{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inżynieria cech"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "<b>Feature Engineering</b> polega na przekształcaniu surowych danych w cechy użyteczne w procesie modelowania. \n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-info\">\n",
    "<b>Info</b> \n",
    "\n",
    "Tworzenie zmiennych predykcyjnych zawsze powinno opierać się przede wszystkim na zrozumieniu danych i wiedzy biznesowej związanej z danym problemem. Istnieje jednak kilka standardowych technik, które można stosować do tworzenia cech w większości problemów, można tu wyróżnić:\n",
    "- przekształcanie zmiennych kategorycznych\n",
    "- zmienne na podstawie cech z daty\n",
    "- zmienne powstałe przez przesunięcie w czasie\n",
    "- transformacje zmiennych numerycznych\n",
    "- zmienne interakcji\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-danger\">\n",
    "<b>Uwaga!</b> \n",
    "    \n",
    "<b>Feature Engineering</b> stanowi jeden z ważniejszych etapów w procesie budowy uczenia maszynowego, dlatego należy mu poświęcić dużo uwagi. Często można uzyskać znacznie lepsze wyniki mając prosty model oparty na cechach predykcyjnych dobrze oddających naturę badanego zjawiska niż budując wyrafinowany model w oparciu o zbyt wąski zbiór zmiennych.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tworzenie przykładowych zbiorów danych"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Zaprezentujemy tutaj 2 przykładowe zbiory danych o różnym charakterze oraz przykłady zmiennych predykcyjnych, które można wygenerować na ich podstawie"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:51.187621Z",
     "start_time": "2021-10-13T07:15:40.687595Z"
    }
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:51.344602Z",
     "start_time": "2021-10-13T07:15:51.332608Z"
    }
   },
   "outputs": [],
   "source": [
    "def generateIceCreamSalesData(month_coef=10000, week_coef=1000, day_coef=100, random_coef =10000):\n",
    "    dates = pd.date_range(start=\"2018-01-01\", end=\"2020-12-31\", freq =\"D\")\n",
    "    df = pd.DataFrame(dates, columns=[\"SalesDate\"])\n",
    "    df[\"Month\"] = df.SalesDate.dt.month\n",
    "    df[\"Week\"] = df.SalesDate.dt.isocalendar().week\n",
    "    df[\"WeekDay\"] = df.SalesDate.dt.dayofweek+1\n",
    "    df[\"IceCreamSales\"] = (-1*np.power(df.Month-6,2)+np.power(6,2))*month_coef + (-1*np.power(df.Week-27,2)+np.power(27,2))*week_coef +df.WeekDay*day_coef +random_coef * np.random.randint(low =1,high =10, size=len(df))\n",
    "    return df.loc[:,[\"SalesDate\",\"IceCreamSales\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:51.668480Z",
     "start_time": "2021-10-13T07:15:51.471482Z"
    }
   },
   "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>SalesDate</th>\n",
       "      <th>IceCreamSales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2018-01-01</td>\n",
       "      <td>253100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2018-01-02</td>\n",
       "      <td>223200</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2018-01-03</td>\n",
       "      <td>173300</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2018-01-04</td>\n",
       "      <td>223400</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2018-01-05</td>\n",
       "      <td>223500</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1091</th>\n",
       "      <td>2020-12-27</td>\n",
       "      <td>134700</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1092</th>\n",
       "      <td>2020-12-28</td>\n",
       "      <td>143100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1093</th>\n",
       "      <td>2020-12-29</td>\n",
       "      <td>93200</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1094</th>\n",
       "      <td>2020-12-30</td>\n",
       "      <td>73300</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1095</th>\n",
       "      <td>2020-12-31</td>\n",
       "      <td>133400</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1096 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      SalesDate  IceCreamSales\n",
       "0    2018-01-01         253100\n",
       "1    2018-01-02         223200\n",
       "2    2018-01-03         173300\n",
       "3    2018-01-04         223400\n",
       "4    2018-01-05         223500\n",
       "...         ...            ...\n",
       "1091 2020-12-27         134700\n",
       "1092 2020-12-28         143100\n",
       "1093 2020-12-29          93200\n",
       "1094 2020-12-30          73300\n",
       "1095 2020-12-31         133400\n",
       "\n",
       "[1096 rows x 2 columns]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "IceCream_df = generateIceCreamSalesData()\n",
    "IceCream_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:51.898539Z",
     "start_time": "2021-10-13T07:15:51.872481Z"
    }
   },
   "outputs": [],
   "source": [
    "def generate_used_cars_data(len_df =1000):\n",
    "    conditions ={\"very_bad\":1,\"bad\":2,\"medium\":3,\"good\":4,\"very_good\":5}\n",
    "    brands =[\"Fiat\",\"Renault\",\"VW\", \"Seat\", \"Skoda\",\"Toyota\", \"Audi\",\"BMW\",\"Mercedes\", \"Bugatti\"]\n",
    "    standard_brands = [\"Fiat\",\"Renault\",\"VW\", \"Seat\", \"Skoda\",\"Toyota\"]\n",
    "    premium_brands = [\"Audi\",\"BMW\",\"Mercedes\"]\n",
    "    luxury_brands = [\"Bugatti\"]\n",
    "    dict_data ={\"condition\":np.random.choice(list(conditions.keys()),size=len_df, p = [0.05, 0.15, 0.3, 0.3,0.2 ]),\n",
    "                \"brand\":np.random.choice(brands,size=len_df, p=[0.1]*10 ),\n",
    "                \"year_manufactured\":np.random.randint(1950,2020,size =len_df)\n",
    "               }\n",
    "    df = pd.DataFrame(dict_data)\n",
    "    df[\"age\"] =2021- df.year_manufactured\n",
    "    df[\"mileage\"] = df.age *np.random.randint(100,10000, len_df)+np.random.randint(100,10000, len_df)\n",
    "    \n",
    "    df[\"selling_price\"] = 100000*df[\"brand\"].isin(standard_brands)+300000*df[\"brand\"].isin(premium_brands)+600000*df[\"brand\"].isin(luxury_brands)\n",
    "    df[\"selling_price\"] /= np.log1p(df.age)+np.log1p(df.mileage)\n",
    "    df[\"condition_num\"] = df.condition.map(lambda x:conditions[x])\n",
    "    df[\"selling_price\"] *=  np.log1p(df[\"condition_num\"])\n",
    "    df.loc[(df[\"brand\"].isin(luxury_brands))&(df.year_manufactured<=1970)&(df[\"condition_num\"]>3),\"selling_price\"] *=\\\n",
    "    np.log1p(df.loc[(df[\"brand\"].isin(luxury_brands))&(df.year_manufactured<=1970)&(df[\"condition_num\"]>3),\"age\"])\n",
    "    df[\"selling_price\"] = np.round(df[\"selling_price\"])\n",
    "    return df.loc[:,[\"selling_price\",\"condition\",\"mileage\",\"brand\",\"year_manufactured\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:52.225195Z",
     "start_time": "2021-10-13T07:15:52.139106Z"
    }
   },
   "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>selling_price</th>\n",
       "      <th>condition</th>\n",
       "      <th>mileage</th>\n",
       "      <th>brand</th>\n",
       "      <th>year_manufactured</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>9847.0</td>\n",
       "      <td>good</td>\n",
       "      <td>338883</td>\n",
       "      <td>Fiat</td>\n",
       "      <td>1985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>7899.0</td>\n",
       "      <td>medium</td>\n",
       "      <td>589817</td>\n",
       "      <td>Renault</td>\n",
       "      <td>1951</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>11231.0</td>\n",
       "      <td>good</td>\n",
       "      <td>79656</td>\n",
       "      <td>Toyota</td>\n",
       "      <td>2001</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>33941.0</td>\n",
       "      <td>good</td>\n",
       "      <td>115919</td>\n",
       "      <td>Mercedes</td>\n",
       "      <td>2009</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>38360.0</td>\n",
       "      <td>very_good</td>\n",
       "      <td>64115</td>\n",
       "      <td>Mercedes</td>\n",
       "      <td>2003</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>995</th>\n",
       "      <td>10204.0</td>\n",
       "      <td>medium</td>\n",
       "      <td>29430</td>\n",
       "      <td>Renault</td>\n",
       "      <td>1995</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>996</th>\n",
       "      <td>36737.0</td>\n",
       "      <td>very_good</td>\n",
       "      <td>119063</td>\n",
       "      <td>Audi</td>\n",
       "      <td>2003</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>997</th>\n",
       "      <td>8692.0</td>\n",
       "      <td>bad</td>\n",
       "      <td>14683</td>\n",
       "      <td>Seat</td>\n",
       "      <td>2001</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>998</th>\n",
       "      <td>16719.0</td>\n",
       "      <td>very_good</td>\n",
       "      <td>15039</td>\n",
       "      <td>Renault</td>\n",
       "      <td>2019</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>999</th>\n",
       "      <td>39886.0</td>\n",
       "      <td>very_good</td>\n",
       "      <td>44529</td>\n",
       "      <td>BMW</td>\n",
       "      <td>2006</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1000 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     selling_price  condition  mileage     brand  year_manufactured\n",
       "0           9847.0       good   338883      Fiat               1985\n",
       "1           7899.0     medium   589817   Renault               1951\n",
       "2          11231.0       good    79656    Toyota               2001\n",
       "3          33941.0       good   115919  Mercedes               2009\n",
       "4          38360.0  very_good    64115  Mercedes               2003\n",
       "..             ...        ...      ...       ...                ...\n",
       "995        10204.0     medium    29430   Renault               1995\n",
       "996        36737.0  very_good   119063      Audi               2003\n",
       "997         8692.0        bad    14683      Seat               2001\n",
       "998        16719.0  very_good    15039   Renault               2019\n",
       "999        39886.0  very_good    44529       BMW               2006\n",
       "\n",
       "[1000 rows x 5 columns]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "UsedCars_df =generate_used_cars_data()\n",
    "UsedCars_df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Cechy tworzone na podstawie danych czasowych"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Pierwszy zbiór danych nazwany tutaj IceCream_df de facto ma charakter szeregu czasowego i patrząc na poniższy wykres mógłby być z powodzeniem przewidywany dedykowanymi metodami do predykcji szeregów czasowych, jednak dla celów pokazania tworzenia zmiennych na podstawie danych o charakterze czasowym potraktujemy to zagadnienie jako problem regresyjny, w którym naszą zmienną celu będzie IceCreamSales."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:53.756282Z",
     "start_time": "2021-10-13T07:15:52.570422Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:title={'center':'IceCream Sales'}, xlabel='SalesDate'>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1080x648 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "IceCream_df.set_index(\"SalesDate\").plot(figsize=(15,9), title=\"IceCream Sales\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wyraźnie widać tutaj, że sprzedaż lodów ma silnie sezonowy charakter, stąd cechy wyciągnięte z daty powinny znacznie ułatwić dobrą predykcję. Funkcję generującą takie cechy zaprezentowano poniżej:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:54.203412Z",
     "start_time": "2021-10-13T07:15:54.184414Z"
    }
   },
   "outputs": [],
   "source": [
    "def extract_date_features(df, date_column):\n",
    "    df_temp = df.copy()\n",
    "    df_temp[date_column + 'Quarter'] = df_temp[date_column].dt.quarter\n",
    "    df_temp[date_column + 'Month'] = df_temp[date_column].dt.month\n",
    "    df_temp[date_column + 'Week'] = df_temp[date_column].dt.isocalendar().week\n",
    "    # poniżej zwracany rozkład to 0-6, dodajemy 1 aby przejsc na bardziej intuicyjne wrtości 1-7\n",
    "    df_temp[date_column + 'WeekDay'] = df_temp[date_column].dt.dayofweek + 1 \n",
    "    df_temp[date_column + 'YearDay'] = df_temp[date_column].dt.dayofyear\n",
    "    df_temp[date_column + 'isWeekend'] = np.where(df_temp[date_column + 'WeekDay']>5,1,0)\n",
    "    return df_temp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:54.836613Z",
     "start_time": "2021-10-13T07:15:54.737426Z"
    }
   },
   "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>SalesDate</th>\n",
       "      <th>IceCreamSales</th>\n",
       "      <th>SalesDateQuarter</th>\n",
       "      <th>SalesDateMonth</th>\n",
       "      <th>SalesDateWeek</th>\n",
       "      <th>SalesDateWeekDay</th>\n",
       "      <th>SalesDateYearDay</th>\n",
       "      <th>SalesDateisWeekend</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2018-01-01</td>\n",
       "      <td>253100</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2018-01-02</td>\n",
       "      <td>223200</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2018-01-03</td>\n",
       "      <td>173300</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2018-01-04</td>\n",
       "      <td>223400</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2018-01-05</td>\n",
       "      <td>223500</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>5</td>\n",
       "      <td>5</td>\n",
       "      <td>0</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1091</th>\n",
       "      <td>2020-12-27</td>\n",
       "      <td>134700</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>52</td>\n",
       "      <td>7</td>\n",
       "      <td>362</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1092</th>\n",
       "      <td>2020-12-28</td>\n",
       "      <td>143100</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>53</td>\n",
       "      <td>1</td>\n",
       "      <td>363</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1093</th>\n",
       "      <td>2020-12-29</td>\n",
       "      <td>93200</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>53</td>\n",
       "      <td>2</td>\n",
       "      <td>364</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1094</th>\n",
       "      <td>2020-12-30</td>\n",
       "      <td>73300</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>53</td>\n",
       "      <td>3</td>\n",
       "      <td>365</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1095</th>\n",
       "      <td>2020-12-31</td>\n",
       "      <td>133400</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>53</td>\n",
       "      <td>4</td>\n",
       "      <td>366</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1096 rows × 8 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      SalesDate  IceCreamSales  SalesDateQuarter  SalesDateMonth  \\\n",
       "0    2018-01-01         253100                 1               1   \n",
       "1    2018-01-02         223200                 1               1   \n",
       "2    2018-01-03         173300                 1               1   \n",
       "3    2018-01-04         223400                 1               1   \n",
       "4    2018-01-05         223500                 1               1   \n",
       "...         ...            ...               ...             ...   \n",
       "1091 2020-12-27         134700                 4              12   \n",
       "1092 2020-12-28         143100                 4              12   \n",
       "1093 2020-12-29          93200                 4              12   \n",
       "1094 2020-12-30          73300                 4              12   \n",
       "1095 2020-12-31         133400                 4              12   \n",
       "\n",
       "      SalesDateWeek  SalesDateWeekDay  SalesDateYearDay  SalesDateisWeekend  \n",
       "0                 1                 1                 1                   0  \n",
       "1                 1                 2                 2                   0  \n",
       "2                 1                 3                 3                   0  \n",
       "3                 1                 4                 4                   0  \n",
       "4                 1                 5                 5                   0  \n",
       "...             ...               ...               ...                 ...  \n",
       "1091             52                 7               362                   1  \n",
       "1092             53                 1               363                   0  \n",
       "1093             53                 2               364                   0  \n",
       "1094             53                 3               365                   0  \n",
       "1095             53                 4               366                   0  \n",
       "\n",
       "[1096 rows x 8 columns]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "IceCream_df_extended = extract_date_features(IceCream_df,\"SalesDate\")\n",
    "IceCream_df_extended"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Cechy powstałe przez przesunięcie w czasie"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Analizując wykres można zaobserwować, że sprzedaż lodów nie jest liniowo zależna od zmiennych takich jak numer kwartału, numer miesiąca czy numer tygodnia, aby uzyskać lepsze wyniki warto zbudować cechy w oparciu o przesunięte w czasie wartości zmiennej celu."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-danger\">\n",
    "<b>Uwaga!</b> \n",
    "    \n",
    "Przy korzystaniu ze zmiennych przesuniętych w czasie, w szczególności opartych o wcześniejsze wartości zmiennej celu nalezy zawsze wziąc pod uwagę jakie dane będziemy mieli dostępne na moment predykcji, inaczej możemy popełnić jeden z najczęstszych błędów czyli <b>data leakage</b>. \n",
    "Przy założeniu, że mamy przewidywać sprzedaż lodów np na rok do przodu wykorzystanie sprzedaży lodów z dnia poprzedniego stanowi data leakage, ponieważ ta informacja nie będzie dostępna na moment predykcji w tej samej formie. Moglibyśmy natomiast uwzględnić sprzedaż lodów sprzed roku, jako, że ta informacja będzie dostępna w tej samej formie.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tutaj dla uproszczenia zakładamy chwilowo, że horyzont predykcji to tylko 1 dzień do przodu, co pozwoli zaprezentować większy zakres zmiennych przesuniętych w czasie."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:57.805667Z",
     "start_time": "2021-10-13T07:15:57.788095Z"
    }
   },
   "outputs": [],
   "source": [
    "def get_shifted_target_values(df, lag_values, date_column, target_column):\n",
    "    df_temp = df.copy()\n",
    "    df_temp =df_temp.sort_values(by=date_column)\n",
    "    for lag in lag_values:\n",
    "        df_temp[target_column +\"_lagged_\" + str(lag)] = df_temp[target_column].shift(lag)\n",
    "    return df_temp"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Uwzględniając, że przewidujemy z horyzontem czasowym tylko na dzień w przód, pierwszym kandydatem na wartość przesunięcia jest 1 dzień, jako, że w danych o charakterze szeregu czasowego podobieństwo kolejnych obserwacji będzie siłą rzeczy relatywnie wysokie. Z analizy wykresu można łatwo wywnioskować, że wartośc przesunięcia 365 dni również będzie miała dużą siłę predykcyjną. Ponadto można tutaj wypróbować 7 dni co powinno być przydatne w przypadku tygodniowej sezonowości danych"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:15:59.183156Z",
     "start_time": "2021-10-13T07:15:59.131159Z"
    }
   },
   "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>SalesDate</th>\n",
       "      <th>IceCreamSales</th>\n",
       "      <th>SalesDateQuarter</th>\n",
       "      <th>SalesDateMonth</th>\n",
       "      <th>SalesDateWeek</th>\n",
       "      <th>SalesDateWeekDay</th>\n",
       "      <th>SalesDateYearDay</th>\n",
       "      <th>SalesDateisWeekend</th>\n",
       "      <th>IceCreamSales_lagged_1</th>\n",
       "      <th>IceCreamSales_lagged_7</th>\n",
       "      <th>IceCreamSales_lagged_365</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2018-01-01</td>\n",
       "      <td>253100</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2018-01-02</td>\n",
       "      <td>213200</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>253100</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2018-01-03</td>\n",
       "      <td>193300</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>213200</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2018-01-04</td>\n",
       "      <td>253400</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>193300</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2018-01-05</td>\n",
       "      <td>213500</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>5</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>253400</td>\n",
       "      <td>&lt;NA&gt;</td>\n",
       "      <td>&lt;NA&gt;</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1091</th>\n",
       "      <td>2020-12-27</td>\n",
       "      <td>154700</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>52</td>\n",
       "      <td>7</td>\n",
       "      <td>362</td>\n",
       "      <td>1</td>\n",
       "      <td>144600</td>\n",
       "      <td>233700</td>\n",
       "      <td>114600</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1092</th>\n",
       "      <td>2020-12-28</td>\n",
       "      <td>73100</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>53</td>\n",
       "      <td>1</td>\n",
       "      <td>363</td>\n",
       "      <td>0</td>\n",
       "      <td>154700</td>\n",
       "      <td>194100</td>\n",
       "      <td>124700</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1093</th>\n",
       "      <td>2020-12-29</td>\n",
       "      <td>63200</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>53</td>\n",
       "      <td>2</td>\n",
       "      <td>364</td>\n",
       "      <td>0</td>\n",
       "      <td>73100</td>\n",
       "      <td>154200</td>\n",
       "      <td>83100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1094</th>\n",
       "      <td>2020-12-30</td>\n",
       "      <td>143300</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>53</td>\n",
       "      <td>3</td>\n",
       "      <td>365</td>\n",
       "      <td>0</td>\n",
       "      <td>63200</td>\n",
       "      <td>164300</td>\n",
       "      <td>143200</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1095</th>\n",
       "      <td>2020-12-31</td>\n",
       "      <td>103400</td>\n",
       "      <td>4</td>\n",
       "      <td>12</td>\n",
       "      <td>53</td>\n",
       "      <td>4</td>\n",
       "      <td>366</td>\n",
       "      <td>0</td>\n",
       "      <td>143300</td>\n",
       "      <td>184400</td>\n",
       "      <td>203300</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1096 rows × 11 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      SalesDate  IceCreamSales  SalesDateQuarter  SalesDateMonth  \\\n",
       "0    2018-01-01         253100                 1               1   \n",
       "1    2018-01-02         213200                 1               1   \n",
       "2    2018-01-03         193300                 1               1   \n",
       "3    2018-01-04         253400                 1               1   \n",
       "4    2018-01-05         213500                 1               1   \n",
       "...         ...            ...               ...             ...   \n",
       "1091 2020-12-27         154700                 4              12   \n",
       "1092 2020-12-28          73100                 4              12   \n",
       "1093 2020-12-29          63200                 4              12   \n",
       "1094 2020-12-30         143300                 4              12   \n",
       "1095 2020-12-31         103400                 4              12   \n",
       "\n",
       "      SalesDateWeek  SalesDateWeekDay  SalesDateYearDay  SalesDateisWeekend  \\\n",
       "0                 1                 1                 1                   0   \n",
       "1                 1                 2                 2                   0   \n",
       "2                 1                 3                 3                   0   \n",
       "3                 1                 4                 4                   0   \n",
       "4                 1                 5                 5                   0   \n",
       "...             ...               ...               ...                 ...   \n",
       "1091             52                 7               362                   1   \n",
       "1092             53                 1               363                   0   \n",
       "1093             53                 2               364                   0   \n",
       "1094             53                 3               365                   0   \n",
       "1095             53                 4               366                   0   \n",
       "\n",
       "      IceCreamSales_lagged_1  IceCreamSales_lagged_7  IceCreamSales_lagged_365  \n",
       "0                       <NA>                    <NA>                      <NA>  \n",
       "1                     253100                    <NA>                      <NA>  \n",
       "2                     213200                    <NA>                      <NA>  \n",
       "3                     193300                    <NA>                      <NA>  \n",
       "4                     253400                    <NA>                      <NA>  \n",
       "...                      ...                     ...                       ...  \n",
       "1091                  144600                  233700                    114600  \n",
       "1092                  154700                  194100                    124700  \n",
       "1093                   73100                  154200                     83100  \n",
       "1094                   63200                  164300                    143200  \n",
       "1095                  143300                  184400                    203300  \n",
       "\n",
       "[1096 rows x 11 columns]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "IceCream_df_extended = get_shifted_target_values(IceCream_df_extended, [1,7,365], \"SalesDate\",\"IceCreamSales\")\n",
    "IceCream_df_extended"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Widzimy, że początek zbioru danych zawiera teraz wartości puste, ze względu na fakt, że dla wartości z roku 2018 nie istniały obserwacje cofnięte o rok. Do pokazywania liczby niepustych wartości w ramce danych przydatna jest funkcja info z biblioteki pandas"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:00.709833Z",
     "start_time": "2021-10-13T07:16:00.680833Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 1096 entries, 0 to 1095\n",
      "Data columns (total 11 columns):\n",
      " #   Column                    Non-Null Count  Dtype         \n",
      "---  ------                    --------------  -----         \n",
      " 0   SalesDate                 1096 non-null   datetime64[ns]\n",
      " 1   IceCreamSales             1096 non-null   Int64         \n",
      " 2   SalesDateQuarter          1096 non-null   int64         \n",
      " 3   SalesDateMonth            1096 non-null   int64         \n",
      " 4   SalesDateWeek             1096 non-null   UInt32        \n",
      " 5   SalesDateWeekDay          1096 non-null   int64         \n",
      " 6   SalesDateYearDay          1096 non-null   int64         \n",
      " 7   SalesDateisWeekend        1096 non-null   int32         \n",
      " 8   IceCreamSales_lagged_1    1095 non-null   Int64         \n",
      " 9   IceCreamSales_lagged_7    1089 non-null   Int64         \n",
      " 10  IceCreamSales_lagged_365  731 non-null    Int64         \n",
      "dtypes: Int64(4), UInt32(1), datetime64[ns](1), int32(1), int64(4)\n",
      "memory usage: 99.5 KB\n"
     ]
    }
   ],
   "source": [
    "IceCream_df_extended.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "oczywiście wartości puste należy potem w jakiś sposób obsłużyć, poprzez ich usunięcie lub imputację"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-info\">\n",
    "<b>Info</b> \n",
    "\n",
    "Tworząc zmienne oparte o cechy przesunięte w czasie warto mieć na uwadze, że dane historyczne mogą zawierać wartości odstające. Dlatego zamiast brać wprost wartość sprzed np. 365 dni można rozważyć wygładzenie wartości stosując medianę z 5 dniowego okna, którego środek stanowi wartość sprzed 365 dni.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-11T12:30:06.208232Z",
     "start_time": "2021-10-11T12:30:06.192598Z"
    }
   },
   "source": [
    "# Cechy kategoryczne i ich transformacje"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Dla zilustrowania transformacji na zmiennych kategorycznych posłużymy się drugim z przygotowanych zbiorów danych, gdzie chcemy przewidzieć cenę sprzedaży używanego samochodu."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:04.773577Z",
     "start_time": "2021-10-13T07:16:04.754574Z"
    }
   },
   "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>selling_price</th>\n",
       "      <th>condition</th>\n",
       "      <th>mileage</th>\n",
       "      <th>brand</th>\n",
       "      <th>year_manufactured</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>32123.0</td>\n",
       "      <td>good</td>\n",
       "      <td>129627</td>\n",
       "      <td>Mercedes</td>\n",
       "      <td>1996</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>62394.0</td>\n",
       "      <td>medium</td>\n",
       "      <td>68446</td>\n",
       "      <td>Bugatti</td>\n",
       "      <td>2013</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>8746.0</td>\n",
       "      <td>medium</td>\n",
       "      <td>218559</td>\n",
       "      <td>Fiat</td>\n",
       "      <td>1987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>9193.0</td>\n",
       "      <td>good</td>\n",
       "      <td>573386</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>1952</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>12340.0</td>\n",
       "      <td>good</td>\n",
       "      <td>27153</td>\n",
       "      <td>Fiat</td>\n",
       "      <td>2005</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   selling_price condition  mileage     brand  year_manufactured\n",
       "0        32123.0      good   129627  Mercedes               1996\n",
       "1        62394.0    medium    68446   Bugatti               2013\n",
       "2         8746.0    medium   218559      Fiat               1987\n",
       "3         9193.0      good   573386     Skoda               1952\n",
       "4        12340.0      good    27153      Fiat               2005"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "UsedCars_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "jak widzimy występują tutaj 2 cechy kategoryczne condition oraz brand, na podstawie których zaprezentujemy które transformacje danych najlepiej zastosować w którym przypadku. Aby umożliwić wykorzystanie tych cech w predykcji niezbędne jest odpowiednie ich przekształcenie w wartości numeryczne"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:06.297398Z",
     "start_time": "2021-10-13T07:16:06.279408Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['good', 'medium', 'bad', 'very_good', 'very_bad'], dtype=object)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "UsedCars_df.condition.unique()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "patrząc na atrybuty zmiennej condition widzimy, że ma ona charakter porządkowy - jesteśmy w stanie łatwo ustalić naturalną kolejność jej atrybutów. Stanowi ona zatem dobrego kandydata do zastosowania <b>kodowania porządkowego</b>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-11T12:47:35.578945Z",
     "start_time": "2021-10-11T12:47:35.566890Z"
    }
   },
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "<b>kodowania porządkowe</b> (ang. *Ordinal encoding*) - reprezentacja każdego atrybutu kodowanej zmiennej jako kolejnej liczby naturalnej \n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "poniżej zaprezentujemy kodowanie porządkowe z użyciem transformera OrdinalEncoder, najpierw jednak podzielimy nasze dane na zbiór treningowy i testowy, aby lepiej odwzorować to, że dane treningowe są oddzielone od danych produkcyjnych, na których model będzie potem stosowany."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:22.871995Z",
     "start_time": "2021-10-13T07:16:10.112336Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import OrdinalEncoder\n",
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tutaj dla uproszczenia stosujemy tylko prosty podział na zbiór treningowy i testowy, generalnie najlepszą praktyką jest wydzielenie osobnego zbioru testowego reprezentującego zdolność modelu do generalizacji na nowych, niewidzianych wcześniej danych i dobór najlepszego zestawu parametrów i transformacji stosując crosswalidację na zbiorze treningowym."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:23.275504Z",
     "start_time": "2021-10-13T07:16:23.261497Z"
    }
   },
   "outputs": [],
   "source": [
    "X_train, X_test = train_test_split(UsedCars_df, test_size =0.25, random_state=42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "następnie tworzymy obiekt klasy OrdinalEncoder, określamy tutaj porządek cechy atrybutów cechy, którą chcemy przetransformować, inaczej mogą one zostać po prostu posortowane alfabetycznie co najczęściej nie będzie odpowiadało ich znaczeniu biznesowemu"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:23.894916Z",
     "start_time": "2021-10-13T07:16:23.860918Z"
    }
   },
   "outputs": [],
   "source": [
    "oe= OrdinalEncoder(categories =[['very_bad', 'bad', 'medium', 'good', 'very_good']],\n",
    "                   handle_unknown ='use_encoded_value', unknown_value=np.NaN)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-danger\">\n",
    "<b>Uwaga!</b> \n",
    "    \n",
    "Kodując atrybuty należy mieć świadomość, że na nowych danych mogą pojawić się niewidziane wcześniej wartości, które należy w jakiś sposób obsłużyć. Domyślnym zachowaniem OrdinalEncoder w takiej sytuacji jest zwrócenie wyjątku, tutaj skorzystaliśmy z przypisania mu ustalonej wartości, gdzie wybraliśmy  przypisanie wartości pustej. Następnie taką wartość można zastąpić np. dominantą ze zbioru treningowego, lub przypisać jej wartość neutralną, którą tutaj byłoby 2 odpowiadające kategorii medium.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Zdecydowanie najlepszym sposobem użycia wszystkich transformerów jest skorzystanie z <b>pipeline</b>, co zostanie zaprezentowane potem. Tutaj zaprezentujemy najprostsze wykorzystanie polegające na skorzystaniu z metod <b>fit</b> i <b>transform</b>. OrdinalEncoder stosujemy tylko do przekształcenia jednej cechy, gdyby było inaczej moglibyśmy wykonać fit na całym zbiorze treningowym a następnie przetransformować zbiory treningowy i testowy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:24.439446Z",
     "start_time": "2021-10-13T07:16:24.425440Z"
    }
   },
   "outputs": [],
   "source": [
    "oe.fit(X_train.condition.values.reshape(-1, 1))\n",
    "X_train[\"condition_transformed\"] = oe.transform(X_train.condition.values.reshape(-1, 1))\n",
    "X_test[\"condition_transformed\"] = oe.transform(X_test.condition.values.reshape(-1, 1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-danger\">\n",
    "<b>Uwaga!</b> \n",
    "    \n",
    "Jeśli nie korzystamy z pipeline należy zwrócić szczególną uwagę żeby metody fit używać tylko na zbiorze treningowym a następnie mając już \"nauczony\" transformer stosować metodę transform na pozostałych zbiorach. Inaczej może dojść do przecieku informacji ze zbioru testowego.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "możemy jeszcze sprawdzić czy przypisanie atrybutów na zbiorach treningowym i testowym jest prawidłowe, a następnie pozbyć się pierwotnej kolumny, wartości pierwotne mogą być łatwo odzyskane z przetransformowanych danych stosując metodę <b>inverse_transform</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:24.997502Z",
     "start_time": "2021-10-13T07:16:24.973502Z"
    }
   },
   "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>condition</th>\n",
       "      <th>condition_transformed</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>323</th>\n",
       "      <td>very_bad</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>bad</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>medium</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>good</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>very_good</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "     condition  condition_transformed\n",
       "323   very_bad                    0.0\n",
       "894        bad                    1.0\n",
       "82      medium                    2.0\n",
       "991       good                    3.0\n",
       "398  very_good                    4.0"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[[\"condition\",\"condition_transformed\"]].drop_duplicates().sort_values(by=\"condition_transformed\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:25.785826Z",
     "start_time": "2021-10-13T07:16:25.749495Z"
    }
   },
   "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>condition</th>\n",
       "      <th>condition_transformed</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>346</th>\n",
       "      <td>very_bad</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>521</th>\n",
       "      <td>bad</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>740</th>\n",
       "      <td>medium</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>513</th>\n",
       "      <td>good</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>737</th>\n",
       "      <td>very_good</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "     condition  condition_transformed\n",
       "346   very_bad                    0.0\n",
       "521        bad                    1.0\n",
       "740     medium                    2.0\n",
       "513       good                    3.0\n",
       "737  very_good                    4.0"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_test[[\"condition\",\"condition_transformed\"]].drop_duplicates().sort_values(by=\"condition_transformed\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:26.610243Z",
     "start_time": "2021-10-13T07:16:26.595250Z"
    }
   },
   "outputs": [],
   "source": [
    "X_train.drop([\"condition\"],axis=1, inplace=True)\n",
    "X_test.drop([\"condition\"],axis=1, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:27.463826Z",
     "start_time": "2021-10-13T07:16:27.441818Z"
    }
   },
   "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>selling_price</th>\n",
       "      <th>mileage</th>\n",
       "      <th>brand</th>\n",
       "      <th>year_manufactured</th>\n",
       "      <th>condition_transformed</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>7986.0</td>\n",
       "      <td>524486</td>\n",
       "      <td>VW</td>\n",
       "      <td>1956</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>10531.0</td>\n",
       "      <td>166856</td>\n",
       "      <td>Toyota</td>\n",
       "      <td>1996</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>9532.0</td>\n",
       "      <td>94191</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>2000</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>7266.0</td>\n",
       "      <td>54210</td>\n",
       "      <td>VW</td>\n",
       "      <td>1954</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>11133.0</td>\n",
       "      <td>264049</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>1985</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>31713.0</td>\n",
       "      <td>127914</td>\n",
       "      <td>Mercedes</td>\n",
       "      <td>1990</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>270</th>\n",
       "      <td>9918.0</td>\n",
       "      <td>185880</td>\n",
       "      <td>Seat</td>\n",
       "      <td>1962</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>4824.0</td>\n",
       "      <td>116011</td>\n",
       "      <td>VW</td>\n",
       "      <td>2007</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>435</th>\n",
       "      <td>11329.0</td>\n",
       "      <td>25782</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>2014</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>8860.0</td>\n",
       "      <td>145116</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>1979</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>750 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     selling_price  mileage     brand  year_manufactured  \\\n",
       "82          7986.0   524486        VW               1956   \n",
       "991        10531.0   166856    Toyota               1996   \n",
       "789         9532.0    94191     Skoda               2000   \n",
       "894         7266.0    54210        VW               1954   \n",
       "398        11133.0   264049     Skoda               1985   \n",
       "..             ...      ...       ...                ...   \n",
       "106        31713.0   127914  Mercedes               1990   \n",
       "270         9918.0   185880      Seat               1962   \n",
       "860         4824.0   116011        VW               2007   \n",
       "435        11329.0    25782     Skoda               2014   \n",
       "102         8860.0   145116     Skoda               1979   \n",
       "\n",
       "     condition_transformed  \n",
       "82                     2.0  \n",
       "991                    3.0  \n",
       "789                    2.0  \n",
       "894                    1.0  \n",
       "398                    4.0  \n",
       "..                     ...  \n",
       "106                    3.0  \n",
       "270                    3.0  \n",
       "860                    0.0  \n",
       "435                    2.0  \n",
       "102                    2.0  \n",
       "\n",
       "[750 rows x 5 columns]"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "widzimy, że do zakodowania pozostała nam jeszcze cecha - brand, w odróżnieniu od poprzednio rozważanej cechy tutaj nie ma oczywistego naturalnego porządku, dlatego należy tutaj zastosować inne podejście"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "<b>kodowania 1 z n</b> (ang. *One-hot encoding*) - reprezentacja każdego atrybutu kodowanej zmiennej jako osobnej zmiennej binarnej, gdzie występowanie rozważanego atrybutu dla danej obserwacji oznaczane jest jako 1 a wszystkie pozostałe atrybuty oznaczane są jako 0 \n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "kodowanie 1 z n stanowi jeden z najczęściej używanych i najbardziej intuicyjnych sposobów kodowania, jednak jego główną wadą jest zwiększanie wymiarowości danych, co zaraz zaprezentujemy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "najpierw pokażemy jak posługiwać się transformerem OneHotEncoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:31.358006Z",
     "start_time": "2021-10-13T07:16:31.349906Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import OneHotEncoder"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tworzymy obiekt klasy OneHotEncoder, podobnie jak w przypadku poprzedniego transformera, tu także domyślnym sposobem obsługi nieznanych atrybutów jest zwracanie wyjątku, my wybieramy handle_unknown=\"ignore\" co sprawi, że nieznane atrybuty reprezentowane będą po prostu jako 0 we wszystkich zakodowanych kolumnach.\n",
    "\n",
    "Natomiast ustawienie sparse=False sprawia, że zwracane dane będą typu np.array zamiast domyślnego sparse matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:32.849003Z",
     "start_time": "2021-10-13T07:16:32.841484Z"
    }
   },
   "outputs": [],
   "source": [
    "ohe = OneHotEncoder(handle_unknown=\"ignore\",sparse=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "należy zwrócić uwagę, że we wszystkich transformerach z biblioteki sklearn stosowane jest to samo API, w związku z tym tutaj analogicznie jak w poprzednim przypadku możemy skorzystać z metod <b>fit</b> i <b>transform</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:34.371505Z",
     "start_time": "2021-10-13T07:16:34.354961Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "rozmiar zakodowanej kolumny brand na danych treningowych to: (750, 10)\n",
      "rozmiar zakodowanej kolumny brand na danych testowych to: (250, 10)\n",
      "[[0. 0. 0. ... 0. 0. 1.]\n",
      " [0. 0. 0. ... 0. 1. 0.]\n",
      " [0. 0. 0. ... 1. 0. 0.]\n",
      " ...\n",
      " [0. 0. 0. ... 0. 0. 1.]\n",
      " [0. 0. 0. ... 1. 0. 0.]\n",
      " [0. 0. 0. ... 1. 0. 0.]]\n"
     ]
    }
   ],
   "source": [
    "ohe.fit(X_train.brand.values.reshape(-1, 1))\n",
    "brand_transformed_train = ohe.transform(X_train.brand.values.reshape(-1, 1))\n",
    "brand_transformed_test = ohe.transform(X_test.brand.values.reshape(-1, 1))\n",
    "print(f\"rozmiar zakodowanej kolumny brand na danych treningowych to: {brand_transformed_train.shape}\")\n",
    "print(f\"rozmiar zakodowanej kolumny brand na danych testowych to: {brand_transformed_test.shape}\")\n",
    "print(brand_transformed_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "jak widzimy kolumna brand reprezentowana jest teraz jako array gdzie każdy atrybut ze zbioru treningowego zaprezentowany jest w osobnej kolumnie binarnej. Można to sprawdzić zliczając liczbę unikalnych atrybutów tej kolumny na zbiorze treningowym"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:35.963830Z",
     "start_time": "2021-10-13T07:16:35.957828Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.brand.nunique()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "kolejność atrybutów odpowiadających kolumnom wynikowego arraya można zobaczyć korzystając z atrybutu categories_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:37.722373Z",
     "start_time": "2021-10-13T07:16:37.709372Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[array(['Audi', 'BMW', 'Bugatti', 'Fiat', 'Mercedes', 'Renault', 'Seat',\n",
       "        'Skoda', 'Toyota', 'VW'], dtype=object)]"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ohe.categories_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "całą transformację wraz z dodaniem przetransformowanej zmiennej do zbioru treningowego można zrealizować za pomocą prostej funkcji"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:39.641779Z",
     "start_time": "2021-10-13T07:16:39.632782Z"
    }
   },
   "outputs": [],
   "source": [
    "def OneHotEncode(X_train, X_test, encoded_column_name, **encoder_kwargs):\n",
    "    \n",
    "    ohe = OneHotEncoder(**encoder_kwargs)\n",
    "    ohe.fit(X_train[encoded_column_name].values.reshape(-1, 1))\n",
    "    transformed_train = ohe.transform(X_train[encoded_column_name].values.reshape(-1, 1))\n",
    "    transformed_test = ohe.transform(X_test[encoded_column_name].values.reshape(-1, 1))\n",
    "    \n",
    "    column_names = [encoded_column_name +\"_\"+category for category in list(ohe.categories_[0]) ]\n",
    "    df_transformed_train = pd.DataFrame(transformed_train, columns = column_names)\n",
    "    df_transformed_test = pd.DataFrame(transformed_test, columns = column_names)\n",
    "    \n",
    "    df_out_train = X_train.reset_index(drop=True).drop([encoded_column_name], axis=1)\n",
    "    df_out_test = X_test.reset_index(drop=True).drop([encoded_column_name], axis=1)\n",
    "    df_out_train = pd.concat([df_out_train, df_transformed_train],axis=1)\n",
    "    df_out_test = pd.concat([df_out_test, df_transformed_test],axis=1)\n",
    "    \n",
    "    return df_out_train, df_out_test"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "powyższa funkcja zwraca zbiór treningowy i testowy po dodaniu odpowiednio nazwanych kolumn powstałych po transformacji obiektem OneHotEncoder i usunięciu pierwotnej zmiennej. Argumenty do OneHotEncoder przekazywane są z pomocą  **encoder_kwargs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:41.132045Z",
     "start_time": "2021-10-13T07:16:41.085522Z"
    }
   },
   "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>selling_price</th>\n",
       "      <th>mileage</th>\n",
       "      <th>year_manufactured</th>\n",
       "      <th>condition_transformed</th>\n",
       "      <th>brand_Audi</th>\n",
       "      <th>brand_BMW</th>\n",
       "      <th>brand_Bugatti</th>\n",
       "      <th>brand_Fiat</th>\n",
       "      <th>brand_Mercedes</th>\n",
       "      <th>brand_Renault</th>\n",
       "      <th>brand_Seat</th>\n",
       "      <th>brand_Skoda</th>\n",
       "      <th>brand_Toyota</th>\n",
       "      <th>brand_VW</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>7986.0</td>\n",
       "      <td>524486</td>\n",
       "      <td>1956</td>\n",
       "      <td>2.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>10531.0</td>\n",
       "      <td>166856</td>\n",
       "      <td>1996</td>\n",
       "      <td>3.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>9532.0</td>\n",
       "      <td>94191</td>\n",
       "      <td>2000</td>\n",
       "      <td>2.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>7266.0</td>\n",
       "      <td>54210</td>\n",
       "      <td>1954</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>11133.0</td>\n",
       "      <td>264049</td>\n",
       "      <td>1985</td>\n",
       "      <td>4.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   selling_price  mileage  year_manufactured  condition_transformed  \\\n",
       "0         7986.0   524486               1956                    2.0   \n",
       "1        10531.0   166856               1996                    3.0   \n",
       "2         9532.0    94191               2000                    2.0   \n",
       "3         7266.0    54210               1954                    1.0   \n",
       "4        11133.0   264049               1985                    4.0   \n",
       "\n",
       "   brand_Audi  brand_BMW  brand_Bugatti  brand_Fiat  brand_Mercedes  \\\n",
       "0         0.0        0.0            0.0         0.0             0.0   \n",
       "1         0.0        0.0            0.0         0.0             0.0   \n",
       "2         0.0        0.0            0.0         0.0             0.0   \n",
       "3         0.0        0.0            0.0         0.0             0.0   \n",
       "4         0.0        0.0            0.0         0.0             0.0   \n",
       "\n",
       "   brand_Renault  brand_Seat  brand_Skoda  brand_Toyota  brand_VW  \n",
       "0            0.0         0.0          0.0           0.0       1.0  \n",
       "1            0.0         0.0          0.0           1.0       0.0  \n",
       "2            0.0         0.0          1.0           0.0       0.0  \n",
       "3            0.0         0.0          0.0           0.0       1.0  \n",
       "4            0.0         0.0          1.0           0.0       0.0  "
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train_ohe, X_test_ohe = OneHotEncode(X_train, X_test, \"brand\", handle_unknown=\"ignore\",sparse=False)\n",
    "X_train_ohe.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-info\">\n",
    "<b>Info</b> \n",
    "\n",
    "Alternatywnie zamiast OneHotEncoder można wykorzystać funkcję <b>get_dummies</b> z biblioteki pandas, która jest nieco prostsza w użyciu, jednak OneHotEncoder jest lepiej dostosowany do obsługi niewidzianych wartości a ponadto znacznie łatwiej go zastosować w ramach pipeline\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "jak widzimy nawet przy tak niskiej liczbie atrybutów wymiarowość naszych danych znacznie wzrosła. W prawdziwych zastosowaniach możemy się spotkać ze zbiorami danych mającymi wiele zmiennych kategorycznych o dziesiątkach lub setkach unikalnych atrybutów, więc metodę OneHotEncoder zaleca się stosować tylko tam, gdzie atrybuty są niezbyt liczne, w pozostałych sytuacjach lepiej skorzystać z <b>kodowania zmienną celu</b>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T09:05:01.108185Z",
     "start_time": "2021-10-12T09:05:01.092691Z"
    }
   },
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "<b>kodowanie zmienną celu</b> (ang. *Target encoding*) - reprezentacja każdego atrybutu kodowanej zmiennej jako estymata średniej wartości zmiennej celu na danych treningowych\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Kodowanie zmienną celu stanowi jedną z najbardziej uniwersalnych metod kodowania zmiennych kategorycznych, nie wymaga naturalnego porządku w danych jak kodowanie porządkowe i nie zwiększa wymiarowości danych jak kodowanie 1 z n.\n",
    "\n",
    "Istnieje wiele różnych sposobów kodowania zmiennej celu, my tutaj posłużymy się metodą James-Stein Encoder dostępną w bibliotece category_encoders"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T09:15:40.326427Z",
     "start_time": "2021-10-12T09:15:40.319428Z"
    }
   },
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "Formuła na zakodowaną wartość k-tego atrybutu kodowanej zmiennej:\n",
    "$$\n",
    "X_{k} = (1-B) *Avg(Y_{k}) +B*Avg(Y)\n",
    "$$\n",
    "\n",
    "    \n",
    "gdzie:\n",
    "    \n",
    "$X_{k}$ - wartość przypisana k-temu atrybutowi kodowanej zmiennej kategorycznej\n",
    "    \n",
    "$Avg(Y_{k})$ - średnia wartości zmiennej celu dla k-tego atrybutu zmiennej kategorycznej\n",
    "    \n",
    "$Avg(Y)$ - globalna średnia zmiennej celu na zbiorze treningowym\n",
    "    \n",
    "$B$ -waga globalnej średniej, wyliczona według formuły:\n",
    "$$\n",
    "B = \\frac{Var(Y_{k})}{Var(Y) + Var(Y_{k})}\n",
    "$$\n",
    "gdzie:\n",
    "\n",
    "$Var(Y_{k})$ - wariancja zmiennej celu dla k-tego atrybutu zmiennej kategorycznej\n",
    "\n",
    "$Var(Y)$ - wariancja zmiennej celu dla całego zbioru treningowego"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "importujemy potrzebną klasę"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:47.637140Z",
     "start_time": "2021-10-13T07:16:46.169136Z"
    }
   },
   "outputs": [],
   "source": [
    "from category_encoders.james_stein import JamesSteinEncoder"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tworzymy obiekt klasy JamesSteinEncoder, wybierając kolumnę brand do przekształcenia, domyślnie przekształcone zostaną wszystkie kolumny kategoryczne"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:48.554893Z",
     "start_time": "2021-10-13T07:16:48.542897Z"
    }
   },
   "outputs": [],
   "source": [
    "jse = JamesSteinEncoder(cols=[\"brand\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tak jak w transformerach z biblioteki sklearn także tutaj posługujemy się metodami <b>fit</b> i <b>transform</b>, jednak jako, że jest to kodowanie zmienną celu niezbędne jest jej podanie do metody fit, dlatego najpierw wydzielimy zmienną celu"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:49.316898Z",
     "start_time": "2021-10-13T07:16:49.305893Z"
    }
   },
   "outputs": [],
   "source": [
    "y_train = X_train.selling_price\n",
    "y_test = X_test.selling_price\n",
    "X_train = X_train.drop(\"selling_price\",axis=1)\n",
    "X_test = X_test.drop(\"selling_price\",axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:50.323299Z",
     "start_time": "2021-10-13T07:16:50.254236Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>brand</th>\n",
       "      <th>year_manufactured</th>\n",
       "      <th>condition_transformed</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>524486</td>\n",
       "      <td>9530.009707</td>\n",
       "      <td>1956</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>166856</td>\n",
       "      <td>10018.691398</td>\n",
       "      <td>1996</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>94191</td>\n",
       "      <td>9636.700001</td>\n",
       "      <td>2000</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>54210</td>\n",
       "      <td>9530.009707</td>\n",
       "      <td>1954</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>264049</td>\n",
       "      <td>9636.700001</td>\n",
       "      <td>1985</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "     mileage         brand  year_manufactured  condition_transformed\n",
       "82    524486   9530.009707               1956                    2.0\n",
       "991   166856  10018.691398               1996                    3.0\n",
       "789    94191   9636.700001               2000                    2.0\n",
       "894    54210   9530.009707               1954                    1.0\n",
       "398   264049   9636.700001               1985                    4.0"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "jse.fit(X_train,y_train)\n",
    "X_train_jse = jse.transform(X_train)\n",
    "X_test_jse = jse.transform(X_test)\n",
    "X_train_jse.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "jak widzimy zmienna brand została teraz zastąpiona estymatami średniej wartości zmiennej celu w zależności od brandu"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Łączenie atrybutów cech kategorycznych"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Oprócz standardowych metod transformacji zmiennych kategorycznych opisanych w poprzedniej sekcji można też stosować różne przekształcenia polegające na łączeniu pierwotnych atrybutów w podgrupy. Takie łączenie może być oparte na podobieństwie atrybutów pod katęm statystycznym np. zbliżony poziom średniej wartości zmiennej celu lub, co bardziej zalecane - na wiedzy domenowej."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tutaj zaprezentujemy prosty przykład łączenia atrybutów oparty na wiedzy domenowej - załóżmy, że znamy podział wszystkich marek samochodów ze zmiennej brand na 3 relatywnie jednolite podgrupy - marki podstawowe, marki premium i marki luksusowe. Na tej podstawie stworzymy nową zmienną korzystając z prostej funkcji:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:54.379596Z",
     "start_time": "2021-10-13T07:16:54.368589Z"
    }
   },
   "outputs": [],
   "source": [
    "def brand_binning(brand):\n",
    "    result =\"standard\"\n",
    "    standard_brands = [\"Fiat\",\"Renault\",\"VW\", \"Seat\", \"Skoda\",\"Toyota\"]\n",
    "    premium_brands = [\"Audi\",\"BMW\",\"Mercedes\"]\n",
    "    luxury_brands = [\"Bugatti\"]\n",
    "    if brand in luxury_brands:\n",
    "        result = \"luxury\"\n",
    "    elif brand in premium_brands:\n",
    "        result = \"premium\"\n",
    "    return result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:55.433592Z",
     "start_time": "2021-10-13T07:16:55.399599Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>brand</th>\n",
       "      <th>year_manufactured</th>\n",
       "      <th>condition_transformed</th>\n",
       "      <th>brand_binned</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>524486</td>\n",
       "      <td>VW</td>\n",
       "      <td>1956</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>166856</td>\n",
       "      <td>Toyota</td>\n",
       "      <td>1996</td>\n",
       "      <td>3.0</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>94191</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>2000</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>54210</td>\n",
       "      <td>VW</td>\n",
       "      <td>1954</td>\n",
       "      <td>1.0</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>264049</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>1985</td>\n",
       "      <td>4.0</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "     mileage   brand  year_manufactured  condition_transformed brand_binned\n",
       "82    524486      VW               1956                    2.0     standard\n",
       "991   166856  Toyota               1996                    3.0     standard\n",
       "789    94191   Skoda               2000                    2.0     standard\n",
       "894    54210      VW               1954                    1.0     standard\n",
       "398   264049   Skoda               1985                    4.0     standard"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[\"brand_binned\"] = X_train[\"brand\"].map(lambda x:brand_binning(x))\n",
    "X_test[\"brand_binned\"] = X_test[\"brand\"].map(lambda x:brand_binning(x))\n",
    "X_train.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "możemy też sprawdzić jak wygląda unikalne przypisanie pomiędzy pierwotną a zgrupowaną zmienną:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:16:56.678205Z",
     "start_time": "2021-10-13T07:16:56.663102Z"
    }
   },
   "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>brand</th>\n",
       "      <th>brand_binned</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>Bugatti</td>\n",
       "      <td>luxury</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>323</th>\n",
       "      <td>Audi</td>\n",
       "      <td>premium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>798</th>\n",
       "      <td>Mercedes</td>\n",
       "      <td>premium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>717</th>\n",
       "      <td>BMW</td>\n",
       "      <td>premium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>VW</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>Toyota</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>Skoda</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>731</th>\n",
       "      <td>Fiat</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>97</th>\n",
       "      <td>Renault</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>523</th>\n",
       "      <td>Seat</td>\n",
       "      <td>standard</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        brand brand_binned\n",
       "5     Bugatti       luxury\n",
       "323      Audi      premium\n",
       "798  Mercedes      premium\n",
       "717       BMW      premium\n",
       "82         VW     standard\n",
       "991    Toyota     standard\n",
       "789     Skoda     standard\n",
       "731      Fiat     standard\n",
       "97    Renault     standard\n",
       "523      Seat     standard"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[[\"brand\",\"brand_binned\"]].drop_duplicates().sort_values(by=\"brand_binned\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "jak widzimy przypisanie zostało przeprowadzone poprawnie, nowa zmienna niesie mniej informacji niż zmienna pierwotna, dlatego zastąpienie nią pierwotnej zmiennej może być pomocne jeśli mamy problem z przeuczeniem modelu. Oczywiście tak powstałą zmienną należy potem jeszcze przetransformować zgodnie z wytycznymi z poprzedniej sekcji."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T10:12:01.320269Z",
     "start_time": "2021-10-12T10:12:01.311270Z"
    }
   },
   "source": [
    "# Cechy oparte na wiedzy domenowej oraz cechy interakcji"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Zaprezentowane powyżej grupowanie marek polegało na przekształceniu zmiennej kategorycznej w oparciu o wiedzę domenową. Tutaj pokażemy przykłady zastosowania wiedzy domenowej na zmiennych numerycznych i interakcji cech numerycznych i katgorycznych."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Przykładem prostej cechy opartej na wiedzy domenowej może być wiek samochodu w momencie sprzedaży. Nie mamy tutaj informacji o dacie transakcji, dlatego można dla ułatwienia założyć, że wszystkie transakcje odbyły się w bieżącym roku."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:00.413063Z",
     "start_time": "2021-10-13T07:17:00.375401Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>brand</th>\n",
       "      <th>year_manufactured</th>\n",
       "      <th>condition_transformed</th>\n",
       "      <th>brand_binned</th>\n",
       "      <th>age</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>524486</td>\n",
       "      <td>VW</td>\n",
       "      <td>1956</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>65</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>166856</td>\n",
       "      <td>Toyota</td>\n",
       "      <td>1996</td>\n",
       "      <td>3.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>94191</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>2000</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>21</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>54210</td>\n",
       "      <td>VW</td>\n",
       "      <td>1954</td>\n",
       "      <td>1.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>67</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>264049</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>1985</td>\n",
       "      <td>4.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>36</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>106</th>\n",
       "      <td>127914</td>\n",
       "      <td>Mercedes</td>\n",
       "      <td>1990</td>\n",
       "      <td>3.0</td>\n",
       "      <td>premium</td>\n",
       "      <td>31</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>270</th>\n",
       "      <td>185880</td>\n",
       "      <td>Seat</td>\n",
       "      <td>1962</td>\n",
       "      <td>3.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>59</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>116011</td>\n",
       "      <td>VW</td>\n",
       "      <td>2007</td>\n",
       "      <td>0.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>435</th>\n",
       "      <td>25782</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>2014</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>145116</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>1979</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>42</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>750 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     mileage     brand  year_manufactured  condition_transformed brand_binned  \\\n",
       "82    524486        VW               1956                    2.0     standard   \n",
       "991   166856    Toyota               1996                    3.0     standard   \n",
       "789    94191     Skoda               2000                    2.0     standard   \n",
       "894    54210        VW               1954                    1.0     standard   \n",
       "398   264049     Skoda               1985                    4.0     standard   \n",
       "..       ...       ...                ...                    ...          ...   \n",
       "106   127914  Mercedes               1990                    3.0      premium   \n",
       "270   185880      Seat               1962                    3.0     standard   \n",
       "860   116011        VW               2007                    0.0     standard   \n",
       "435    25782     Skoda               2014                    2.0     standard   \n",
       "102   145116     Skoda               1979                    2.0     standard   \n",
       "\n",
       "     age  \n",
       "82    65  \n",
       "991   25  \n",
       "789   21  \n",
       "894   67  \n",
       "398   36  \n",
       "..   ...  \n",
       "106   31  \n",
       "270   59  \n",
       "860   14  \n",
       "435    7  \n",
       "102   42  \n",
       "\n",
       "[750 rows x 6 columns]"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[\"age\"] = 2021 - X_train[\"year_manufactured\"]\n",
    "X_test[\"age\"] = 2021 - X_test[\"year_manufactured\"]\n",
    "X_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Zmienne interakcji można wygenerować stosując np. iloczyny poszczególnych zmiennych numerycznych, tutaj wydaje się to nie być najlepszy pomysł, ponieważ intuicyjnie czujemy, że np mnożenie przebiegu * rok produkcji nie będzie zbyt dobrą cechą predykcyjną. Można natomiast wyliczyć np średni roczny przebieg dzieląc przebieg przez dodany powyżej wiek samochodu."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:02.036080Z",
     "start_time": "2021-10-13T07:17:02.002080Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>brand</th>\n",
       "      <th>year_manufactured</th>\n",
       "      <th>condition_transformed</th>\n",
       "      <th>brand_binned</th>\n",
       "      <th>age</th>\n",
       "      <th>avg_yearly_mileage</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>524486</td>\n",
       "      <td>VW</td>\n",
       "      <td>1956</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>65</td>\n",
       "      <td>8069.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>166856</td>\n",
       "      <td>Toyota</td>\n",
       "      <td>1996</td>\n",
       "      <td>3.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>25</td>\n",
       "      <td>6674.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>94191</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>2000</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>21</td>\n",
       "      <td>4485.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>54210</td>\n",
       "      <td>VW</td>\n",
       "      <td>1954</td>\n",
       "      <td>1.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>67</td>\n",
       "      <td>809.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>264049</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>1985</td>\n",
       "      <td>4.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>36</td>\n",
       "      <td>7335.0</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>127914</td>\n",
       "      <td>Mercedes</td>\n",
       "      <td>1990</td>\n",
       "      <td>3.0</td>\n",
       "      <td>premium</td>\n",
       "      <td>31</td>\n",
       "      <td>4126.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>270</th>\n",
       "      <td>185880</td>\n",
       "      <td>Seat</td>\n",
       "      <td>1962</td>\n",
       "      <td>3.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>59</td>\n",
       "      <td>3151.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>116011</td>\n",
       "      <td>VW</td>\n",
       "      <td>2007</td>\n",
       "      <td>0.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>14</td>\n",
       "      <td>8286.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>435</th>\n",
       "      <td>25782</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>2014</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>7</td>\n",
       "      <td>3683.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>145116</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>1979</td>\n",
       "      <td>2.0</td>\n",
       "      <td>standard</td>\n",
       "      <td>42</td>\n",
       "      <td>3455.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>750 rows × 7 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     mileage     brand  year_manufactured  condition_transformed brand_binned  \\\n",
       "82    524486        VW               1956                    2.0     standard   \n",
       "991   166856    Toyota               1996                    3.0     standard   \n",
       "789    94191     Skoda               2000                    2.0     standard   \n",
       "894    54210        VW               1954                    1.0     standard   \n",
       "398   264049     Skoda               1985                    4.0     standard   \n",
       "..       ...       ...                ...                    ...          ...   \n",
       "106   127914  Mercedes               1990                    3.0      premium   \n",
       "270   185880      Seat               1962                    3.0     standard   \n",
       "860   116011        VW               2007                    0.0     standard   \n",
       "435    25782     Skoda               2014                    2.0     standard   \n",
       "102   145116     Skoda               1979                    2.0     standard   \n",
       "\n",
       "     age  avg_yearly_mileage  \n",
       "82    65              8069.0  \n",
       "991   25              6674.0  \n",
       "789   21              4485.0  \n",
       "894   67               809.0  \n",
       "398   36              7335.0  \n",
       "..   ...                 ...  \n",
       "106   31              4126.0  \n",
       "270   59              3151.0  \n",
       "860   14              8286.0  \n",
       "435    7              3683.0  \n",
       "102   42              3455.0  \n",
       "\n",
       "[750 rows x 7 columns]"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[\"avg_yearly_mileage\"] = np.round(X_train[\"mileage\"]/X_train[\"age\"])\n",
    "X_test[\"avg_yearly_mileage\"] = np.round(X_test[\"mileage\"]/X_test[\"age\"])\n",
    "X_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Innym przykładem zmiennej opartej stricte na wiedzy domenowej łączecej ze sobą informacje ze zmiennych numerycznych i kategorycznych mogła by być informacja o klasycznych modelach. Załóżmy, że modele luksusowych marek wyprodukowane przed 1970 rokiem są modelami klasycznymi i jeśli ich stan jest co najmniej dobry to ich cena jest znacznie wyższa niż by to wynikało z wieku auta, gdzie normalnie spodziewamy się ujemnej relacji z ceną sprzedaży."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:03.554608Z",
     "start_time": "2021-10-13T07:17:03.535315Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.0    739\n",
       "1.0     11\n",
       "Name: is_classic, dtype: int64"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.loc[(X_train.brand_binned==\"luxury\")&(X_train.year_manufactured<=1970)&(X_train.condition_transformed>2.0),\"is_classic\"]=1\n",
    "X_train[\"is_classic\"] = X_train[\"is_classic\"].fillna(0)\n",
    "X_test.loc[(X_test.brand_binned==\"luxury\")&(X_test.year_manufactured<=1970)&(X_test.condition_transformed>2.0),\"is_classic\"]=1\n",
    "X_test[\"is_classic\"] = X_test[\"is_classic\"].fillna(0)\n",
    "X_train[\"is_classic\"].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T10:57:29.776134Z",
     "start_time": "2021-10-12T10:57:29.751133Z"
    }
   },
   "source": [
    "Takie przypadki są stosunkowo rzadkie, ale tego typu zmienna może pomóc w dokładniejszym przewidzeniu tych obserwacji."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Cechy oparte na transformacji zmiennych numerycznych"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Dość powszechnie spotykaną praktyką jest stosowanie nieliniowych transformacji zmiennych numerycznych, co zazwyczaj sprzyja uzyskaniu lepszej jakości predykcji ze względu na fakt, że modele najlepiej radzą sobie gdy rozkłady zmiennych są zbliżone do rozkładu normalnego."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T11:56:28.907518Z",
     "start_time": "2021-10-12T11:56:28.875293Z"
    }
   },
   "source": [
    "Często spotykaną jest po prostu logarytmowanie zmiennych numerycznych, tutaj natomiast posłużymy się gotowym transformerem z biblioteki sklearn, a mianowicie <b>PowerTransformer</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:07.512038Z",
     "start_time": "2021-10-13T07:17:07.498042Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import PowerTransformer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "zanim dokonamy transformacji sprawdźmy jak wyglądają rozkłady zmiennych numerycznych, ograniczymy się tutaj do przebiegu, wieku i średniego przebiegu"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:09.592613Z",
     "start_time": "2021-10-13T07:17:08.864544Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[<AxesSubplot:title={'center':'mileage'}>,\n",
       "        <AxesSubplot:title={'center':'age'}>],\n",
       "       [<AxesSubplot:title={'center':'avg_yearly_mileage'}>,\n",
       "        <AxesSubplot:>]], dtype=object)"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1080x648 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "num_columns = [\"mileage\", \"age\",\"avg_yearly_mileage\"]\n",
    "X_train[num_columns].hist(figsize=(15,9), bins=30)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "można tu zaobserwować, że rozkład przebiegu jest silnie skośny, pozostałe 2 zmienne mają rozkład zbliżony do jednostajnego"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:10.658130Z",
     "start_time": "2021-10-13T07:17:10.645132Z"
    }
   },
   "outputs": [],
   "source": [
    "pt=PowerTransformer(standardize=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tworzymy obiekt klasy PowerTransformer, korzystamy tutaj z domyślnej transformacji Yeo-Johnson, zmieniamy natomiast ustawienie o sprowadzeniu do rozkładu o średniej równej 0 i odchyleniu standardowym równym 1, które domyślnie jest ustawione na True"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-info\">\n",
    "<b>Info</b> \n",
    "\n",
    "Domyślna transformacja czyli metoda Yeo-Johnsona może być stosowana niezależnie od znaku transformowanych zmiennych, natomiast alternatywna transformacja Box-Cox wymaga ściśle dodatnich zmiennych na wejściu.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:12.567395Z",
     "start_time": "2021-10-13T07:17:12.504256Z"
    }
   },
   "outputs": [],
   "source": [
    "pt.fit(X_train[num_columns])\n",
    "X_train_num_transformed = pd.DataFrame(pt.transform(X_train[num_columns]), columns =num_columns)\n",
    "X_test_num_transformed =  pd.DataFrame(pt.transform(X_test[num_columns]), columns =num_columns)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "analogicznie jak przy poprzednio stosowanych transformerach korzystamy tutaj z metod fit i transform, a następnie konwertujemy wynik na ramkę danych."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Przeanalizujmy jak mocno zmieniły się rozkłady poszczególnych zmiennych:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:15.319229Z",
     "start_time": "2021-10-13T07:17:14.579501Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[<AxesSubplot:title={'center':'mileage'}>,\n",
       "        <AxesSubplot:title={'center':'age'}>],\n",
       "       [<AxesSubplot:title={'center':'avg_yearly_mileage'}>,\n",
       "        <AxesSubplot:>]], dtype=object)"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1080x648 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "X_train_num_transformed.hist(figsize=(15,9), bins=30)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:17.300950Z",
     "start_time": "2021-10-13T07:17:16.516675Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[<AxesSubplot:title={'center':'mileage'}>,\n",
       "        <AxesSubplot:title={'center':'age'}>],\n",
       "       [<AxesSubplot:title={'center':'avg_yearly_mileage'}>,\n",
       "        <AxesSubplot:>]], dtype=object)"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1080x648 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "X_test_num_transformed.hist(figsize=(15,9), bins=30)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Zmienna mileage na zbiorze treningowym z silnie skośnej nabrała cech rozkładu normalnego, pozostałe przypadki nie wyglądają jednak na rozkład normalny.\n",
    "\n",
    "Wypróbujmy inny sposób modyfikacji rozkładu - <b>QuantileTransformer</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:18.505059Z",
     "start_time": "2021-10-13T07:17:18.494059Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import QuantileTransformer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T12:33:51.065807Z",
     "start_time": "2021-10-12T12:33:51.054809Z"
    }
   },
   "source": [
    "tworzymy obiekt klasy QuantileTransformer, zamiast bazowego rozkładu jednostajnego wybierając rozkład normalny"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:19.603548Z",
     "start_time": "2021-10-13T07:17:19.595552Z"
    }
   },
   "outputs": [],
   "source": [
    "qt = QuantileTransformer(output_distribution=\"normal\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:20.741079Z",
     "start_time": "2021-10-13T07:17:20.712098Z"
    }
   },
   "outputs": [],
   "source": [
    "qt.fit(X_train[num_columns])\n",
    "X_train_num_transformed = pd.DataFrame(qt.transform(X_train[num_columns]), columns =num_columns)\n",
    "X_test_num_transformed =  pd.DataFrame(qt.transform(X_test[num_columns]), columns =num_columns)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:22.402716Z",
     "start_time": "2021-10-13T07:17:21.796969Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[<AxesSubplot:title={'center':'mileage'}>,\n",
       "        <AxesSubplot:title={'center':'age'}>],\n",
       "       [<AxesSubplot:title={'center':'avg_yearly_mileage'}>,\n",
       "        <AxesSubplot:>]], dtype=object)"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1080x648 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "X_train_num_transformed.hist(figsize=(15,9), bins=30)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:24.287340Z",
     "start_time": "2021-10-13T07:17:23.467093Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[<AxesSubplot:title={'center':'mileage'}>,\n",
       "        <AxesSubplot:title={'center':'age'}>],\n",
       "       [<AxesSubplot:title={'center':'avg_yearly_mileage'}>,\n",
       "        <AxesSubplot:>]], dtype=object)"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1080x648 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "X_test_num_transformed.hist(figsize=(15,9), bins=30)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Jak widzimy rozkłady są teraz zupełnie inne, na zbiorze treningowym bardzo zbliżone do rozkładu normalnego, jednak na zbiorze testowym juz dość znacznie się różnią, co może wynikać po części z niewielkiego rozmiaru próby."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T12:44:21.258422Z",
     "start_time": "2021-10-12T12:44:21.242568Z"
    }
   },
   "source": [
    "<div class=\"alert alert-block alert-info\">\n",
    "<b>Info</b> \n",
    "\n",
    "To jaki dokładnie zestaw transformacji ostatecznie wybrać powinno wynikać przede wszystkim z obserwacji ustalonej wcześniej metryki jakości modelu.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T05:54:13.280903Z",
     "start_time": "2021-10-13T05:54:13.268867Z"
    }
   },
   "source": [
    "# Łączenie zmiennych numerycznych w przedziały"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T05:58:15.399488Z",
     "start_time": "2021-10-13T05:58:15.382490Z"
    }
   },
   "source": [
    "podobnie jak w przypadku zmiennych kategorycznych, zmienne numeryczne również można przekształcać grupując razem pewne ich wartości. Proces taki nazywamy <b>kubełkowaniem</b>."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "<b>Kubełkowanie</b> (ang. *binning*) polega na łączeniu wartości zmiennych numerycznych w przedziały, ma to na celu zmniejszenie możliwości przeuczenia modelu.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wybór takich przedziałów może występować zarówno na podstawie cech rozkładu zmiennych (np wartości kwartyli) jak i wiedzy domenowej, tutaj zaprezentujemy obie możliwości."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T06:29:15.187589Z",
     "start_time": "2021-10-13T06:29:15.174595Z"
    }
   },
   "source": [
    "do podziału wg statystyk pozycyjnych z rozkładu wykorzystamy funkcję <b>qcut</b> z biblioteki pandas"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:26.802519Z",
     "start_time": "2021-10-13T07:17:26.750194Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>mileage_binned_4</th>\n",
       "      <th>mileage_binned_10</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>524486</td>\n",
       "      <td>(293524.25, 699892.0]</td>\n",
       "      <td>(429566.5, 699892.0]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>166856</td>\n",
       "      <td>(162418.5, 293524.25]</td>\n",
       "      <td>(162418.5, 212552.8]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>94191</td>\n",
       "      <td>(61161.25, 162418.5]</td>\n",
       "      <td>(78348.9, 116729.4]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>54210</td>\n",
       "      <td>(3784.999, 61161.25]</td>\n",
       "      <td>(48631.6, 78348.9]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>264049</td>\n",
       "      <td>(162418.5, 293524.25]</td>\n",
       "      <td>(212552.8, 265707.6]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>127914</td>\n",
       "      <td>(61161.25, 162418.5]</td>\n",
       "      <td>(116729.4, 162418.5]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>270</th>\n",
       "      <td>185880</td>\n",
       "      <td>(162418.5, 293524.25]</td>\n",
       "      <td>(162418.5, 212552.8]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>116011</td>\n",
       "      <td>(61161.25, 162418.5]</td>\n",
       "      <td>(78348.9, 116729.4]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>435</th>\n",
       "      <td>25782</td>\n",
       "      <td>(3784.999, 61161.25]</td>\n",
       "      <td>(25757.7, 48631.6]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>145116</td>\n",
       "      <td>(61161.25, 162418.5]</td>\n",
       "      <td>(116729.4, 162418.5]</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>750 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     mileage       mileage_binned_4     mileage_binned_10\n",
       "82    524486  (293524.25, 699892.0]  (429566.5, 699892.0]\n",
       "991   166856  (162418.5, 293524.25]  (162418.5, 212552.8]\n",
       "789    94191   (61161.25, 162418.5]   (78348.9, 116729.4]\n",
       "894    54210   (3784.999, 61161.25]    (48631.6, 78348.9]\n",
       "398   264049  (162418.5, 293524.25]  (212552.8, 265707.6]\n",
       "..       ...                    ...                   ...\n",
       "106   127914   (61161.25, 162418.5]  (116729.4, 162418.5]\n",
       "270   185880  (162418.5, 293524.25]  (162418.5, 212552.8]\n",
       "860   116011   (61161.25, 162418.5]   (78348.9, 116729.4]\n",
       "435    25782   (3784.999, 61161.25]    (25757.7, 48631.6]\n",
       "102   145116   (61161.25, 162418.5]  (116729.4, 162418.5]\n",
       "\n",
       "[750 rows x 3 columns]"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[\"mileage_binned_4\"] = pd.qcut(X_train[\"mileage\"], q=4)\n",
    "X_train[\"mileage_binned_10\"] = pd.qcut(X_train[\"mileage\"], q=10)\n",
    "X_train[[\"mileage\",\"mileage_binned_4\",\"mileage_binned_10\"]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Stworzyliśmy tutaj 2 nowe cechy na zbiorze treningowym w oparciu o kwartyle i decyle rozkładu zmiennej mileage. Jak widzimy poszczególnym wartościom przypisane zostały wartości przedziałów do których one wpadają w danym podziale na kubełki."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "jeśli chcemy samodzielnie określić etykiety przedziałów można posłużyć się parametrem labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:28.921670Z",
     "start_time": "2021-10-13T07:17:28.889069Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>mileage_binned_4</th>\n",
       "      <th>mileage_binned_10</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>524486</td>\n",
       "      <td>4</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>166856</td>\n",
       "      <td>3</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>94191</td>\n",
       "      <td>2</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>54210</td>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>264049</td>\n",
       "      <td>3</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>127914</td>\n",
       "      <td>2</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>270</th>\n",
       "      <td>185880</td>\n",
       "      <td>3</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>116011</td>\n",
       "      <td>2</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>435</th>\n",
       "      <td>25782</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>145116</td>\n",
       "      <td>2</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>750 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     mileage mileage_binned_4 mileage_binned_10\n",
       "82    524486                4                10\n",
       "991   166856                3                 6\n",
       "789    94191                2                 4\n",
       "894    54210                1                 3\n",
       "398   264049                3                 7\n",
       "..       ...              ...               ...\n",
       "106   127914                2                 5\n",
       "270   185880                3                 6\n",
       "860   116011                2                 4\n",
       "435    25782                1                 2\n",
       "102   145116                2                 5\n",
       "\n",
       "[750 rows x 3 columns]"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[\"mileage_binned_4\"] = pd.qcut(X_train[\"mileage\"], q=4, labels=np.arange(1,5))\n",
    "X_train[\"mileage_binned_10\"] = pd.qcut(X_train[\"mileage\"], q=10, labels=np.arange(1,11))\n",
    "X_train[[\"mileage\",\"mileage_binned_4\",\"mileage_binned_10\"]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-danger\">\n",
    "<b>Uwaga!</b> \n",
    "    \n",
    "Wszystkie przekształcenia wykorzystujące informacje o rozkładach cech należy implementować najpierw na zbiorze treningowym, a następnie w oparciu o rozkład ze zbioru treningowego - na zbiorze testowym. Inaczej wykorzystujemy informacje ze zbioru testowego i przestaje on być niezależny.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tutaj możemy skorzystać z parametru <b>retbins</b> i dostać granice kubełków a następnie w oparciu o te granice dokonać identycznego podziału na zbiorze testowym"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:31.262868Z",
     "start_time": "2021-10-13T07:17:31.238876Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([  3785.  ,  61161.25, 162418.5 , 293524.25, 699892.  ])"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "_, bins = pd.qcut(X_train[\"mileage\"], q=4, retbins=True)\n",
    "bins"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "alternatywnym podejściem jest użycie funkcji <b>cut</b> z biblioteki pandas, która pozwala podzielić wartości zmiennej na zdefiniowaną liczbę przedziałów o identycznej szerokości lub przypisać je do zdefiniowanych samodzielnie przedziałów"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "załóżmy, że jeśli chodzi o wiek samochodu znaczenie mają przedziały do 3 lat, od 3 do 7, od 7 do 12, 12-25 oraz ponad 25. Możemy łatwo dokonać takiego przypisania przydzielając poszczególnym kubełkom odpowiednie nazwy:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:33.515740Z",
     "start_time": "2021-10-13T07:17:33.488642Z"
    }
   },
   "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>age</th>\n",
       "      <th>age_binned</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>65</td>\n",
       "      <td>extremely_old</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>25</td>\n",
       "      <td>very_old</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>21</td>\n",
       "      <td>very_old</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>67</td>\n",
       "      <td>extremely_old</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>36</td>\n",
       "      <td>extremely_old</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>31</td>\n",
       "      <td>extremely_old</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>270</th>\n",
       "      <td>59</td>\n",
       "      <td>extremely_old</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>14</td>\n",
       "      <td>very_old</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>435</th>\n",
       "      <td>7</td>\n",
       "      <td>middle_age</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>42</td>\n",
       "      <td>extremely_old</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>750 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     age     age_binned\n",
       "82    65  extremely_old\n",
       "991   25       very_old\n",
       "789   21       very_old\n",
       "894   67  extremely_old\n",
       "398   36  extremely_old\n",
       "..   ...            ...\n",
       "106   31  extremely_old\n",
       "270   59  extremely_old\n",
       "860   14       very_old\n",
       "435    7     middle_age\n",
       "102   42  extremely_old\n",
       "\n",
       "[750 rows x 2 columns]"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[\"age_binned\"] = pd.cut(X_train[\"age\"],[0,3,7,12,25,100], labels =[\"new\",\"middle_age\",\"old\",\"very_old\",\"extremely_old\"] )\n",
    "X_train[[\"age\",\"age_binned\"]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Normalizacja i Standaryzacja"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Załóżmy, że finalnie mamy zbiór danych złożony stricte ze zmiennych numerycznych:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:35.815133Z",
     "start_time": "2021-10-13T07:17:35.791137Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>condition_transformed</th>\n",
       "      <th>age</th>\n",
       "      <th>avg_yearly_mileage</th>\n",
       "      <th>is_classic</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>524486</td>\n",
       "      <td>2.0</td>\n",
       "      <td>65</td>\n",
       "      <td>8069.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>166856</td>\n",
       "      <td>3.0</td>\n",
       "      <td>25</td>\n",
       "      <td>6674.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>94191</td>\n",
       "      <td>2.0</td>\n",
       "      <td>21</td>\n",
       "      <td>4485.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>54210</td>\n",
       "      <td>1.0</td>\n",
       "      <td>67</td>\n",
       "      <td>809.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>264049</td>\n",
       "      <td>4.0</td>\n",
       "      <td>36</td>\n",
       "      <td>7335.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>127914</td>\n",
       "      <td>3.0</td>\n",
       "      <td>31</td>\n",
       "      <td>4126.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>270</th>\n",
       "      <td>185880</td>\n",
       "      <td>3.0</td>\n",
       "      <td>59</td>\n",
       "      <td>3151.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>116011</td>\n",
       "      <td>0.0</td>\n",
       "      <td>14</td>\n",
       "      <td>8286.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>435</th>\n",
       "      <td>25782</td>\n",
       "      <td>2.0</td>\n",
       "      <td>7</td>\n",
       "      <td>3683.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>145116</td>\n",
       "      <td>2.0</td>\n",
       "      <td>42</td>\n",
       "      <td>3455.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>750 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     mileage  condition_transformed  age  avg_yearly_mileage  is_classic\n",
       "82    524486                    2.0   65              8069.0         0.0\n",
       "991   166856                    3.0   25              6674.0         0.0\n",
       "789    94191                    2.0   21              4485.0         0.0\n",
       "894    54210                    1.0   67               809.0         0.0\n",
       "398   264049                    4.0   36              7335.0         0.0\n",
       "..       ...                    ...  ...                 ...         ...\n",
       "106   127914                    3.0   31              4126.0         0.0\n",
       "270   185880                    3.0   59              3151.0         0.0\n",
       "860   116011                    0.0   14              8286.0         0.0\n",
       "435    25782                    2.0    7              3683.0         0.0\n",
       "102   145116                    2.0   42              3455.0         0.0\n",
       "\n",
       "[750 rows x 5 columns]"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train = X_train.loc[:,[\"mileage\",\"condition_transformed\", \"age\",\"avg_yearly_mileage\",\"is_classic\"]]\n",
    "X_test = X_test.loc[:,[\"mileage\",\"condition_transformed\", \"age\",\"avg_yearly_mileage\",\"is_classic\"]]\n",
    "X_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "możemy łatwo zaobserwować, że zmienne znacznie różnią się pod względem średniej czy wariancji:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:37.305187Z",
     "start_time": "2021-10-13T07:17:37.248078Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>condition_transformed</th>\n",
       "      <th>age</th>\n",
       "      <th>avg_yearly_mileage</th>\n",
       "      <th>is_classic</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>198670.12</td>\n",
       "      <td>2.41</td>\n",
       "      <td>36.94</td>\n",
       "      <td>5458.12</td>\n",
       "      <td>0.01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>158883.17</td>\n",
       "      <td>1.16</td>\n",
       "      <td>20.34</td>\n",
       "      <td>2766.60</td>\n",
       "      <td>0.12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>3785.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>2.00</td>\n",
       "      <td>118.00</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>61161.25</td>\n",
       "      <td>2.00</td>\n",
       "      <td>19.00</td>\n",
       "      <td>3306.75</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>162418.50</td>\n",
       "      <td>2.00</td>\n",
       "      <td>37.00</td>\n",
       "      <td>5615.00</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>293524.25</td>\n",
       "      <td>3.00</td>\n",
       "      <td>55.00</td>\n",
       "      <td>7816.25</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>699892.00</td>\n",
       "      <td>4.00</td>\n",
       "      <td>71.00</td>\n",
       "      <td>12205.00</td>\n",
       "      <td>1.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         mileage  condition_transformed     age  avg_yearly_mileage  \\\n",
       "count     750.00                 750.00  750.00              750.00   \n",
       "mean   198670.12                   2.41   36.94             5458.12   \n",
       "std    158883.17                   1.16   20.34             2766.60   \n",
       "min      3785.00                   0.00    2.00              118.00   \n",
       "25%     61161.25                   2.00   19.00             3306.75   \n",
       "50%    162418.50                   2.00   37.00             5615.00   \n",
       "75%    293524.25                   3.00   55.00             7816.25   \n",
       "max    699892.00                   4.00   71.00            12205.00   \n",
       "\n",
       "       is_classic  \n",
       "count      750.00  \n",
       "mean         0.01  \n",
       "std          0.12  \n",
       "min          0.00  \n",
       "25%          0.00  \n",
       "50%          0.00  \n",
       "75%          0.00  \n",
       "max          1.00  "
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.round(X_train.describe(),2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T12:57:38.502082Z",
     "start_time": "2021-10-12T12:57:38.483083Z"
    }
   },
   "source": [
    "Niektóre metody modelowania są zależne od odległości bądź wariancji zmiennych, więc chcąc aby wpływ danej zmiennej na predykcje wynikał przede wszystkim z jej związku ze zmienną celu, a nie ze skali, powinniśmy zastosować <b>normalizację</b> lub <b>standaryzację</b>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "zacznijmy od normalizacji, zrealizujemy ją za pomocą kolejnego transformera z biblioteki sklearn: MinMaxScaler"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T12:59:31.456780Z",
     "start_time": "2021-10-12T12:59:31.451782Z"
    }
   },
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "<b>Normalizacja</b> polega na przekształcaniu zmiennej do zakresu wartości <0,1> poprzez odjęcie minumum i podzielenie przez różnicę pomiędzy maksimum a minimum z rozkładu.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:40.148991Z",
     "start_time": "2021-10-13T07:17:40.135140Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import MinMaxScaler"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T05:15:34.463942Z",
     "start_time": "2021-10-13T05:15:34.450911Z"
    }
   },
   "source": [
    "tworzymy obiekt klasy MinMaxScaler korzystając z bazowych ustawień, możliwa jest zmiana docelowego zakresu wartości z bazowego <0,1> na dowolnie wybrany poprzez użycie parametru <b>feature_range</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:41.519658Z",
     "start_time": "2021-10-13T07:17:41.505312Z"
    }
   },
   "outputs": [],
   "source": [
    "mm = MinMaxScaler()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Metody są oczywiście analogiczne jak w poprzednich przypadkach, konwertujemy wyniki z powrotem na ramki danych aby łatwiej było je oglądać"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:42.980277Z",
     "start_time": "2021-10-13T07:17:42.922226Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>condition_transformed</th>\n",
       "      <th>age</th>\n",
       "      <th>avg_yearly_mileage</th>\n",
       "      <th>is_classic</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.28</td>\n",
       "      <td>0.60</td>\n",
       "      <td>0.51</td>\n",
       "      <td>0.44</td>\n",
       "      <td>0.01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>0.23</td>\n",
       "      <td>0.29</td>\n",
       "      <td>0.29</td>\n",
       "      <td>0.23</td>\n",
       "      <td>0.12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.08</td>\n",
       "      <td>0.50</td>\n",
       "      <td>0.25</td>\n",
       "      <td>0.26</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.23</td>\n",
       "      <td>0.50</td>\n",
       "      <td>0.51</td>\n",
       "      <td>0.45</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.42</td>\n",
       "      <td>0.75</td>\n",
       "      <td>0.77</td>\n",
       "      <td>0.64</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       mileage  condition_transformed     age  avg_yearly_mileage  is_classic\n",
       "count   750.00                 750.00  750.00              750.00      750.00\n",
       "mean      0.28                   0.60    0.51                0.44        0.01\n",
       "std       0.23                   0.29    0.29                0.23        0.12\n",
       "min       0.00                   0.00    0.00                0.00        0.00\n",
       "25%       0.08                   0.50    0.25                0.26        0.00\n",
       "50%       0.23                   0.50    0.51                0.45        0.00\n",
       "75%       0.42                   0.75    0.77                0.64        0.00\n",
       "max       1.00                   1.00    1.00                1.00        1.00"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mm.fit(X_train)\n",
    "X_train_mm_scaled = pd.DataFrame(mm.transform(X_train), columns=X_train.columns)\n",
    "X_test_mm_scaled = pd.DataFrame(mm.transform(X_test), columns=X_test.columns)\n",
    "np.round(X_train_mm_scaled.describe(),2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "jak widzimy wszystkie cechy zostały teraz przeniesione do tego samego zakresu wartości <0,1>, ale ich średnia i odchylenie standardowe nie są identyczne. Oczywiście nie mamy gwarancji, że na nowych danych nie pojawią się wartości spoza wykresu widzianego na zbiorze treningowym. Jeśli chcemy być pewni, że znormalizowane zmienne zachowają swój zakres wartości należy tworząc obiekt klasy MinMaxScaler ustawić parametr <b>clip</b> na True."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-danger\">\n",
    "<b>Uwaga!</b> \n",
    "    \n",
    "Nie zaleca się korzystania z MinMaxScaler dla zmiennych posiadających <b>wartości odstające</b>. Jeśli takie wartości nie zostaną wcześniej prawidłowo obsłużone, to większość \"normalnych\" wartości zmiennej będzie \"upchana\" w małym zakresie wartości co nie będzie sprzyjać jej wartości dla predykcji. \n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "zaprezentujemy teraz przykład standaryzacji z wykorzystaniem transformera StandardScaler z biblioteki sklearn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-12T13:00:59.929173Z",
     "start_time": "2021-10-12T13:00:59.914162Z"
    }
   },
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "<b>Standaryzacja</b> polega na przekształcaniu zmiennej do rozkładu o wartości oczekiwanej 0 i odchyleniu standardowym 1 poprzez odjęcie średniej i podzielenie przez odchylenie standardowe\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:46.352480Z",
     "start_time": "2021-10-13T07:17:46.341479Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import StandardScaler"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tworzymy obiekt klasy StandardScaler korzystając z bazowych ustawień"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:47.954486Z",
     "start_time": "2021-10-13T07:17:47.945485Z"
    }
   },
   "outputs": [],
   "source": [
    "ss =StandardScaler()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Metody rownież są analogiczne jak w poprzednich przypadkach, konwertujemy wyniki z powrotem na ramki danych aby łatwiej było je oglądać"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:17:49.512341Z",
     "start_time": "2021-10-13T07:17:49.423237Z"
    }
   },
   "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>mileage</th>\n",
       "      <th>condition_transformed</th>\n",
       "      <th>age</th>\n",
       "      <th>avg_yearly_mileage</th>\n",
       "      <th>is_classic</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "      <td>750.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.00</td>\n",
       "      <td>-0.00</td>\n",
       "      <td>-0.00</td>\n",
       "      <td>-0.00</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>-1.23</td>\n",
       "      <td>-2.08</td>\n",
       "      <td>-1.72</td>\n",
       "      <td>-1.93</td>\n",
       "      <td>-0.12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>-0.87</td>\n",
       "      <td>-0.35</td>\n",
       "      <td>-0.88</td>\n",
       "      <td>-0.78</td>\n",
       "      <td>-0.12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>-0.23</td>\n",
       "      <td>-0.35</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.06</td>\n",
       "      <td>-0.12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.60</td>\n",
       "      <td>0.51</td>\n",
       "      <td>0.89</td>\n",
       "      <td>0.85</td>\n",
       "      <td>-0.12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>3.16</td>\n",
       "      <td>1.37</td>\n",
       "      <td>1.68</td>\n",
       "      <td>2.44</td>\n",
       "      <td>8.20</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       mileage  condition_transformed     age  avg_yearly_mileage  is_classic\n",
       "count   750.00                 750.00  750.00              750.00      750.00\n",
       "mean      0.00                  -0.00   -0.00               -0.00        0.00\n",
       "std       1.00                   1.00    1.00                1.00        1.00\n",
       "min      -1.23                  -2.08   -1.72               -1.93       -0.12\n",
       "25%      -0.87                  -0.35   -0.88               -0.78       -0.12\n",
       "50%      -0.23                  -0.35    0.00                0.06       -0.12\n",
       "75%       0.60                   0.51    0.89                0.85       -0.12\n",
       "max       3.16                   1.37    1.68                2.44        8.20"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ss.fit(X_train)\n",
    "X_train_ss_scaled = pd.DataFrame(ss.transform(X_train), columns=X_train.columns)\n",
    "X_test_ss_scaled = pd.DataFrame(ss.transform(X_test), columns=X_test.columns)\n",
    "np.round(X_train_ss_scaled.describe(),2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T05:47:31.194364Z",
     "start_time": "2021-10-13T05:47:31.178364Z"
    }
   },
   "source": [
    "W przeciwieństwie do poprzedniego przekształcenia zmienne mają identyczną średnią i odchylenie standardowe, za to różnią sie zakresami wartości."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T05:50:25.897677Z",
     "start_time": "2021-10-13T05:50:25.880675Z"
    }
   },
   "source": [
    "<div class=\"alert alert-block alert-info\">\n",
    "<b>Info</b> \n",
    "\n",
    "Jako że StandardScaler opiera się na użyciu średniej i odchylenia standardowego z transformowanej zmiennej, również nie jest on wolny od wpływu wartości odstających. Jeśli chcemy być pewni, że wartości odstające nie będą miały wpływu na skalę przekształconej zmiennej można skorzystać z klasy <b>RobustScaler</b> \n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Łączenie różnych transformacji w ramach pipeline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "cofnijmy się teraz na chwilę do etapu generowania danych i zobaczmy jak można w łatwy i niezawodny sposób łączyć ze sobą różne transformacje z wykorzystaniem obiektów Pipeline oraz ColumnTransformer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-success\">\n",
    "<b>Definicja</b> \n",
    "\n",
    "<b>Pipeline</b> jest to przepływ danych przez ułożone w kolejności moduły wykonujące ustalone transformacje, zazwyczaj ostatnim elementem jest model predykcyjny\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T08:30:51.418449Z",
     "start_time": "2021-10-13T08:30:51.387402Z"
    }
   },
   "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>condition</th>\n",
       "      <th>mileage</th>\n",
       "      <th>brand</th>\n",
       "      <th>age</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>very_good</td>\n",
       "      <td>69205</td>\n",
       "      <td>Skoda</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>991</th>\n",
       "      <td>good</td>\n",
       "      <td>322639</td>\n",
       "      <td>VW</td>\n",
       "      <td>46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>789</th>\n",
       "      <td>good</td>\n",
       "      <td>86970</td>\n",
       "      <td>Fiat</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>894</th>\n",
       "      <td>very_good</td>\n",
       "      <td>524559</td>\n",
       "      <td>Fiat</td>\n",
       "      <td>66</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>398</th>\n",
       "      <td>medium</td>\n",
       "      <td>18954</td>\n",
       "      <td>Audi</td>\n",
       "      <td>35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>good</td>\n",
       "      <td>209892</td>\n",
       "      <td>Renault</td>\n",
       "      <td>38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>270</th>\n",
       "      <td>very_good</td>\n",
       "      <td>108168</td>\n",
       "      <td>Seat</td>\n",
       "      <td>51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>good</td>\n",
       "      <td>39632</td>\n",
       "      <td>Mercedes</td>\n",
       "      <td>59</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>435</th>\n",
       "      <td>medium</td>\n",
       "      <td>267041</td>\n",
       "      <td>Fiat</td>\n",
       "      <td>50</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>good</td>\n",
       "      <td>24438</td>\n",
       "      <td>Audi</td>\n",
       "      <td>26</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>750 rows × 4 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     condition  mileage     brand  age\n",
       "82   very_good    69205     Skoda    6\n",
       "991       good   322639        VW   46\n",
       "789       good    86970      Fiat   20\n",
       "894  very_good   524559      Fiat   66\n",
       "398     medium    18954      Audi   35\n",
       "..         ...      ...       ...  ...\n",
       "106       good   209892   Renault   38\n",
       "270  very_good   108168      Seat   51\n",
       "860       good    39632  Mercedes   59\n",
       "435     medium   267041      Fiat   50\n",
       "102       good    24438      Audi   26\n",
       "\n",
       "[750 rows x 4 columns]"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "UsedCars_df =generate_used_cars_data()\n",
    "X = UsedCars_df.drop(\"selling_price\",axis=1)\n",
    "y = UsedCars_df[\"selling_price\"]\n",
    "X[\"age\"] = 2021 - X[\"year_manufactured\"]\n",
    "X.drop(\"year_manufactured\", axis=1, inplace=True)\n",
    "X_train, X_test, y_train, y_test = train_test_split(X,y, test_size=0.25, random_state=42)\n",
    "X_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "zdefiniujmy teraz grupy zmiennych, które będziemy poddawać poszczególnym transformacjom:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T08:31:37.861867Z",
     "start_time": "2021-10-13T08:31:37.843794Z"
    }
   },
   "outputs": [],
   "source": [
    "columns_for_ordinal_encoding =[\"condition\"]\n",
    "columns_for_target_encoding =[\"brand\"]\n",
    "numerical_columns =[\"age\",\"mileage\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "następnie importujemy niezbędne klasy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T07:37:51.258106Z",
     "start_time": "2021-10-13T07:37:50.986228Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.compose import ColumnTransformer\n",
    "from sklearn.pipeline import Pipeline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "definiujemy pipeline do przetwarzania poszczególnych grup kolumn, pipeline mogą zawierać wiele kroków, tutaj dla uproszczenia wykorzystamy jednoelementowe"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T08:46:33.279578Z",
     "start_time": "2021-10-13T08:46:33.264583Z"
    }
   },
   "outputs": [],
   "source": [
    "pipeline_oe = Pipeline(steps =[(\"OrdinalEncoder\",\n",
    "                            OrdinalEncoder(\n",
    "                               categories =[['very_bad', 'bad', 'medium', 'good', 'very_good']],\n",
    "                               handle_unknown ='use_encoded_value', unknown_value=np.NaN))])\n",
    "\n",
    "pipeline_jse = Pipeline(steps=[(\"JamesSteinEncoder\",JamesSteinEncoder())])\n",
    "\n",
    "pipeline_num = Pipeline(steps=[(\"PowerTransformer\",PowerTransformer(standardize=False))])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T08:47:50.362158Z",
     "start_time": "2021-10-13T08:47:50.346529Z"
    }
   },
   "source": [
    "następnie przypisujemy zmienne do poszczególnych transformacji, korzystając z obiektu ColumnTransformer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T08:50:55.791722Z",
     "start_time": "2021-10-13T08:50:55.784725Z"
    }
   },
   "outputs": [],
   "source": [
    "column_transformer = ColumnTransformer(\n",
    "                        transformers=[\n",
    "                         ('categorical_oe', pipeline_oe, columns_for_ordinal_encoding),\n",
    "                         ('categorical_jse', pipeline_jse, columns_for_target_encoding),\n",
    "                         ('numerical', pipeline_num, numerical_columns)\n",
    "                         ])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "sam ColumnTransformer również może być częścią pipeline, przykładowo możemy na koniec zastosować standaryzacje"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T08:59:34.098331Z",
     "start_time": "2021-10-13T08:59:34.091331Z"
    }
   },
   "outputs": [],
   "source": [
    "preprocessing_pipeline = Pipeline(steps = [\n",
    "                        (\"column_transformer\", column_transformer),\n",
    "                        (\"scaler\",StandardScaler())\n",
    "                    ])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T09:08:30.550361Z",
     "start_time": "2021-10-13T09:08:30.533364Z"
    }
   },
   "source": [
    "jeśli używamy pipeline bez modelu predykcyjnego na końcu to stosujemy te same metody co przy zwykłych transformerach"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T08:59:37.863310Z",
     "start_time": "2021-10-13T08:59:37.797282Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Pipeline(steps=[('column_transformer',\n",
       "                 ColumnTransformer(transformers=[('categorical_oe',\n",
       "                                                  Pipeline(steps=[('OrdinalEncoder',\n",
       "                                                                   OrdinalEncoder(categories=[['very_bad',\n",
       "                                                                                               'bad',\n",
       "                                                                                               'medium',\n",
       "                                                                                               'good',\n",
       "                                                                                               'very_good']],\n",
       "                                                                                  handle_unknown='use_encoded_value',\n",
       "                                                                                  unknown_value=nan))]),\n",
       "                                                  ['condition']),\n",
       "                                                 ('categorical_jse',\n",
       "                                                  Pipeline(steps=[('JamesSteinEncoder',\n",
       "                                                                   JamesSteinEncoder())]),\n",
       "                                                  ['brand']),\n",
       "                                                 ('numerical',\n",
       "                                                  Pipeline(steps=[('PowerTransformer',\n",
       "                                                                   PowerTransformer(standardize=False))]),\n",
       "                                                  ['age', 'mileage'])])),\n",
       "                ('scaler', StandardScaler())])"
      ]
     },
     "execution_count": 81,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preprocessing_pipeline.fit(X_train,y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "po uruchomieniu metody fit widzimy wszystkie kroki całego pipeline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T09:04:04.397527Z",
     "start_time": "2021-10-13T09:04:04.362537Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "rozmiar zbioru treningowego po transformacji: (750, 4)\n",
      "rozmiar zbioru testowego po transformacji: (250, 4)\n",
      "[[ 1.34093122 -0.77016684 -1.5987142  -0.59924888]\n",
      " [ 0.46641086 -0.80925124  0.58475157  0.97365003]\n",
      " [ 0.46641086 -0.7939586  -0.68985786 -0.40448688]\n",
      " ...\n",
      " [ 0.46641086  0.70876962  1.13401097 -1.02756173]\n",
      " [-0.4081095  -0.7939586   0.75822769  0.74397369]\n",
      " [ 0.46641086  0.80959101 -0.36550292 -1.35095193]]\n"
     ]
    }
   ],
   "source": [
    "X_train_transformed = preprocessing_pipeline.transform(X_train)\n",
    "X_test_transformed = preprocessing_pipeline.transform(X_test)\n",
    "print(f\"rozmiar zbioru treningowego po transformacji: {X_train_transformed.shape}\")\n",
    "print(f\"rozmiar zbioru testowego po transformacji: {X_test_transformed.shape}\")\n",
    "print(X_train_transformed)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T09:11:50.317268Z",
     "start_time": "2021-10-13T09:11:50.299272Z"
    }
   },
   "source": [
    "jak widzimy zwrócony zbiór treningowy ma typ danych array a jego wymiary odpowiadają tym przed transformacją, zgodność wymiarów zależy jednak od zastosowanych transformacji"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "jeśli pipeline jest zakończony modelem predykcyjnym jego metody są identyczne jak metody modelu, czyli korzystamy z fit i predict, tak jak w poniższym przykładzie"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T09:15:56.032669Z",
     "start_time": "2021-10-13T09:15:56.026673Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.linear_model import LinearRegression\n",
    "\n",
    "final_pipeline = Pipeline(steps = [\n",
    "                (\"preprocessing\", preprocessing_pipeline),\n",
    "                (\"model\", LinearRegression())\n",
    "                ])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T09:19:56.002592Z",
     "start_time": "2021-10-13T09:19:55.993593Z"
    }
   },
   "source": [
    "poniżej pokazujemy jak wytrenować pipeline w oparciu o zbiór treningowy a następnie dokonać predykcji na zbiorze testowym"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-10-13T09:20:56.132669Z",
     "start_time": "2021-10-13T09:20:56.071678Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0     9649.925584\n",
       "1    30782.788094\n",
       "2    44681.214268\n",
       "3     5967.815692\n",
       "4    45684.785298\n",
       "Name: prediction, dtype: float64"
      ]
     },
     "execution_count": 91,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "final_pipeline.fit(X_train, y_train)\n",
    "X_test_predicted = pd.Series(final_pipeline.predict(X_test),name=\"prediction\")\n",
    "X_test_predicted.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Podsumowanie"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Powyżej zaprezentowaliśmy kilka różnych przykładów przekształcania surowych danych w cechy predykcyjne przydatne przy modelowaniu. Niektóre transformacje można z powodzeniem stosować przy pracy nad innymi problemami, inne należy raczej potraktować jako inspirację. Ważne aby prace nad Feature Engineering poprzedzone były dobrym zrozumieniem danych. Niezwykle ważne jest tutaj aby już na tym etapie stosować odpowiednią strategię walidacyjną, aby zapobiec korzystaniu z informacji ze zbioru testowego, inaczej nasze wyniki nie będą miarodajne."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.7"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {
    "height": "calc(100% - 180px)",
    "left": "10px",
    "top": "150px",
    "width": "426.667px"
   },
   "toc_section_display": true,
   "toc_window_display": true
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}