sklearn管道正确使用

我在python中有一个数据帧,它有一个名为“datetime”的日期时间字段。使用管道和FeatureUnion,我试图提取day、month、weekday和isBusinessday。为了提取这些特性,我编写了自定义代码。

class itemselector(BaseEstimator, TransformerMixin):
    def __init__(self, key):
        self.key = key

    def transform(self, X):
        return X[self.key]

    def fit(self, X, y=None):
        return self


    f_df = Pipeline([

       ('union', FeatureUnion([
    ('date', Pipeline([
        ('sitem', itemselector('pickup_datetime')),
        ('sday', Extract_date()),
    ])),
    ('month', Pipeline([
        ('sitem', itemselector('pickup_datetime')),
        ('smonth', Extract_month()),
    ])),
])),

    ])

当我运行这段代码时,我会得到一个列表作为输出。例如:

df = f_df.fit_transform(df_train[:5])

输出:

[14 12 19  6 26  3  6  1  4  3]  // it has both day and month.  it is not expected output 

但我是一个日月分离的人。我该怎么做?我的代码出了什么问题?谁能帮我找到它吗?

更新

总结一下我的问题,我得到了输出形状 (10,) (5,2)

更新1

class Extract_date(BaseEstimator, TransformerMixin):
    def fit(self, X):
        print('one')
        return self

    def transform(self, X):
        return X.apply(lambda y: y.day)


class Extract_month(BaseEstimator, TransformerMixin):
    def fit(self, X, **atr):
        print('two')
        return self

    def transform(self, X):
        return X.apply(lambda y: y.month)