在sklearn pipeline中对指定列应用KBinsDiscretizer返回DataFrame报错求助

报错触发是因为ColumnTransformer的目标列传入了单个字符串'numeric_col_to_change'，sklearn会将该列作为1D的Series传入后续的KBinsDiscretizer，而该转换器要求输入必须为2D结构（哪怕仅有1列也需要是(n_samples, 1)的格式）。

另外当前的自定义PandasColumnTransformer存在两个隐藏问题：

• 未设置remainder='passthrough'，会默认丢弃未指定变换的列
• 输出列名直接复用原输入的列名，若列顺序发生变化会导致列名错位

1. 将单列选择改为列表格式，避免传入1D数据
2. 给PandasColumnTransformer添加remainder='passthrough'参数保留原始列
3. 优化PandasColumnTransformer的列名生成逻辑，适配列顺序变化

import pandas as pd
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import KBinsDiscretizer
from sklearn.pipeline import Pipeline


class PandasColumnTransformer(ColumnTransformer):
    def get_feature_names_out(self, input_features=None):
        feature_names = []
        for name, trans, cols, _ in self._iter(fitted=True):
            if trans == 'drop':
                continue
            elif trans == 'passthrough':
                feature_names.extend(cols)
            else:
                feature_names.extend(trans.get_feature_names_out(cols))
        return feature_names

    def transform(self, X: pd.DataFrame) -> pd.DataFrame:
        return pd.DataFrame(super().transform(X), columns=self.get_feature_names_out(), index=X.index)

    def fit_transform(self, X: pd.DataFrame, y=None) -> pd.DataFrame:
        return pd.DataFrame(super().fit_transform(X, y), columns=self.get_feature_names_out(), index=X.index)


class PandasKBinsDiscretizer(KBinsDiscretizer):
    def __init__(self, n_bins):
        super().__init__(n_bins, encode='ordinal')

    def get_feature_names_out(self, input_features=None):
        return input_features

    def transform(self, X):
        self.col_names = list(X.columns.values)
        X = super().transform(X)
        return pd.DataFrame(X, columns=self.col_names)


binner_on_numeric = PandasColumnTransformer(
    transformers=[
        ("binner", PandasKBinsDiscretizer(2), ['numeric_col_to_change'])
    ],
    remainder='passthrough' # 保留未指定变换的列
)


pp = Pipeline([('binner_just_numeric', binner_on_numeric)])

d = {'numeric_col_not_to_change': [1, 2, 1, 2, 1, 2],
     'numeric_col_to_change': [1, 2, 3, 4, 5, 6]}

df = pd.DataFrame(data=d)

res = pp.fit_transform(df)

assert isinstance(res, pd.DataFrame)

运行后输出的res为标准pandas DataFrame格式，numeric_col_to_change会被转换为0/1的分箱结果，其余列保持原始值，assert语句不会触发报错。

内容的提问来源于stack exchange，提问作者Yehoshaphat Schellekens