Machine Learning : M-Estimate

m_estimate_encoder.py

for col in df.columns :
    if df[col].dtype == "object" and df[col].nunique() > 10 :
        print(col) #카테고리가 많은 feature일수록 target encoding이 필요하다
        if df[col].value_counts().any() < 5 : 
            print("required smoothing") #rare category가 존재하는 feature들은 smoothing을 적용한다.
            

#encoding split이랑 train split 나누기
X_encode = df.sample(frac=0.20, random_state=0)#인코더 fitting용, encoding split
y_encode = X_encode.pop("target")#인코더로 transfrom할거, train split
X_pretrain = X.drop(X_encode.index)
y_train = y[X_pretrain.index]

from category_encoders import MEstimateEncoder
#해당 feature로 인코더 만들기
encoder = MEstimateEncoder(cols=["feature_1"], m=5.0)

# encoding split에 fit하기
encoder.fit(X_encode, y_encode)

# encoder로 train split 변환
X_train = encoder.transform(X_pretrain)