n",
    "train_item = np.unique(sales_data[\"item_code\"],return_counts=True)\n",
    "train_item"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "id": "833c2454",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Unique item codes: 1754\n",
      "大类编码 [307 301 308 305 303 306 302 304]\n",
      "细类编码 [403 405 404 412 410 401 407 408 402 409 406 411]\n",
      "地区编码 [104 102 101 105 103]\n"
     ]
    }
   ],
   "source": [
    "# 当然这是做了异常值处理之后的\n",
    "print('Unique item codes:', len(sales_data['item_code'].unique()))  # 1421种\n",
    "print('大类编码', sales_data[\"first_cate_code\"].unique())  # 8大类\n",
    "print('细类编码', sales_data[\"second_cate_code\"].unique()) # 12小类\n",
    "print('地区编码', sales_data[\"sales_region_code\"].unique()) # 5个地区\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "87701d6f",
   "metadata": {},
   "source": [
    "## 检验这四类code是否一一对应\n",
    "对后续如何处理很关键"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "id": "98116680",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "first_cate_code  second_cate_code\n",
       "301              405                  21457\n",
       "302              408                  94785\n",
       "303              401                  60167\n",
       "                 406                   5187\n",
       "                 410                   5011\n",
       "                 411                   1437\n",
       "304              409                   9996\n",
       "305              412                  75165\n",
       "306              402                  12262\n",
       "                 407                 207185\n",
       "307              403                  50855\n",
       "308              404                  54072\n",
       "dtype: int64"
      ]
     },
     "execution_count": 152,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 大类和细类是否一一对应\n",
    "grouped_data = sales_data.groupby(['first_cate_code', 'second_cate_code']).size() #.reset_index(name='counts')\n",
    "print(len(grouped_data))\n",
    "grouped_data\n",
    "# 发现大类和细类是一一对应的"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "id": "8897938c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "有 1754 种商品只属于一个second_cate_code\n",
      "每个second_cate_code中包含的商品数量：\n",
      "    second_cate_code  counts\n",
      "0                401     183\n",
      "1                402      15\n",
      "2                403     163\n",
      "3                404     140\n",
      "4                405      36\n",
      "5                406      14\n",
      "6                407     498\n",
      "7                408     404\n",
      "8                409      71\n",
      "9                410      31\n",
      "10               411       7\n",
      "11               412     192\n"
     ]
    }
   ],
   "source": [
    "# 统计每个second_cate_code中包含的商品种类数量\n",
    "second_cate_count = sales_data.groupby('second_cate_code')['item_code'].nunique().reset_index(name='counts')\n",
    "# second_cate_count\n",
    "# 统计每个商品所属的second_cate_code数量，并检查是否有某个商品只属于一个second_cate_code\n",
    "item_counts = sales_data.groupby('item_code')['second_cate_code'].nunique()\n",
    "unique_items = item_counts[item_counts == 1].index.tolist()\n",
    "if len(unique_items) > 0:\n",
    "    print(f\"有 {len(unique_items)} 种商品只属于一个second_cate_code\")\n",
    "else:\n",
    "    print(\"所有商品都属于多个second_cate_code\")\n",
    "\n",
    "# 输出每个second_cate_code中包含的商品数量\n",
    "print(\"每个second_cate_code中包含的商品数量：\")\n",
    "print(second_cate_count)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "id": "3a0f310b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "有 355 种商品只在一个地区销售\n",
      "有 241 种商品在两个地区销售\n",
      "有 224 种商品在三个地区销售\n",
      "有 828 种商品在四个地区销售\n",
      "有 106 种商品在五个地区销售\n",
      "每个sales_region_codee中包含的商品数量：\n",
      "   sales_region_code  counts\n",
      "0                101    1290\n",
      "1                102    1250\n",
      "2                103    1203\n",
      "3                104     257\n",
      "4                105    1351\n"
     ]
    }
   ],
   "source": [
    "## 地区和商品\n",
    "# 统计每个sales_region_code中包含的商品种类数量\n",
    "sales_region_count = sales_data.groupby('sales_region_code')['item_code'].nunique().reset_index(name='counts')\n",
    "# second_cate_count\n",
    "# 统计每个商品所属的sales_region_code数量，并检查是否有某个商品只属于一个sales_region_code\n",
    "item_counts = sales_data.groupby('item_code')['sales_region_code'].nunique()\n",
    "unique_items = item_counts[item_counts == 1].index.tolist()\n",
    "two_items = item_counts[item_counts == 2].index.tolist()\n",
    "three_items = item_counts[item_counts == 3].index.tolist()\n",
    "four_items = item_counts[item_counts == 4].index.tolist()\n",
    "five_items = item_counts[item_counts == 5].index.tolist()\n",
    "if len(unique_items) > 0:\n",
    "    print(f\"有 {len(unique_items)} 种商品只在一个地区销售\")\n",
    "if len(two_items) > 0:\n",
    "    print(f\"有 {len(two_items)} 种商品在两个地区销售\")\n",
    "if len(three_items) > 0:  \n",
    "    print(f\"有 {len(three_items)} 种商品在三个地区销售\")\n",
    "if len(four_items) > 0:\n",
    "    print(f\"有 {len(four_items)} 种商品在四个地区销售\")\n",
    "if len(five_items) > 0:\n",
    "    print(f\"有 {len(five_items)} 种商品在五个地区销售\")\n",
    "\n",
    "# 输出每个sales_region_code中包含的商品数量\n",
    "print(\"每个sales_region_codee中包含的商品数量：\")\n",
    "print(sales_region_count)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "08431a5b",
   "metadata": {},
   "source": [
    "- 简单总结\n",
    "   - 一级分类和二级分类是一对多的\n",
    "   - 每个商品只有一个类别，不会对用多种\n",
    "   - 1421 种商品中，只有 332 种商品是只在一个地区销售的，其他的都在多个地区销售"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "03df2f16",
   "metadata": {},
   "source": [
    "## 编码label_encoding\n",
    "- 将地区、item_code、类别进行顺序编码"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "id": "ea93feb8",
   "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_date</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>sales_chan_name</th>\n",
       "      <th>item_price</th>\n",
       "      <th>ord_qty</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2015-09-01</td>\n",
       "      <td>104</td>\n",
       "      <td>22069</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>1</td>\n",
       "      <td>1114.0</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2015-09-01</td>\n",
       "      <td>104</td>\n",
       "      <td>20028</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "      <td>1</td>\n",
       "      <td>1012.0</td>\n",
       "      <td>12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>21183</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>0</td>\n",
       "      <td>428.0</td>\n",
       "      <td>109</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>20448</td>\n",
       "      <td>308</td>\n",
       "      <td>404</td>\n",
       "      <td>0</td>\n",
       "      <td>962.0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>21565</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>1</td>\n",
       "      <td>1400.0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  order_date  sales_region_code  item_code  first_cate_code  second_cate_code  \\\n",
       "0 2015-09-01                104      22069              307               403   \n",
       "1 2015-09-01                104      20028              301               405   \n",
       "2 2015-09-02                104      21183              307               403   \n",
       "3 2015-09-02                104      20448              308               404   \n",
       "4 2015-09-02                104      21565              307               403   \n",
       "\n",
       "   sales_chan_name  item_price  ord_qty  \n",
       "0                1      1114.0       19  \n",
       "1                1      1012.0       12  \n",
       "2                0       428.0      109  \n",
       "3                0       962.0        3  \n",
       "4                1      1400.0        3  "
      ]
     },
     "execution_count": 155,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6ef4ed32",
   "metadata": {},
   "source": [
    "这里test有train中没有的商品类别（没有更新）\n",
    "包括\n",
    "> y contains previously unseen labels: [20011, 20025, 20198, 20273, 20324, 20401, 20449, 20456, 20469, 20503, 20549, 20701, 20721, 20736, 20750, 20793, 20841, 20869, 20951, 20999, 21085, 21184, 21393, 21416, 21523, 21606, 21667, 21687, 21823, 21867, 21879, 21954, 22075]\n",
    "\n",
    "所以就先不对item code进行编码了"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "id": "741f97ba",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "# from sklearn.preprocessing import LabelEncoder\n",
    "# # 地区\n",
    "# LE1 = LabelEncoder()\n",
    "# sales_data[\"sales_region_code\"] = LE1.fit_transform(sales_data.sales_region_code )\n",
    "# test[\"sales_region_code\"] = LE1.transform(test.sales_region_code)\n",
    "# # itemcode\n",
    "# # LE2 = LabelEncoder()\n",
    "# # sales_data[\"item_code\"] = LE2.fit_transform(sales_data.item_code )\n",
    "# # test[\"item_code\"] = LE2.transform( test.item_code )\n",
    "# # 大类\n",
    "# LE3 = LabelEncoder()\n",
    "# sales_data[\"first_cate_code\"] = LE3.fit_transform(sales_data.first_cate_code )\n",
    "# test[\"first_cate_code\"] = LE3.transform( test.first_cate_code )\n",
    "# # 细类\n",
    "# LE4 = LabelEncoder()\n",
    "# sales_data[\"second_cate_code\"] = LE4.fit_transform(sales_data.second_cate_code )\n",
    "# test[\"second_cate_code\"] = LE4.transform( test.second_cate_code )\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 157,
   "id": "93d24b41",
   "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_date</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>sales_chan_name</th>\n",
       "      <th>item_price</th>\n",
       "      <th>ord_qty</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2015-09-01</td>\n",
       "      <td>104</td>\n",
       "      <td>22069</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>1</td>\n",
       "      <td>1114.0</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2015-09-01</td>\n",
       "      <td>104</td>\n",
       "      <td>20028</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "      <td>1</td>\n",
       "      <td>1012.0</td>\n",
       "      <td>12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>21183</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>0</td>\n",
       "      <td>428.0</td>\n",
       "      <td>109</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>20448</td>\n",
       "      <td>308</td>\n",
       "      <td>404</td>\n",
       "      <td>0</td>\n",
       "      <td>962.0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>21565</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>1</td>\n",
       "      <td>1400.0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  order_date  sales_region_code  item_code  first_cate_code  second_cate_code  \\\n",
       "0 2015-09-01                104      22069              307               403   \n",
       "1 2015-09-01                104      20028              301               405   \n",
       "2 2015-09-02                104      21183              307               403   \n",
       "3 2015-09-02                104      20448              308               404   \n",
       "4 2015-09-02                104      21565              307               403   \n",
       "\n",
       "   sales_chan_name  item_price  ord_qty  \n",
       "0                1      1114.0       19  \n",
       "1                1      1012.0       12  \n",
       "2                0       428.0      109  \n",
       "3                0       962.0        3  \n",
       "4                1      1400.0        3  "
      ]
     },
     "execution_count": 157,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0b040271",
   "metadata": {},
   "source": [
    "# 2. Preprocessing"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "df98009b",
   "metadata": {},
   "source": [
    "- 标签平滑处理：取对数，其实应该放在不同的时间精度不同，比如按月预测的话，应该先按月合成再取对数\n",
    "- 按照月份递增的顺序，创建一个包含月份、地区和商品的组合的矩阵df。\n",
    "    - Item_cnt_day（ord_qty）被求和为Item_cnt_month。\n",
    "- product()函数也可以用于求取一个可迭代对象中自己各个元素的全排列组合\n",
    "\n",
    "- 因为每一个月，每个region种所出售的item情况不同，很多时候可能是0（没有这个记录），但是这都要参与到我们的建模中。所以我们要先将所有的情况都列出来。然后再与现有数据合并，缺失的数据就填补为0，这样构建数据集"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a6303889",
   "metadata": {},
   "source": [
    "- ？？？\n",
    "- 这里暂时按月合成的时候，先忽略线上、线下的因素，全部一股脑合并先，需要补充eda，对不同商品线上线下销售情况进行EDA，再进行分析"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 158,
   "id": "abff876f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# # 销量取对数\n",
    "# import math\n",
    "# sales_data[\"ord_qty\"] = list(map(lambda x : x if x== np.NaN else math.log(x+1,2),sales_data[\"ord_qty\"]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 159,
   "id": "e1f13693",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 生成一列月份编号\n",
    "sales_data[\"month\"] = sales_data.order_date.dt.month\n",
    "sales_data[\"year\"] = sales_data.order_date.dt.year\n",
    "sales_data[\"date_block_num\"] = (sales_data[\"year\"]-sales_data[\"year\"].min())*12+sales_data[\"month\"]-9 # 减8是因为第一个月是9月份\n",
    "\n",
    "# 输出结果\n",
    "del sales_data[\"month\"],sales_data[\"year\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 160,
   "id": "d7e78511",
   "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_date</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>sales_chan_name</th>\n",
       "      <th>item_price</th>\n",
       "      <th>ord_qty</th>\n",
       "      <th>date_block_num</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>597689</th>\n",
       "      <td>2018-12-20</td>\n",
       "      <td>102</td>\n",
       "      <td>20994</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "      <td>1</td>\n",
       "      <td>843.0</td>\n",
       "      <td>59</td>\n",
       "      <td>39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>597690</th>\n",
       "      <td>2018-12-20</td>\n",
       "      <td>102</td>\n",
       "      <td>21875</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "      <td>1</td>\n",
       "      <td>762.0</td>\n",
       "      <td>502</td>\n",
       "      <td>39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>597691</th>\n",
       "      <td>2018-12-20</td>\n",
       "      <td>102</td>\n",
       "      <td>20215</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "      <td>1</td>\n",
       "      <td>2013.0</td>\n",
       "      <td>106</td>\n",
       "      <td>39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>597692</th>\n",
       "      <td>2018-12-20</td>\n",
       "      <td>102</td>\n",
       "      <td>20195</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "      <td>1</td>\n",
       "      <td>2120.0</td>\n",
       "      <td>187</td>\n",
       "      <td>39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>597693</th>\n",
       "      <td>2018-12-20</td>\n",
       "      <td>102</td>\n",
       "      <td>20321</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "      <td>1</td>\n",
       "      <td>1244.0</td>\n",
       "      <td>205</td>\n",
       "      <td>39</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       order_date  sales_region_code  item_code  first_cate_code  \\\n",
       "597689 2018-12-20                102      20994              302   \n",
       "597690 2018-12-20                102      21875              302   \n",
       "597691 2018-12-20                102      20215              302   \n",
       "597692 2018-12-20                102      20195              302   \n",
       "597693 2018-12-20                102      20321              302   \n",
       "\n",
       "        second_cate_code  sales_chan_name  item_price  ord_qty  date_block_num  \n",
       "597689               408                1       843.0       59              39  \n",
       "597690               408                1       762.0      502              39  \n",
       "597691               408                1      2013.0      106              39  \n",
       "597692               408                1      2120.0      187              39  \n",
       "597693               408                1      1244.0      205              39  "
      ]
     },
     "execution_count": 160,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# sales_data.head()\n",
    "sales_data.tail()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "59a295d9",
   "metadata": {},
   "source": [
    "尝试直接剔除104地区\n",
    "实际上应该不需要"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "130f989b",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 161,
   "id": "af7a93dd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.19086503982543945"
      ]
     },
     "execution_count": 161,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from itertools import product\n",
    "import time\n",
    "ts = time.time()\n",
    "matrix = []\n",
    "cols  = [\"date_block_num\",\"sales_region_code\", \"item_code\"]\n",
    "for i in range(40):\n",
    "    sales = sales_data[sales_data.date_block_num == i]\n",
    "    region_list = sales.sales_region_code.unique() # （不）剔除了104地区\n",
    "    item_list = sales.item_code.unique()\n",
    "    matrix.append( np.array(list( product( [i],region_list , item_list) ), dtype = np.int16) )\n",
    "\n",
    "matrix = pd.DataFrame( np.vstack(matrix), columns = cols ) # 堆叠\n",
    "matrix[\"date_block_num\"] = matrix[\"date_block_num\"].astype(np.int8)\n",
    "matrix[\"sales_region_code\"] = matrix[\"sales_region_code\"].astype(np.int8)\n",
    "matrix[\"item_code\"] = matrix[\"item_code\"].astype(np.int16)\n",
    "matrix.sort_values( cols, inplace = True )\n",
    "time.time()- ts "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "id": "688cc900",
   "metadata": {},
   "outputs": [],
   "source": [
    "# add revenue to train df 收入\n",
    "sales_data[\"revenue\"] = sales_data[\"item_price\"] * sales_data[\"ord_qty\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 163,
   "id": "b2fb3d5d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.12241411209106445"
      ]
     },
     "execution_count": 163,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "group = sales_data.groupby( [\"date_block_num\", \"sales_region_code\", \"item_code\"] ).agg( {\"ord_qty\": [\"sum\"]} )\n",
    "group.columns = [\"item_cnt_month\"]\n",
    "group.reset_index( inplace = True) # 这一步重置索引，直接把层级索引转换为列\n",
    "matrix = pd.merge( matrix, group, on = cols, how = \"left\" ) # 转换后，二者就可以merge了\n",
    "matrix[\"item_cnt_month\"] = matrix[\"item_cnt_month\"].fillna(0).astype(np.float16) # 填补缺失值\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 164,
   "id": "f4222e92",
   "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>date_block_num</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>item_cnt_month</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20003</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20006</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20007</td>\n",
       "      <td>3092.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20008</td>\n",
       "      <td>308.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20009</td>\n",
       "      <td>260.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   date_block_num  sales_region_code  item_code  item_cnt_month\n",
       "0               0                101      20003             0.0\n",
       "1               0                101      20006             0.0\n",
       "2               0                101      20007          3092.0\n",
       "3               0                101      20008           308.0\n",
       "4               0                101      20009           260.0"
      ]
     },
     "execution_count": 164,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# matrix就是每月、每个地区、每个商品的销量\n",
    "matrix.head()  "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "de81c6a1",
   "metadata": {},
   "source": [
    "- 为第40个月设置一下test集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "id": "77a66e3b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 对测试集添加一个月份列，同时设置一下三列的数据格式\n",
    "test[\"date_block_num\"] = 40\n",
    "test[\"date_block_num\"] = test[\"date_block_num\"].astype(np.int8)\n",
    "test[\"sales_region_code\"] = test.sales_region_code.astype(np.int8)\n",
    "test[\"item_code\"] = test.item_code.astype(np.int16)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "45c15bc0",
   "metadata": {},
   "source": [
    "- 合并训练集和测试集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "id": "ab2cba33",
   "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>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>item_code</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>22069</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "      <td>20028</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>21183</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>308</td>\n",
       "      <td>404</td>\n",
       "      <td>20448</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>21565</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   first_cate_code  second_cate_code  item_code\n",
       "0              307               403      22069\n",
       "1              301               405      20028\n",
       "2              307               403      21183\n",
       "3              308               404      20448\n",
       "4              307               403      21565"
      ]
     },
     "execution_count": 166,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "map_cate_df = sales_data[[\"first_cate_code\",\"second_cate_code\",\"item_code\"]].drop_duplicates()\n",
    "map_cate_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "id": "be0d090e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# matrix.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "id": "69b494f2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 给matrix补充上类别属性\n",
    "matrix = pd.merge(matrix, map_cate_df, on = [\"item_code\"], how = \"left\")\n",
    "matrix[\"first_cate_code\"] = matrix[\"first_cate_code\"].astype(np.int16)\n",
    "matrix[\"second_cate_code\"] = matrix[\"second_cate_code\"].astype(np.int16)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 169,
   "id": "b30d72a9",
   "metadata": {},
   "outputs": [],
   "source": [
    "# matrix.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 170,
   "id": "5ebac8fe",
   "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>date_block_num</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>item_cnt_month</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20003</td>\n",
       "      <td>0.0</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20006</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20007</td>\n",
       "      <td>3092.0</td>\n",
       "      <td>305</td>\n",
       "      <td>412</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20008</td>\n",
       "      <td>308.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20009</td>\n",
       "      <td>260.0</td>\n",
       "      <td>306</td>\n",
       "      <td>407</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>107696</th>\n",
       "      <td>39</td>\n",
       "      <td>105</td>\n",
       "      <td>22080</td>\n",
       "      <td>16.0</td>\n",
       "      <td>306</td>\n",
       "      <td>407</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>107697</th>\n",
       "      <td>39</td>\n",
       "      <td>105</td>\n",
       "      <td>22081</td>\n",
       "      <td>7.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>107698</th>\n",
       "      <td>39</td>\n",
       "      <td>105</td>\n",
       "      <td>22082</td>\n",
       "      <td>0.0</td>\n",
       "      <td>305</td>\n",
       "      <td>412</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>107699</th>\n",
       "      <td>39</td>\n",
       "      <td>105</td>\n",
       "      <td>22083</td>\n",
       "      <td>1124.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>107700</th>\n",
       "      <td>39</td>\n",
       "      <td>105</td>\n",
       "      <td>22084</td>\n",
       "      <td>0.0</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>107701 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        date_block_num  sales_region_code  item_code  item_cnt_month  \\\n",
       "0                    0                101      20003             0.0   \n",
       "1                    0                101      20006             0.0   \n",
       "2                    0                101      20007          3092.0   \n",
       "3                    0                101      20008           308.0   \n",
       "4                    0                101      20009           260.0   \n",
       "...                ...                ...        ...             ...   \n",
       "107696              39                105      22080            16.0   \n",
       "107697              39                105      22081             7.0   \n",
       "107698              39                105      22082             0.0   \n",
       "107699              39                105      22083          1124.0   \n",
       "107700              39                105      22084             0.0   \n",
       "\n",
       "        first_cate_code  second_cate_code  \n",
       "0                   301               405  \n",
       "1                   307               403  \n",
       "2                   305               412  \n",
       "3                   303               401  \n",
       "4                   306               407  \n",
       "...                 ...               ...  \n",
       "107696              306               407  \n",
       "107697              303               401  \n",
       "107698              305               412  \n",
       "107699              303               401  \n",
       "107700              302               408  \n",
       "\n",
       "[107701 rows x 6 columns]"
      ]
     },
     "execution_count": 170,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 171,
   "id": "42cf3c0a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.006050825119018555"
      ]
     },
     "execution_count": 171,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 合并训练和测试\n",
    "ts = time.time()\n",
    "# cols  = [\"date_block_num\",\"sales_region_code\", \"item_code\"]\n",
    "\n",
    "matrix = pd.concat([matrix, test], ignore_index=True, sort=False, keys=cols) # 按三列，添加？\n",
    "matrix.fillna( 0, inplace = True )\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 172,
   "id": "20a9ca0e",
   "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>date_block_num</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>item_cnt_month</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20003</td>\n",
       "      <td>0.0</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20006</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20007</td>\n",
       "      <td>3092.0</td>\n",
       "      <td>305</td>\n",
       "      <td>412</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20008</td>\n",
       "      <td>308.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20009</td>\n",
       "      <td>260.0</td>\n",
       "      <td>306</td>\n",
       "      <td>407</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110315</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22066</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110316</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22072</td>\n",
       "      <td>0.0</td>\n",
       "      <td>305</td>\n",
       "      <td>412</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110317</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22075</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110318</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22083</td>\n",
       "      <td>0.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110319</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22084</td>\n",
       "      <td>0.0</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>110320 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        date_block_num  sales_region_code  item_code  item_cnt_month  \\\n",
       "0                    0                101      20003             0.0   \n",
       "1                    0                101      20006             0.0   \n",
       "2                    0                101      20007          3092.0   \n",
       "3                    0                101      20008           308.0   \n",
       "4                    0                101      20009           260.0   \n",
       "...                ...                ...        ...             ...   \n",
       "110315              40                105      22066             0.0   \n",
       "110316              40                105      22072             0.0   \n",
       "110317              40                105      22075             0.0   \n",
       "110318              40                105      22083             0.0   \n",
       "110319              40                105      22084             0.0   \n",
       "\n",
       "        first_cate_code  second_cate_code  \n",
       "0                   301               405  \n",
       "1                   307               403  \n",
       "2                   305               412  \n",
       "3                   303               401  \n",
       "4                   306               407  \n",
       "...                 ...               ...  \n",
       "110315              307               403  \n",
       "110316              305               412  \n",
       "110317              307               403  \n",
       "110318              303               401  \n",
       "110319              302               408  \n",
       "\n",
       "[110320 rows x 6 columns]"
      ]
     },
     "execution_count": 172,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "58615d0b",
   "metadata": {},
   "source": [
    "取对数，+1是因为销量有可能为0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 173,
   "id": "58d4f19c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# # 销量取对数\n",
    "import math\n",
    "if is_need_log:\n",
    "    matrix[\"item_cnt_month\"] = list(map(lambda x : x if x== np.NaN else math.log(x+1,2),matrix[\"item_cnt_month\"]))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0442ddfb",
   "metadata": {},
   "source": [
    "#  3. Feature Engineering"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "40d13b7a",
   "metadata": {},
   "source": [
    "## 3.1 Add lag features to matrix df."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 174,
   "id": "ae5b5a9f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define a lag feature function\n",
    "def lag_feature( df,lags, cols ):\n",
    "    for col in cols:\n",
    "        print(col)\n",
    "        tmp = df[[\"date_block_num\", \"sales_region_code\",\"item_code\",col ]]\n",
    "        for i in lags:\n",
    "            shifted = tmp.copy()\n",
    "            shifted.columns = [\"date_block_num\", \"sales_region_code\",\"item_code\", col + \"_lag_\"+str(i)]\n",
    "            shifted.date_block_num = shifted.date_block_num + i\n",
    "            df = pd.merge(df, shifted, on=[\"date_block_num\", \"sales_region_code\",\"item_code\"], how='left')\n",
    "    return df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d551ce2b",
   "metadata": {},
   "source": [
    "- Add item_cnt_month lag features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "id": "ce12809a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "item_cnt_month\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.12216329574584961"
      ]
     },
     "execution_count": 175,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "matrix = lag_feature( matrix, [1,2,3], [\"item_cnt_month\"] ) # 添加该商店的该商品一个月、两个月、三个月的销量\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7902376e",
   "metadata": {},
   "source": [
    "- Add the previous month's average item_cnt. 上个月的平均销量"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 176,
   "id": "256ccd7f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "date_avg_item_cnt\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.07225799560546875"
      ]
     },
     "execution_count": 176,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "# 计算每个月的平均销量（这个月的所有大杂烩）\n",
    "group = matrix.groupby( [\"date_block_num\"] ).agg({\"item_cnt_month\" : [\"mean\"]})\n",
    "group.columns = [\"date_avg_item_cnt\"]\n",
    "group.reset_index(inplace = True)\n",
    "\n",
    "# 在matrix中添加  前一个月的平均销量 作为一个lag feature\n",
    "matrix = pd.merge(matrix, group, on = [\"date_block_num\"], how = \"left\")\n",
    "matrix.date_avg_item_cnt = matrix[\"date_avg_item_cnt\"].astype(np.float16)\n",
    "matrix = lag_feature( matrix, [1], [\"date_avg_item_cnt\"] )\n",
    "matrix.drop( [\"date_avg_item_cnt\"], axis = 1, inplace = True )\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bde4eee4",
   "metadata": {},
   "source": [
    "- Add lag values of item_cnt_month for month / item_id."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 177,
   "id": "920a962f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "date_item_avg_item_cnt\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.16543793678283691"
      ]
     },
     "execution_count": 177,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "# 计算每个月该商品的平均销量（这个月的该商品）\n",
    "group = matrix.groupby(['date_block_num', 'item_code']).agg({'item_cnt_month': ['mean']})\n",
    "group.columns = [ 'date_item_avg_item_cnt' ]\n",
    "group.reset_index(inplace=True)\n",
    "\n",
    "# 添加 前三个月该商品的平均销量\n",
    "matrix = pd.merge(matrix, group, on=['date_block_num','item_code'], how='left')\n",
    "matrix.date_item_avg_item_cnt = matrix['date_item_avg_item_cnt'].astype(np.float16)\n",
    "matrix = lag_feature(matrix, [1,2,3], ['date_item_avg_item_cnt'])\n",
    "matrix.drop(['date_item_avg_item_cnt'], axis=1, inplace=True)\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9e874b14",
   "metadata": {},
   "source": [
    "- Add lag values for item_cnt_month for every month / region combination."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 178,
   "id": "22a65673",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "date_region_avg_item_cnt\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "<ipython-input-178-1f9a7b212985>:8: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/pandas-docs/stable/indexing.html#attribute-access\n",
      "  matrix.date_avg_item_cnt = matrix[\"date_region_avg_item_cnt\"].astype(np.float16)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.18927478790283203"
      ]
     },
     "execution_count": 178,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "# 计算 该地区的平均销量\n",
    "group = matrix.groupby( [\"date_block_num\",\"sales_region_code\"] ).agg({\"item_cnt_month\" : [\"mean\"]})\n",
    "group.columns = [\"date_region_avg_item_cnt\"]\n",
    "group.reset_index(inplace = True)\n",
    "\n",
    "matrix = pd.merge(matrix, group, on = [\"date_block_num\",\"sales_region_code\"], how = \"left\")\n",
    "matrix.date_avg_item_cnt = matrix[\"date_region_avg_item_cnt\"].astype(np.float16)\n",
    "matrix = lag_feature( matrix, [1,2,3], [\"date_region_avg_item_cnt\"] )\n",
    "matrix.drop( [\"date_region_avg_item_cnt\"], axis = 1, inplace = True )\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2369b88c",
   "metadata": {},
   "source": [
    "- Add lag values for item_cnt_month for month/region/item."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 179,
   "id": "18bf6c9d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "date_region_item_avg_item_cnt\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "<ipython-input-179-4c9cee418e72>:8: UserWarning: Pandas doesn't allow columns to be created via a new attribute name - see https://pandas.pydata.org/pandas-docs/stable/indexing.html#attribute-access\n",
      "  matrix.date_avg_item_cnt = matrix[\"date_region_item_avg_item_cnt\"].astype(np.float16)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.2372748851776123"
      ]
     },
     "execution_count": 179,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "# 计算每月该地区、该商品的平均销量\n",
    "group = matrix.groupby( [\"date_block_num\",\"sales_region_code\",\"item_code\"] ).agg({\"item_cnt_month\" : [\"mean\"]})\n",
    "group.columns = [\"date_region_item_avg_item_cnt\"]\n",
    "group.reset_index(inplace = True)\n",
    "\n",
    "matrix = pd.merge(matrix, group, on = [\"date_block_num\",\"sales_region_code\",\"item_code\"], how = \"left\")\n",
    "matrix.date_avg_item_cnt = matrix[\"date_region_item_avg_item_cnt\"].astype(np.float16)\n",
    "matrix = lag_feature( matrix, [1,2,3], [\"date_region_item_avg_item_cnt\"] )\n",
    "matrix.drop( [\"date_region_item_avg_item_cnt\"], axis = 1, inplace = True )\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "db025b26",
   "metadata": {},
   "source": [
    "- Add lag values for item_cnt_month for month/region/item second_cate(subtype)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 180,
   "id": "da595deb",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "date_region_subtype_avg_item_cnt\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.12911415100097656"
      ]
     },
     "execution_count": 180,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "# 计算每月该地区、该子类商品的 销售量\n",
    "group = matrix.groupby(['date_block_num', 'sales_region_code', 'second_cate_code']).agg({'item_cnt_month': ['mean']})\n",
    "group.columns = ['date_region_subtype_avg_item_cnt']\n",
    "group.reset_index(inplace=True)\n",
    "\n",
    "# 添加上一月的\n",
    "matrix = pd.merge(matrix, group, on=['date_block_num', 'sales_region_code', 'second_cate_code'], how='left')\n",
    "matrix.date_region_subtype_avg_item_cnt = matrix['date_region_subtype_avg_item_cnt'].astype(np.float16)\n",
    "matrix = lag_feature(matrix, [1], ['date_region_subtype_avg_item_cnt'])\n",
    "matrix.drop(['date_region_subtype_avg_item_cnt'], axis=1, inplace=True)\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dd3d0d2a",
   "metadata": {},
   "source": [
    "- 其实还可以补充一个大类商品的销售情况（有需要再说吧）"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e988a3d0",
   "metadata": {},
   "source": [
    "## 3.2 add price trend feature\n",
    "* Add average item price on to matix df. \n",
    "* Add lag values of item price per month.\n",
    "* Add delta price values - how current month average pirce relates to global average.\n",
    "\n",
    "前面算的一直是销量，这里开始添加的特征是价格\n",
    "虽然要求预测的是销量，但是价格与销量息息相关，也是重要的特征\n",
    "\n",
    "Price trend features，最后只有一个'delta_price_lag'特征，具体看代码注释"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 181,
   "id": "9d0bb94d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['date_block_num', 'sales_region_code', 'item_code', 'item_cnt_month',\n",
       "       'first_cate_code', 'second_cate_code', 'item_cnt_month_lag_1',\n",
       "       'item_cnt_month_lag_2', 'item_cnt_month_lag_3',\n",
       "       'date_avg_item_cnt_lag_1', 'date_item_avg_item_cnt_lag_1',\n",
       "       'date_item_avg_item_cnt_lag_2', 'date_item_avg_item_cnt_lag_3',\n",
       "       'date_region_avg_item_cnt_lag_1', 'date_region_avg_item_cnt_lag_2',\n",
       "       'date_region_avg_item_cnt_lag_3', 'date_region_item_avg_item_cnt_lag_1',\n",
       "       'date_region_item_avg_item_cnt_lag_2',\n",
       "       'date_region_item_avg_item_cnt_lag_3',\n",
       "       'date_region_subtype_avg_item_cnt_lag_1'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 181,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "matrix.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 182,
   "id": "0467b8bd",
   "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_date</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>sales_chan_name</th>\n",
       "      <th>item_price</th>\n",
       "      <th>ord_qty</th>\n",
       "      <th>date_block_num</th>\n",
       "      <th>revenue</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2015-09-01</td>\n",
       "      <td>104</td>\n",
       "      <td>22069</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>1</td>\n",
       "      <td>1114.0</td>\n",
       "      <td>19</td>\n",
       "      <td>0</td>\n",
       "      <td>21166.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2015-09-01</td>\n",
       "      <td>104</td>\n",
       "      <td>20028</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "      <td>1</td>\n",
       "      <td>1012.0</td>\n",
       "      <td>12</td>\n",
       "      <td>0</td>\n",
       "      <td>12144.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>21183</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>0</td>\n",
       "      <td>428.0</td>\n",
       "      <td>109</td>\n",
       "      <td>0</td>\n",
       "      <td>46652.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>20448</td>\n",
       "      <td>308</td>\n",
       "      <td>404</td>\n",
       "      <td>0</td>\n",
       "      <td>962.0</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>2886.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2015-09-02</td>\n",
       "      <td>104</td>\n",
       "      <td>21565</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>1</td>\n",
       "      <td>1400.0</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>4200.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  order_date  sales_region_code  item_code  first_cate_code  second_cate_code  \\\n",
       "0 2015-09-01                104      22069              307               403   \n",
       "1 2015-09-01                104      20028              301               405   \n",
       "2 2015-09-02                104      21183              307               403   \n",
       "3 2015-09-02                104      20448              308               404   \n",
       "4 2015-09-02                104      21565              307               403   \n",
       "\n",
       "   sales_chan_name  item_price  ord_qty  date_block_num  revenue  \n",
       "0                1      1114.0       19               0  21166.0  \n",
       "1                1      1012.0       12               0  12144.0  \n",
       "2                0       428.0      109               0  46652.0  \n",
       "3                0       962.0        3               0   2886.0  \n",
       "4                1      1400.0        3               0   4200.0  "
      ]
     },
     "execution_count": 182,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 183,
   "id": "e4f96073",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "date_item_avg_item_price\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "1.3211040496826172"
      ]
     },
     "execution_count": 183,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "# 添加该商品的平均价格\n",
    "group = sales_data.groupby( [\"item_code\"] ).agg({\"item_price\": [\"mean\"]})\n",
    "group.columns = [\"item_avg_item_price\"]\n",
    "group.reset_index(inplace = True)\n",
    "\n",
    "matrix = matrix.merge( group, on = [\"item_code\"], how = \"left\" )\n",
    "matrix[\"item_avg_item_price\"] = matrix.item_avg_item_price.astype(np.float16)\n",
    "\n",
    "\n",
    "\n",
    "# 添加该商品每月的平均价格，前三个月的\n",
    "group = sales_data.groupby( [\"date_block_num\",\"item_code\"] ).agg( {\"item_price\": [\"mean\"]} )\n",
    "group.columns = [\"date_item_avg_item_price\"]\n",
    "group.reset_index(inplace = True)\n",
    "\n",
    "matrix = matrix.merge(group, on = [\"date_block_num\",\"item_code\"], how = \"left\")\n",
    "matrix[\"date_item_avg_item_price\"] = matrix.date_item_avg_item_price.astype(np.float16)\n",
    "lags = [1, 2, 3]\n",
    "matrix = lag_feature( matrix, lags, [\"date_item_avg_item_price\"] ) # 添加价格\n",
    "for i in lags: # 添加价格趋势\n",
    "    matrix[\"delta_price_lag_\" + str(i) ] = (matrix[\"date_item_avg_item_price_lag_\" + str(i)]- matrix[\"item_avg_item_price\"] )/ matrix[\"item_avg_item_price\"]\n",
    "\n",
    "\n",
    "def select_trends(row) :\n",
    "    for i in lags:\n",
    "        if row[\"delta_price_lag_\" + str(i)]:\n",
    "            return row[\"delta_price_lag_\" + str(i)]\n",
    "    return 0\n",
    "\n",
    "# 这个函数看起来像是，该商品有上一个月的趋势，就取上一个月的。有上上个月的，就取上上个月的。。。如果前三个月都没有销售的价格，那么就取0\n",
    "matrix[\"delta_price_lag\"] = matrix.apply(select_trends, axis = 1)\n",
    "matrix[\"delta_price_lag\"] = matrix.delta_price_lag.astype( np.float16 )\n",
    "matrix[\"delta_price_lag\"].fillna( 0 ,inplace = True)\n",
    "\n",
    "# 删除该商品的平均价格，商品的每月平均价格，和每月价格趋势。最后仅仅保留了一个delta_price_lag\n",
    "features_to_drop = [\"item_avg_item_price\", \"date_item_avg_item_price\"]\n",
    "for i in lags:\n",
    "    features_to_drop.append(\"date_item_avg_item_price_lag_\" + str(i) )\n",
    "    features_to_drop.append(\"delta_price_lag_\" + str(i) )\n",
    "matrix.drop(features_to_drop, axis = 1, inplace = True)\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "58508ba6",
   "metadata": {},
   "source": [
    "* Add total shop revenue per month to matix df. \n",
    "* Add lag values of revenue per month.\n",
    "* Add delta revenue values - how current month revenue relates to global average.\n",
    "\n",
    "我们已经知道validation set和test set都会出现train set所没有的新商品，借助商店revenue趋势特征可以帮助算法预测新商品在现有店铺的销量\n",
    "\n",
    "该商店的特征同上，但是泰迪杯没有这样的商店列，所以可以用大类或者小类的商品销售趋势替代\n",
    "暂时用的是细类，其实可以优化为细类和地区的组合"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 184,
   "id": "62622529",
   "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>date_block_num</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>item_cnt_month</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>item_cnt_month_lag_1</th>\n",
       "      <th>item_cnt_month_lag_2</th>\n",
       "      <th>item_cnt_month_lag_3</th>\n",
       "      <th>date_avg_item_cnt_lag_1</th>\n",
       "      <th>...</th>\n",
       "      <th>date_item_avg_item_cnt_lag_2</th>\n",
       "      <th>date_item_avg_item_cnt_lag_3</th>\n",
       "      <th>date_region_avg_item_cnt_lag_1</th>\n",
       "      <th>date_region_avg_item_cnt_lag_2</th>\n",
       "      <th>date_region_avg_item_cnt_lag_3</th>\n",
       "      <th>date_region_item_avg_item_cnt_lag_1</th>\n",
       "      <th>date_region_item_avg_item_cnt_lag_2</th>\n",
       "      <th>date_region_item_avg_item_cnt_lag_3</th>\n",
       "      <th>date_region_subtype_avg_item_cnt_lag_1</th>\n",
       "      <th>delta_price_lag</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20003</td>\n",
       "      <td>0.0</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20006</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20007</td>\n",
       "      <td>3092.0</td>\n",
       "      <td>305</td>\n",
       "      <td>412</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20008</td>\n",
       "      <td>308.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>20009</td>\n",
       "      <td>260.0</td>\n",
       "      <td>306</td>\n",
       "      <td>407</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110315</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22066</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>7224.0</td>\n",
       "      <td>10536.0</td>\n",
       "      <td>5556.0</td>\n",
       "      <td>300.75</td>\n",
       "      <td>...</td>\n",
       "      <td>4584.00</td>\n",
       "      <td>3054.0</td>\n",
       "      <td>337.75</td>\n",
       "      <td>486.0</td>\n",
       "      <td>900.5</td>\n",
       "      <td>7224.0</td>\n",
       "      <td>10536.0</td>\n",
       "      <td>5556.0</td>\n",
       "      <td>433.750</td>\n",
       "      <td>0.091797</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110316</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22072</td>\n",
       "      <td>0.0</td>\n",
       "      <td>305</td>\n",
       "      <td>412</td>\n",
       "      <td>488.0</td>\n",
       "      <td>1842.0</td>\n",
       "      <td>1082.0</td>\n",
       "      <td>300.75</td>\n",
       "      <td>...</td>\n",
       "      <td>460.50</td>\n",
       "      <td>270.5</td>\n",
       "      <td>337.75</td>\n",
       "      <td>486.0</td>\n",
       "      <td>900.5</td>\n",
       "      <td>488.0</td>\n",
       "      <td>1842.0</td>\n",
       "      <td>1082.0</td>\n",
       "      <td>278.250</td>\n",
       "      <td>-0.006069</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110317</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22075</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110318</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22083</td>\n",
       "      <td>0.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "      <td>1124.0</td>\n",
       "      <td>1344.0</td>\n",
       "      <td>7420.0</td>\n",
       "      <td>300.75</td>\n",
       "      <td>...</td>\n",
       "      <td>2326.00</td>\n",
       "      <td>5108.0</td>\n",
       "      <td>337.75</td>\n",
       "      <td>486.0</td>\n",
       "      <td>900.5</td>\n",
       "      <td>1124.0</td>\n",
       "      <td>1344.0</td>\n",
       "      <td>7420.0</td>\n",
       "      <td>217.375</td>\n",
       "      <td>-0.006493</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110319</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22084</td>\n",
       "      <td>0.0</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>11.0</td>\n",
       "      <td>300.75</td>\n",
       "      <td>...</td>\n",
       "      <td>239.25</td>\n",
       "      <td>146.0</td>\n",
       "      <td>337.75</td>\n",
       "      <td>486.0</td>\n",
       "      <td>900.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>11.0</td>\n",
       "      <td>197.125</td>\n",
       "      <td>-0.010872</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>110320 rows × 21 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        date_block_num  sales_region_code  item_code  item_cnt_month  \\\n",
       "0                    0                101      20003             0.0   \n",
       "1                    0                101      20006             0.0   \n",
       "2                    0                101      20007          3092.0   \n",
       "3                    0                101      20008           308.0   \n",
       "4                    0                101      20009           260.0   \n",
       "...                ...                ...        ...             ...   \n",
       "110315              40                105      22066             0.0   \n",
       "110316              40                105      22072             0.0   \n",
       "110317              40                105      22075             0.0   \n",
       "110318              40                105      22083             0.0   \n",
       "110319              40                105      22084             0.0   \n",
       "\n",
       "        first_cate_code  second_cate_code  item_cnt_month_lag_1  \\\n",
       "0                   301               405                   NaN   \n",
       "1                   307               403                   NaN   \n",
       "2                   305               412                   NaN   \n",
       "3                   303               401                   NaN   \n",
       "4                   306               407                   NaN   \n",
       "...                 ...               ...                   ...   \n",
       "110315              307               403                7224.0   \n",
       "110316              305               412                 488.0   \n",
       "110317              307               403                   NaN   \n",
       "110318              303               401                1124.0   \n",
       "110319              302               408                   0.0   \n",
       "\n",
       "        item_cnt_month_lag_2  item_cnt_month_lag_3  date_avg_item_cnt_lag_1  \\\n",
       "0                        NaN                   NaN                      NaN   \n",
       "1                        NaN                   NaN                      NaN   \n",
       "2                        NaN                   NaN                      NaN   \n",
       "3                        NaN                   NaN                      NaN   \n",
       "4                        NaN                   NaN                      NaN   \n",
       "...                      ...                   ...                      ...   \n",
       "110315               10536.0                5556.0                   300.75   \n",
       "110316                1842.0                1082.0                   300.75   \n",
       "110317                   NaN                   NaN                      NaN   \n",
       "110318                1344.0                7420.0                   300.75   \n",
       "110319                   0.0                  11.0                   300.75   \n",
       "\n",
       "        ...  date_item_avg_item_cnt_lag_2  date_item_avg_item_cnt_lag_3  \\\n",
       "0       ...                           NaN                           NaN   \n",
       "1       ...                           NaN                           NaN   \n",
       "2       ...                           NaN                           NaN   \n",
       "3       ...                           NaN                           NaN   \n",
       "4       ...                           NaN                           NaN   \n",
       "...     ...                           ...                           ...   \n",
       "110315  ...                       4584.00                        3054.0   \n",
       "110316  ...                        460.50                         270.5   \n",
       "110317  ...                           NaN                           NaN   \n",
       "110318  ...                       2326.00                        5108.0   \n",
       "110319  ...                        239.25                         146.0   \n",
       "\n",
       "        date_region_avg_item_cnt_lag_1  date_region_avg_item_cnt_lag_2  \\\n",
       "0                                  NaN                             NaN   \n",
       "1                                  NaN                             NaN   \n",
       "2                                  NaN                             NaN   \n",
       "3                                  NaN                             NaN   \n",
       "4                                  NaN                             NaN   \n",
       "...                                ...                             ...   \n",
       "110315                          337.75                           486.0   \n",
       "110316                          337.75                           486.0   \n",
       "110317                             NaN                             NaN   \n",
       "110318                          337.75                           486.0   \n",
       "110319                          337.75                           486.0   \n",
       "\n",
       "        date_region_avg_item_cnt_lag_3  date_region_item_avg_item_cnt_lag_1  \\\n",
       "0                                  NaN                                  NaN   \n",
       "1                                  NaN                                  NaN   \n",
       "2                                  NaN                                  NaN   \n",
       "3                                  NaN                                  NaN   \n",
       "4                                  NaN                                  NaN   \n",
       "...                                ...                                  ...   \n",
       "110315                           900.5                               7224.0   \n",
       "110316                           900.5                                488.0   \n",
       "110317                             NaN                                  NaN   \n",
       "110318                           900.5                               1124.0   \n",
       "110319                           900.5                                  0.0   \n",
       "\n",
       "        date_region_item_avg_item_cnt_lag_2  \\\n",
       "0                                       NaN   \n",
       "1                                       NaN   \n",
       "2                                       NaN   \n",
       "3                                       NaN   \n",
       "4                                       NaN   \n",
       "...                                     ...   \n",
       "110315                              10536.0   \n",
       "110316                               1842.0   \n",
       "110317                                  NaN   \n",
       "110318                               1344.0   \n",
       "110319                                  0.0   \n",
       "\n",
       "        date_region_item_avg_item_cnt_lag_3  \\\n",
       "0                                       NaN   \n",
       "1                                       NaN   \n",
       "2                                       NaN   \n",
       "3                                       NaN   \n",
       "4                                       NaN   \n",
       "...                                     ...   \n",
       "110315                               5556.0   \n",
       "110316                               1082.0   \n",
       "110317                                  NaN   \n",
       "110318                               7420.0   \n",
       "110319                                 11.0   \n",
       "\n",
       "        date_region_subtype_avg_item_cnt_lag_1  delta_price_lag  \n",
       "0                                          NaN         0.000000  \n",
       "1                                          NaN         0.000000  \n",
       "2                                          NaN         0.000000  \n",
       "3                                          NaN         0.000000  \n",
       "4                                          NaN         0.000000  \n",
       "...                                        ...              ...  \n",
       "110315                                 433.750         0.091797  \n",
       "110316                                 278.250        -0.006069  \n",
       "110317                                     NaN         0.000000  \n",
       "110318                                 217.375        -0.006493  \n",
       "110319                                 197.125        -0.010872  \n",
       "\n",
       "[110320 rows x 21 columns]"
      ]
     },
     "execution_count": 184,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 185,
   "id": "5c7ddaf1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "delta_revenue\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.19498991966247559"
      ]
     },
     "execution_count": 185,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "group = sales_data.groupby( [\"date_block_num\",\"second_cate_code\"] ).agg({\"revenue\": [\"sum\"] })\n",
    "group.columns = [\"date_subtype_revenue\"]\n",
    "group.reset_index(inplace = True)\n",
    "\n",
    "matrix = matrix.merge( group , on = [\"date_block_num\", \"second_cate_code\"], how = \"left\" )\n",
    "matrix['date_subtype_revenue'] = matrix['date_subtype_revenue'].astype(np.float32)\n",
    "\n",
    "group = group.groupby([\"second_cate_code\"]).agg({ \"date_block_num\":[\"mean\"] })\n",
    "group.columns = [\"subtype_avg_revenue\"]\n",
    "group.reset_index(inplace = True )\n",
    "\n",
    "matrix = matrix.merge( group, on = [\"second_cate_code\"], how = \"left\" )\n",
    "matrix[\"subtype_avg_revenue\"] = matrix.subtype_avg_revenue.astype(np.float32)\n",
    "matrix[\"delta_revenue\"] = (matrix['date_subtype_revenue'] - matrix['subtype_avg_revenue']) / matrix['subtype_avg_revenue']\n",
    "matrix[\"delta_revenue\"] = matrix[\"delta_revenue\"]. astype(np.float32)\n",
    "\n",
    "matrix = lag_feature(matrix, [1], [\"delta_revenue\"])\n",
    "matrix[\"delta_revenue_lag_1\"] = matrix[\"delta_revenue_lag_1\"].astype(np.float32)\n",
    "matrix.drop( [\"date_subtype_revenue\", \"subtype_avg_revenue\", \"delta_revenue\"] ,axis = 1, inplace = True)\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "44869c9f",
   "metadata": {},
   "source": [
    "- Add month and number of days in each month to matrix df.\n",
    "\n",
    "添加月份、和该月有多少天的两个特征"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 186,
   "id": "8560c680",
   "metadata": {},
   "outputs": [],
   "source": [
    "matrix[\"month\"] = (matrix[\"date_block_num\"]+9) % 12 # 9个月开始\n",
    "days = pd.Series([31,28,31,30,31,30,31,31,30,31,30,31])\n",
    "matrix[\"days\"] = matrix[\"month\"].map(days).astype(np.int8)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6331a7cd",
   "metadata": {},
   "source": [
    "Add the month of each shop and item first sale.\n",
    "\n",
    "'item_subtype_last_sale'和'item_subtype_sale'用以纪录距离最近一次销售之间隔了几个月，通过它可以和之前月份的数据建立关联。'item_subtype_first_sale'和'item_first_sale'则是用于表示新商品的特征\n",
    "\n",
    "该商品从第一个销售到现在有几个月了"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 187,
   "id": "fb01e770",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.014854907989501953"
      ]
     },
     "execution_count": 187,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "matrix[\"item_subtype_first_sale\"] = matrix[\"date_block_num\"] - matrix.groupby([\"item_code\",\"second_cate_code\"])[\"date_block_num\"].transform('min')\n",
    "matrix[\"item_first_sale\"] = matrix[\"date_block_num\"] - matrix.groupby([\"item_code\"])[\"date_block_num\"].transform('min')\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e51ff0f6",
   "metadata": {},
   "source": [
    "Delete first three months from matrix. They don't have lag values."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 188,
   "id": "754b950a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.016722917556762695"
      ]
     },
     "execution_count": 188,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ts = time.time()\n",
    "matrix = matrix[matrix[\"date_block_num\"] > 3]\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 189,
   "id": "cc7403d7",
   "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>9940</th>\n",
       "      <th>9941</th>\n",
       "      <th>9942</th>\n",
       "      <th>9943</th>\n",
       "      <th>9944</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>date_block_num</th>\n",
       "      <td>4.000000</td>\n",
       "      <td>4.000000e+00</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>4.000000e+00</td>\n",
       "      <td>4.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>sales_region_code</th>\n",
       "      <td>101.000000</td>\n",
       "      <td>1.010000e+02</td>\n",
       "      <td>101.000000</td>\n",
       "      <td>1.010000e+02</td>\n",
       "      <td>1.010000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>item_code</th>\n",
       "      <td>20003.000000</td>\n",
       "      <td>2.000500e+04</td>\n",
       "      <td>20006.000000</td>\n",
       "      <td>2.000800e+04</td>\n",
       "      <td>2.000900e+04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>item_cnt_month</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>3.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>3.570000e+02</td>\n",
       "      <td>1.000000e+01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>first_cate_code</th>\n",
       "      <td>301.000000</td>\n",
       "      <td>3.060000e+02</td>\n",
       "      <td>307.000000</td>\n",
       "      <td>3.030000e+02</td>\n",
       "      <td>3.060000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>second_cate_code</th>\n",
       "      <td>405.000000</td>\n",
       "      <td>4.070000e+02</td>\n",
       "      <td>403.000000</td>\n",
       "      <td>4.010000e+02</td>\n",
       "      <td>4.070000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>item_cnt_month_lag_1</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.300000e+02</td>\n",
       "      <td>2.060000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>item_cnt_month_lag_2</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>7.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>8.100000e+01</td>\n",
       "      <td>9.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>item_cnt_month_lag_3</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.620000e+02</td>\n",
       "      <td>4.230000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_avg_item_cnt_lag_1</th>\n",
       "      <td>573.500000</td>\n",
       "      <td>5.735000e+02</td>\n",
       "      <td>573.500000</td>\n",
       "      <td>5.735000e+02</td>\n",
       "      <td>5.735000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_item_avg_item_cnt_lag_1</th>\n",
       "      <td>408.250000</td>\n",
       "      <td>1.020312e+01</td>\n",
       "      <td>81.812500</td>\n",
       "      <td>5.870000e+02</td>\n",
       "      <td>2.867500e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_item_avg_item_cnt_lag_2</th>\n",
       "      <td>304.500000</td>\n",
       "      <td>3.199219e+00</td>\n",
       "      <td>376.500000</td>\n",
       "      <td>2.042500e+02</td>\n",
       "      <td>7.218750e+01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_item_avg_item_cnt_lag_3</th>\n",
       "      <td>1060.000000</td>\n",
       "      <td>1.799805e+00</td>\n",
       "      <td>701.000000</td>\n",
       "      <td>2.392500e+02</td>\n",
       "      <td>7.445000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_region_avg_item_cnt_lag_1</th>\n",
       "      <td>772.000000</td>\n",
       "      <td>7.720000e+02</td>\n",
       "      <td>772.000000</td>\n",
       "      <td>7.720000e+02</td>\n",
       "      <td>7.720000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_region_avg_item_cnt_lag_2</th>\n",
       "      <td>366.500000</td>\n",
       "      <td>3.665000e+02</td>\n",
       "      <td>366.500000</td>\n",
       "      <td>3.665000e+02</td>\n",
       "      <td>3.665000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_region_avg_item_cnt_lag_3</th>\n",
       "      <td>379.750000</td>\n",
       "      <td>3.797500e+02</td>\n",
       "      <td>379.750000</td>\n",
       "      <td>3.797500e+02</td>\n",
       "      <td>3.797500e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_region_item_avg_item_cnt_lag_1</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.300000e+02</td>\n",
       "      <td>2.060000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_region_item_avg_item_cnt_lag_2</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>7.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>8.100000e+01</td>\n",
       "      <td>9.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_region_item_avg_item_cnt_lag_3</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.620000e+02</td>\n",
       "      <td>4.230000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>date_region_subtype_avg_item_cnt_lag_1</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.211000e+03</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>7.185000e+02</td>\n",
       "      <td>1.211000e+03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>delta_price_lag</th>\n",
       "      <td>-0.057953</td>\n",
       "      <td>3.256226e-02</td>\n",
       "      <td>0.070618</td>\n",
       "      <td>-3.829956e-02</td>\n",
       "      <td>-3.530884e-02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>delta_revenue_lag_1</th>\n",
       "      <td>389511.656250</td>\n",
       "      <td>2.354960e+07</td>\n",
       "      <td>628721.562500</td>\n",
       "      <td>5.590607e+06</td>\n",
       "      <td>2.354960e+07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>month</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000e+00</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000e+00</td>\n",
       "      <td>1.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>days</th>\n",
       "      <td>28.000000</td>\n",
       "      <td>2.800000e+01</td>\n",
       "      <td>28.000000</td>\n",
       "      <td>2.800000e+01</td>\n",
       "      <td>2.800000e+01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>item_subtype_first_sale</th>\n",
       "      <td>4.000000</td>\n",
       "      <td>3.000000e+00</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>4.000000e+00</td>\n",
       "      <td>4.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>item_first_sale</th>\n",
       "      <td>4.000000</td>\n",
       "      <td>3.000000e+00</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>4.000000e+00</td>\n",
       "      <td>4.000000e+00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                 9940          9941  \\\n",
       "date_block_num                               4.000000  4.000000e+00   \n",
       "sales_region_code                          101.000000  1.010000e+02   \n",
       "item_code                                20003.000000  2.000500e+04   \n",
       "item_cnt_month                               0.000000  3.000000e+00   \n",
       "first_cate_code                            301.000000  3.060000e+02   \n",
       "second_cate_code                           405.000000  4.070000e+02   \n",
       "item_cnt_month_lag_1                         0.000000  0.000000e+00   \n",
       "item_cnt_month_lag_2                         0.000000  7.000000e+00   \n",
       "item_cnt_month_lag_3                         0.000000  0.000000e+00   \n",
       "date_avg_item_cnt_lag_1                    573.500000  5.735000e+02   \n",
       "date_item_avg_item_cnt_lag_1               408.250000  1.020312e+01   \n",
       "date_item_avg_item_cnt_lag_2               304.500000  3.199219e+00   \n",
       "date_item_avg_item_cnt_lag_3              1060.000000  1.799805e+00   \n",
       "date_region_avg_item_cnt_lag_1             772.000000  7.720000e+02   \n",
       "date_region_avg_item_cnt_lag_2             366.500000  3.665000e+02   \n",
       "date_region_avg_item_cnt_lag_3             379.750000  3.797500e+02   \n",
       "date_region_item_avg_item_cnt_lag_1          0.000000  0.000000e+00   \n",
       "date_region_item_avg_item_cnt_lag_2          0.000000  7.000000e+00   \n",
       "date_region_item_avg_item_cnt_lag_3          0.000000  0.000000e+00   \n",
       "date_region_subtype_avg_item_cnt_lag_1       0.000000  1.211000e+03   \n",
       "delta_price_lag                             -0.057953  3.256226e-02   \n",
       "delta_revenue_lag_1                     389511.656250  2.354960e+07   \n",
       "month                                        1.000000  1.000000e+00   \n",
       "days                                        28.000000  2.800000e+01   \n",
       "item_subtype_first_sale                      4.000000  3.000000e+00   \n",
       "item_first_sale                              4.000000  3.000000e+00   \n",
       "\n",
       "                                                 9942          9943  \\\n",
       "date_block_num                               4.000000  4.000000e+00   \n",
       "sales_region_code                          101.000000  1.010000e+02   \n",
       "item_code                                20006.000000  2.000800e+04   \n",
       "item_cnt_month                               0.000000  3.570000e+02   \n",
       "first_cate_code                            307.000000  3.030000e+02   \n",
       "second_cate_code                           403.000000  4.010000e+02   \n",
       "item_cnt_month_lag_1                         0.000000  2.300000e+02   \n",
       "item_cnt_month_lag_2                         0.000000  8.100000e+01   \n",
       "item_cnt_month_lag_3                         0.000000  2.620000e+02   \n",
       "date_avg_item_cnt_lag_1                    573.500000  5.735000e+02   \n",
       "date_item_avg_item_cnt_lag_1                81.812500  5.870000e+02   \n",
       "date_item_avg_item_cnt_lag_2               376.500000  2.042500e+02   \n",
       "date_item_avg_item_cnt_lag_3               701.000000  2.392500e+02   \n",
       "date_region_avg_item_cnt_lag_1             772.000000  7.720000e+02   \n",
       "date_region_avg_item_cnt_lag_2             366.500000  3.665000e+02   \n",
       "date_region_avg_item_cnt_lag_3             379.750000  3.797500e+02   \n",
       "date_region_item_avg_item_cnt_lag_1          0.000000  2.300000e+02   \n",
       "date_region_item_avg_item_cnt_lag_2          0.000000  8.100000e+01   \n",
       "date_region_item_avg_item_cnt_lag_3          0.000000  2.620000e+02   \n",
       "date_region_subtype_avg_item_cnt_lag_1       0.000000  7.185000e+02   \n",
       "delta_price_lag                              0.070618 -3.829956e-02   \n",
       "delta_revenue_lag_1                     628721.562500  5.590607e+06   \n",
       "month                                        1.000000  1.000000e+00   \n",
       "days                                        28.000000  2.800000e+01   \n",
       "item_subtype_first_sale                      4.000000  4.000000e+00   \n",
       "item_first_sale                              4.000000  4.000000e+00   \n",
       "\n",
       "                                                9944  \n",
       "date_block_num                          4.000000e+00  \n",
       "sales_region_code                       1.010000e+02  \n",
       "item_code                               2.000900e+04  \n",
       "item_cnt_month                          1.000000e+01  \n",
       "first_cate_code                         3.060000e+02  \n",
       "second_cate_code                        4.070000e+02  \n",
       "item_cnt_month_lag_1                    2.060000e+02  \n",
       "item_cnt_month_lag_2                    9.000000e+00  \n",
       "item_cnt_month_lag_3                    4.230000e+02  \n",
       "date_avg_item_cnt_lag_1                 5.735000e+02  \n",
       "date_item_avg_item_cnt_lag_1            2.867500e+02  \n",
       "date_item_avg_item_cnt_lag_2            7.218750e+01  \n",
       "date_item_avg_item_cnt_lag_3            7.445000e+02  \n",
       "date_region_avg_item_cnt_lag_1          7.720000e+02  \n",
       "date_region_avg_item_cnt_lag_2          3.665000e+02  \n",
       "date_region_avg_item_cnt_lag_3          3.797500e+02  \n",
       "date_region_item_avg_item_cnt_lag_1     2.060000e+02  \n",
       "date_region_item_avg_item_cnt_lag_2     9.000000e+00  \n",
       "date_region_item_avg_item_cnt_lag_3     4.230000e+02  \n",
       "date_region_subtype_avg_item_cnt_lag_1  1.211000e+03  \n",
       "delta_price_lag                        -3.530884e-02  \n",
       "delta_revenue_lag_1                     2.354960e+07  \n",
       "month                                   1.000000e+00  \n",
       "days                                    2.800000e+01  \n",
       "item_subtype_first_sale                 4.000000e+00  \n",
       "item_first_sale                         4.000000e+00  "
      ]
     },
     "execution_count": 189,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "matrix.head().T"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 190,
   "id": "d3e4b594",
   "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>date_block_num</th>\n",
       "      <th>sales_region_code</th>\n",
       "      <th>item_code</th>\n",
       "      <th>item_cnt_month</th>\n",
       "      <th>first_cate_code</th>\n",
       "      <th>second_cate_code</th>\n",
       "      <th>item_cnt_month_lag_1</th>\n",
       "      <th>item_cnt_month_lag_2</th>\n",
       "      <th>item_cnt_month_lag_3</th>\n",
       "      <th>date_avg_item_cnt_lag_1</th>\n",
       "      <th>...</th>\n",
       "      <th>date_region_item_avg_item_cnt_lag_1</th>\n",
       "      <th>date_region_item_avg_item_cnt_lag_2</th>\n",
       "      <th>date_region_item_avg_item_cnt_lag_3</th>\n",
       "      <th>date_region_subtype_avg_item_cnt_lag_1</th>\n",
       "      <th>delta_price_lag</th>\n",
       "      <th>delta_revenue_lag_1</th>\n",
       "      <th>month</th>\n",
       "      <th>days</th>\n",
       "      <th>item_subtype_first_sale</th>\n",
       "      <th>item_first_sale</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>9940</th>\n",
       "      <td>4</td>\n",
       "      <td>101</td>\n",
       "      <td>20003</td>\n",
       "      <td>0.0</td>\n",
       "      <td>301</td>\n",
       "      <td>405</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>573.50</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000</td>\n",
       "      <td>-0.057953</td>\n",
       "      <td>3.895117e+05</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9941</th>\n",
       "      <td>4</td>\n",
       "      <td>101</td>\n",
       "      <td>20005</td>\n",
       "      <td>3.0</td>\n",
       "      <td>306</td>\n",
       "      <td>407</td>\n",
       "      <td>0.0</td>\n",
       "      <td>7.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>573.50</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>7.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1211.000</td>\n",
       "      <td>0.032562</td>\n",
       "      <td>2.354960e+07</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>3</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9942</th>\n",
       "      <td>4</td>\n",
       "      <td>101</td>\n",
       "      <td>20006</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>573.50</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.070618</td>\n",
       "      <td>6.287216e+05</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9943</th>\n",
       "      <td>4</td>\n",
       "      <td>101</td>\n",
       "      <td>20008</td>\n",
       "      <td>357.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "      <td>230.0</td>\n",
       "      <td>81.0</td>\n",
       "      <td>262.0</td>\n",
       "      <td>573.50</td>\n",
       "      <td>...</td>\n",
       "      <td>230.0</td>\n",
       "      <td>81.0</td>\n",
       "      <td>262.0</td>\n",
       "      <td>718.500</td>\n",
       "      <td>-0.038300</td>\n",
       "      <td>5.590607e+06</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9944</th>\n",
       "      <td>4</td>\n",
       "      <td>101</td>\n",
       "      <td>20009</td>\n",
       "      <td>10.0</td>\n",
       "      <td>306</td>\n",
       "      <td>407</td>\n",
       "      <td>206.0</td>\n",
       "      <td>9.0</td>\n",
       "      <td>423.0</td>\n",
       "      <td>573.50</td>\n",
       "      <td>...</td>\n",
       "      <td>206.0</td>\n",
       "      <td>9.0</td>\n",
       "      <td>423.0</td>\n",
       "      <td>1211.000</td>\n",
       "      <td>-0.035309</td>\n",
       "      <td>2.354960e+07</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110315</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22066</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>7224.0</td>\n",
       "      <td>10536.0</td>\n",
       "      <td>5556.0</td>\n",
       "      <td>300.75</td>\n",
       "      <td>...</td>\n",
       "      <td>7224.0</td>\n",
       "      <td>10536.0</td>\n",
       "      <td>5556.0</td>\n",
       "      <td>433.750</td>\n",
       "      <td>0.091797</td>\n",
       "      <td>3.829482e+06</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>24</td>\n",
       "      <td>24</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110316</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22072</td>\n",
       "      <td>0.0</td>\n",
       "      <td>305</td>\n",
       "      <td>412</td>\n",
       "      <td>488.0</td>\n",
       "      <td>1842.0</td>\n",
       "      <td>1082.0</td>\n",
       "      <td>300.75</td>\n",
       "      <td>...</td>\n",
       "      <td>488.0</td>\n",
       "      <td>1842.0</td>\n",
       "      <td>1082.0</td>\n",
       "      <td>278.250</td>\n",
       "      <td>-0.006069</td>\n",
       "      <td>2.064940e+06</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110317</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22075</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307</td>\n",
       "      <td>403</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110318</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22083</td>\n",
       "      <td>0.0</td>\n",
       "      <td>303</td>\n",
       "      <td>401</td>\n",
       "      <td>1124.0</td>\n",
       "      <td>1344.0</td>\n",
       "      <td>7420.0</td>\n",
       "      <td>300.75</td>\n",
       "      <td>...</td>\n",
       "      <td>1124.0</td>\n",
       "      <td>1344.0</td>\n",
       "      <td>7420.0</td>\n",
       "      <td>217.375</td>\n",
       "      <td>-0.006493</td>\n",
       "      <td>3.506226e+06</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>6</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>110319</th>\n",
       "      <td>40</td>\n",
       "      <td>105</td>\n",
       "      <td>22084</td>\n",
       "      <td>0.0</td>\n",
       "      <td>302</td>\n",
       "      <td>408</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>11.0</td>\n",
       "      <td>300.75</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>11.0</td>\n",
       "      <td>197.125</td>\n",
       "      <td>-0.010872</td>\n",
       "      <td>6.008424e+06</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>24</td>\n",
       "      <td>24</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>100380 rows × 26 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        date_block_num  sales_region_code  item_code  item_cnt_month  \\\n",
       "9940                 4                101      20003             0.0   \n",
       "9941                 4                101      20005             3.0   \n",
       "9942                 4                101      20006             0.0   \n",
       "9943                 4                101      20008           357.0   \n",
       "9944                 4                101      20009            10.0   \n",
       "...                ...                ...        ...             ...   \n",
       "110315              40                105      22066             0.0   \n",
       "110316              40                105      22072             0.0   \n",
       "110317              40                105      22075             0.0   \n",
       "110318              40                105      22083             0.0   \n",
       "110319              40                105      22084             0.0   \n",
       "\n",
       "        first_cate_code  second_cate_code  item_cnt_month_lag_1  \\\n",
       "9940                301               405                   0.0   \n",
       "9941                306               407                   0.0   \n",
       "9942                307               403                   0.0   \n",
       "9943                303               401                 230.0   \n",
       "9944                306               407                 206.0   \n",
       "...                 ...               ...                   ...   \n",
       "110315              307               403                7224.0   \n",
       "110316              305               412                 488.0   \n",
       "110317              307               403                   0.0   \n",
       "110318              303               401                1124.0   \n",
       "110319              302               408                   0.0   \n",
       "\n",
       "        item_cnt_month_lag_2  item_cnt_month_lag_3  date_avg_item_cnt_lag_1  \\\n",
       "9940                     0.0                   0.0                   573.50   \n",
       "9941                     7.0                   0.0                   573.50   \n",
       "9942                     0.0                   0.0                   573.50   \n",
       "9943                    81.0                 262.0                   573.50   \n",
       "9944                     9.0                 423.0                   573.50   \n",
       "...                      ...                   ...                      ...   \n",
       "110315               10536.0                5556.0                   300.75   \n",
       "110316                1842.0                1082.0                   300.75   \n",
       "110317                   0.0                   0.0                     0.00   \n",
       "110318                1344.0                7420.0                   300.75   \n",
       "110319                   0.0                  11.0                   300.75   \n",
       "\n",
       "        ...  date_region_item_avg_item_cnt_lag_1  \\\n",
       "9940    ...                                  0.0   \n",
       "9941    ...                                  0.0   \n",
       "9942    ...                                  0.0   \n",
       "9943    ...                                230.0   \n",
       "9944    ...                                206.0   \n",
       "...     ...                                  ...   \n",
       "110315  ...                               7224.0   \n",
       "110316  ...                                488.0   \n",
       "110317  ...                                  0.0   \n",
       "110318  ...                               1124.0   \n",
       "110319  ...                                  0.0   \n",
       "\n",
       "        date_region_item_avg_item_cnt_lag_2  \\\n",
       "9940                                    0.0   \n",
       "9941                                    7.0   \n",
       "9942                                    0.0   \n",
       "9943                                   81.0   \n",
       "9944                                    9.0   \n",
       "...                                     ...   \n",
       "110315                              10536.0   \n",
       "110316                               1842.0   \n",
       "110317                                  0.0   \n",
       "110318                               1344.0   \n",
       "110319                                  0.0   \n",
       "\n",
       "        date_region_item_avg_item_cnt_lag_3  \\\n",
       "9940                                    0.0   \n",
       "9941                                    0.0   \n",
       "9942                                    0.0   \n",
       "9943                                  262.0   \n",
       "9944                                  423.0   \n",
       "...                                     ...   \n",
       "110315                               5556.0   \n",
       "110316                               1082.0   \n",
       "110317                                  0.0   \n",
       "110318                               7420.0   \n",
       "110319                                 11.0   \n",
       "\n",
       "        date_region_subtype_avg_item_cnt_lag_1  delta_price_lag  \\\n",
       "9940                                     0.000        -0.057953   \n",
       "9941                                  1211.000         0.032562   \n",
       "9942                                     0.000         0.070618   \n",
       "9943                                   718.500        -0.038300   \n",
       "9944                                  1211.000        -0.035309   \n",
       "...                                        ...              ...   \n",
       "110315                                 433.750         0.091797   \n",
       "110316                                 278.250        -0.006069   \n",
       "110317                                   0.000         0.000000   \n",
       "110318                                 217.375        -0.006493   \n",
       "110319                                 197.125        -0.010872   \n",
       "\n",
       "        delta_revenue_lag_1  month  days  item_subtype_first_sale  \\\n",
       "9940           3.895117e+05      1    28                        4   \n",
       "9941           2.354960e+07      1    28                        3   \n",
       "9942           6.287216e+05      1    28                        4   \n",
       "9943           5.590607e+06      1    28                        4   \n",
       "9944           2.354960e+07      1    28                        4   \n",
       "...                     ...    ...   ...                      ...   \n",
       "110315         3.829482e+06      1    28                       24   \n",
       "110316         2.064940e+06      1    28                        4   \n",
       "110317         0.000000e+00      1    28                        0   \n",
       "110318         3.506226e+06      1    28                        6   \n",
       "110319         6.008424e+06      1    28                       24   \n",
       "\n",
       "        item_first_sale  \n",
       "9940                  4  \n",
       "9941                  3  \n",
       "9942                  4  \n",
       "9943                  4  \n",
       "9944                  4  \n",
       "...                 ...  \n",
       "110315               24  \n",
       "110316                4  \n",
       "110317                0  \n",
       "110318                6  \n",
       "110319               24  \n",
       "\n",
       "[100380 rows x 26 columns]"
      ]
     },
     "execution_count": 190,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "matrix.fillna(0)  #需要吗"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 191,
   "id": "f3355c43",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 补充一下编码\n",
    "# itemcode\n",
    "# LE2 = LabelEncoder()\n",
    "# matrix[\"item_code\"] = LE2.fit_transform(matrix.item_code )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 192,
   "id": "96f46293",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "matrix.to_csv(\"./中间数据/月度数据无编码.csv\",index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3771a8b8",
   "metadata": {},
   "source": [
    "# 3.Modelling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 193,
   "id": "dcf8e035",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "import gc\n",
    "import pickle\n",
    "from xgboost import XGBRegressor\n",
    "# import lightgbm as lgb\n",
    "from matplotlib.pylab import rcParams\n",
    "rcParams['figure.figsize'] = 12, 4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 194,
   "id": "485758c5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "30"
      ]
     },
     "execution_count": 194,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = matrix.copy()  # 使用copy，不要直接=\n",
    "del matrix\n",
    "gc.collect() # 释放内存"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 195,
   "id": "5e364c10",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "9940       4\n",
       "9941       4\n",
       "9942       4\n",
       "9943       4\n",
       "9944       4\n",
       "          ..\n",
       "110315    40\n",
       "110316    40\n",
       "110317    40\n",
       "110318    40\n",
       "110319    40\n",
       "Name: date_block_num, Length: 100380, dtype: int8"
      ]
     },
     "execution_count": 195,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.date_block_num"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cf9f2d7d",
   "metadata": {},
   "source": [
    "Use month 38、39 as validation for training."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6c9be1bb",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8eb34064",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 196,
   "id": "59b2a958",
   "metadata": {},
   "outputs": [],
   "source": [
    "valid = 38  # 最终这里是38，前期可以是37\n",
    "# 38 我们检查的就是验证集，37检查的就是验证集或者测试集\n",
    "\n",
    "X_train = data[data.date_block_num < valid].drop(['item_cnt_month'], axis=1)\n",
    "Y_train = data[data.date_block_num < valid]['item_cnt_month']\n",
    "X_valid = data[data.date_block_num.isin([valid,valid+1])].drop(['item_cnt_month'], axis=1)\n",
    "Y_valid = data[data.date_block_num.isin([valid,valid+1])]['item_cnt_month']\n",
    "X_test = data[data.date_block_num == valid+2].drop(['item_cnt_month'], axis=1)\n",
    "\n",
    "\n",
    "# 最后是要剔除的\n",
    "Y_test = data[data.date_block_num == valid+2]['item_cnt_month']\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 197,
   "id": "9562456a",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Y_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "id": "0845bb4b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot: xlabel='item_cnt_month'>"
      ]
     },
     "execution_count": 198,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 测一下数据分布，不知道这里是否需要归一化\n",
    "plt.figure(figsize=(10,4))\n",
    "plt.xlim(-100, 3000)\n",
    "flierprops = dict(marker='o', markerfacecolor='purple', markersize=6,\n",
    "                  linestyle='none', markeredgecolor='black')\n",
    "sns.boxplot(x=Y_train, flierprops=flierprops)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 199,
   "id": "c742c107",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 划分完就删删删\n",
    "del data\n",
    "gc.collect();"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "530ea20d",
   "metadata": {},
   "source": [
    "lightGBM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 200,
   "id": "b180cfd7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.22567391395568848"
      ]
     },
     "execution_count": 200,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# lightGBM\n",
    "\n",
    "from lightgbm import LGBMRegressor\n",
    "ts = time.time()\n",
    "learning_rate=0.034 \n",
    "max_depth=66 \n",
    "num_leaves=200 \n",
    "model = LGBMRegressor(learning_rate=learning_rate,\n",
    "                      objective = 'tweedie',\n",
    "                  max_depth=max_depth,\n",
    "                  num_leaves=num_leaves,\n",
    "                  n_jobs=-1,\n",
    "                  n_estimators=100)\n",
    " \n",
    "model.fit(X_train,Y_train)\n",
    "\n",
    "time.time() - ts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "27e988cc",
   "metadata": {},
   "source": [
    "XGBOOST"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "09043138",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 201,
   "id": "19f91eac",
   "metadata": {},
   "outputs": [],
   "source": [
    "Y_valid_pred = model.predict(X_valid) # 验证集的预测y\n",
    "Y_test_pred = model.predict(X_test) # 预测集的预测y\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 202,
   "id": "10d92fb2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 这里取决于我们要检查的是预测集还是验证集的效果\n",
    "\n",
    "# 如果是验证集\n",
    "Y_pred = Y_valid_pred\n",
    "Y_valid = Y_valid\n",
    "\n",
    "# 如果是测试集\n",
    "# X_valid = X_test\n",
    "# Y_pred = Y_test_pred\n",
    "# Y_valid = Y_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 203,
   "id": "4461e7d5",
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "submission = pd.DataFrame({\n",
    "#     \"ID\": Y_valid.index, \n",
    "    \"date_block_num\":X_valid.date_block_num,\n",
    "    \"sales_region_code\":X_valid.sales_region_code,\n",
    "    \"item_code\":X_valid.item_code,\n",
    "    \"first_cate_code\":X_valid.first_cate_code,\n",
    "    \"second_cate_code\":X_valid.second_cate_code,\n",
    "    \"item_cnt_month_valid\": Y_valid,\n",
    "    \"item_cnt_month_pred\": Y_pred\n",
    "})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 204,
   "id": "e34250de",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "# 剔除所有item_cnt_month_valid为0的\n",
    "submission = submission[submission[\"item_cnt_month_valid\"] != 0]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 205,
   "id": "12e9fd30",
   "metadata": {},
   "outputs": [],
   "source": [
    "# # 还原编码\n",
    "# submission[\"sales_region_code\"] = LE1.inverse_transform(submission[\"sales_region_code\"])\n",
    "# submission[\"first_cate_code\"] = LE3.inverse_transform(submission[\"first_cate_code\"])\n",
    "# submission[\"second_cate_code\"] = LE4.inverse_transform(submission[\"second_cate_code\"])\n",
    "# submission[\"item_code\"] = LE2.inverse_transform(submission[\"item_code\"])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 206,
   "id": "b71e818d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 把对数转换回来\n",
    "if is_need_log:\n",
    "    submission[\"item_cnt_month_valid\"] = list(map(lambda x : 2**x-1,submission[\"item_cnt_month_valid\"] ))\n",
    "    submission[\"item_cnt_month_pred\"] = list(map(lambda x : 2**x-1,submission[\"item_cnt_month_pred\"] ))\n",
    "# 填补负值\n",
    "submission[\"item_cnt_month_pred\"] =list(map(lambda x : x if x>0 else 0.1,submission[\"item_cnt_month_pred\"] ))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 207,
   "id": "6821b3ef",
   "metadata": {},
   "outputs": [],
   "source": [
    "submission.to_csv('lgbm_submission.csv', index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fe2e87d1",
   "metadata": {},
   "source": [
    "用tweedie指标评价"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 208,
   "id": "be677969",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tweedie deviance: 4852890700180273.0\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import mean_tweedie_deviance\n",
    "\n",
    "# 计算模型在两列数据上的Tweedie偏差\n",
    "tweedie_variance_power = 1.5\n",
    "tweedie_dev = mean_tweedie_deviance(submission[\"item_cnt_month_valid\"], submission[\"item_cnt_month_pred\"], power=tweedie_variance_power)\n",
    "    \n",
    "# 打印Tweedie偏差值\n",
    "print(\"Tweedie deviance:\", tweedie_dev)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 209,
   "id": "a8b2d5e5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(91237, 25)"
      ]
     },
     "execution_count": 209,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 210,
   "id": "d5336444",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot: title={'center': 'Feature importance'}, xlabel='Feature importance', ylabel='Features'>"
      ]
     },
     "execution_count": 210,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x1008 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from lightgbm import plot_importance\n",
    "\n",
    "def plot_features(booster, figsize):    \n",
    "    fig, ax = plt.subplots(1,1,figsize=figsize)\n",
    "    return plot_importance(booster=booster, ax=ax)\n",
    "\n",
    "plot_features(model, (10,14))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b5f06b78",
   "metadata": {},
   "source": [
    "最后其实是没有这步的，前期我们检验一下有没有过拟合"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 211,
   "id": "4d3085c6",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Y_test_pred =  model.predict(X_test)\n",
    "\n",
    "# # 计算模型在两列数据上的Tweedie偏差\n",
    "# tweedie_variance_power = 1\n",
    "# tweedie_dev = mean_tweedie_deviance(Y_test_pred, Y_test, power=tweedie_variance_power)\n",
    "\n",
    "# # 打印Tweedie偏差值\n",
    "# print(\"Tweedie deviance:\", tweedie_dev)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "33551347",
   "metadata": {},
   "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}