2.1 One hot encoding
2.2 Count and Frequency encoding
2.3 Target encoding / Mean encoding
2.4 Ordinal encoding
2.5 Weight of Evidence
2.6 Rare label encoding
2.7 Helmert encoding
2.8 Probability Ratio Encoding
2.9 Label encoding
2.10 Feature hashing
2.11 Binary encoding & BaseN encoding

將使用這個data-frame，有兩個獨立變數或特徵(features)和一個標籤(label or Target)，共有十筆資料。

import pandas as pd
import numpy as np
data = {'Temperature': ['Hot','Cold','Very Hot','Warm','Hot','Warm','Warm','Hot','Hot','Cold'],
        'Color': ['Red','Yellow','Blue','Blue','Red','Yellow','Red','Yellow','Yellow','Yellow'],
        'Target':[1,1,1,0,1,0,1,0,1,1]}

df = pd.DataFrame(data, columns = ['Temperature', 'Color', 'Target'])

補充 2.5 Weight of Evidence 加上程式範例
首先計算 Temperature 的每個類別中，屬於 Target(1) 及 Target(0) 的百分比。例如 Hot 類別中 Target是1， 有三筆；Target是0，有一筆。所以在Hot類別中，target=1的百分比是0.75，target=0的百分比是0.25。

# target = 1 i.e. Good = 1
woe_df = df.groupby('Temperature')['Target'].mean()
woe_df = pd.DataFrame(woe_df)
# remove the column name 'Target' to 'Good'
woe_df = woe_df.rename(columns={'Target':'Good'})
# Calculate Bad probability :  1 - Good probability
woe_df['Bad'] = 1-woe_df.Good
# add a small vlaue to avoid divide by zero in denominator 
# 加入一數值 避免被除數=0
woe_df['Bad'] = np.where(woe_df['Bad']==0, 0.000001, woe_df['Bad'])
# Compute the WoE
woe_df['WoE'] = np.log(woe_df.Good/woe_df.Bad)
woe_df

計算出WOE，我們將WOE值加入原來資料中

# Map the WOE value back to each row of data-frame
# 將 WOE 加入資料集的每一筆資料
df.loc[:, 'WoE_Encode'] = df['Temperature'].map(woe_df['WoE'])
df

2.8 Probability Ratio Encoding

Probability Ratio Encoding 類似 Weight of Evidence(WoE)，唯一的不同是這個方法使用比例(Ratio)而不是自然對數(Natural Log)。

# target = 1 i.e. Good = 1
pr_df = df.groupby('Temperature')['Target'].mean()
pr_df = pd.DataFrame(pr_df)
# remove the column name 'Target' to 'Good'
pr_df = pr_df.rename(columns={'Target':'Good'})
# Calculate Bad probability :  1 - Good probability
pr_df['Bad'] = 1-pr_df.Good
# add a small vlaue to avoid divide by zero in denominator 
# 加入一數值 避免被除數=0
pr_df['Bad'] = np.where(pr_df['Bad']==0, 0.000001, pr_df['Bad'])
# Compute the Probability Ratio
pr_df['PR'] = pr_df.Good/pr_df.Bad
pr_df

計算出Probability Ratio value，我們將值加入原來資料中

# Map the Probability Ratio value back to each row of data-frame
# 將 Probability Ratio value 加入資料集的每一筆資料
df.loc[:, 'PR_Encode'] = df['Temperature'].map(pr_df['PR'])
df

2.9 Label encoding

這個方法給每個類別一個1到N數字，N個是類別的總數。這個方法有一個缺點是，即使類別之間沒有順序等關係，這個方法仍會認為類別間有順序或其他關係存在。例如下面例子看起來似乎有(Cold < Hot < Very Hot < Warm...0 < 1< 2 < 3)關係存在。

使用Scikit-learn

from sklearn.preprocessing import LabelEncoder
df['Temp_label_encoded'] = LabelEncoder().fit_transform(df.Temperature)
df

df.loc[:, 'Temp_factorize_encode'] = pd.factorize(df['Temperature'])[0].reshape(-1,1)
df