FeatureUnion 跟 pipeline 真的好用，可以暴力法選出特徵設定跟模型的參數

隨意找個介紹
https://www.kaggle.com/baghern/a-deep-dive-into-sklearn-pipelines

from sklearn.base import BaseEstimator, TransformerMixin

class RollingWindow(BaseEstimator, TransformerMixin):
    def __init__(self, window):
        self.window = window
        
    def transform(self, X, y=None):
        for col in X.columns:
            data=pd.merge(data, data[col].rolling(window=self.window).agg(['sum','std','mean','max','min','median','kurt','skew']),on='date',suffixes=('','_'+col) )
        data=data.dropna(axis=1,how='all')
        data=data[self.window-1:]
        data=data.fillna(0)
        return data
    
    def fit(self, X, y=None):
        return self 

pipeline=Pipeline([
            ('Rolling',RollingWindow() ),
            ('GBR',GradientBoostingRegressor() ),
            ])   
param = { 
            'Rolling__window':[4,5,6],
            'GBR__n_estimators': [100,200,300]
         }
         
         
clf = GridSearchCV(pipeline, param, cv=5)
 
clf.fit(X_train, y_train)

大概是這樣，有空再寫詳細一點