如何在scikit-learn的FunctionTransformer中使用feature_names_out？

错误原因 #

feature_names_out是scikit-learn 1.2.0版本才新增的参数，你的scikit-learn版本低于1.2.0，所以初始化FunctionTransformer时会抛出TypeError: __init__() got an unexpected keyword argument 'feature_names_out'。

解决方案 #

方案1：升级scikit-learn版本 #

执行以下命令将scikit-learn升级到1.2.0及以上版本：

pip install --upgrade scikit-learn

升级后，原代码即可正常运行：

from sklearn.preprocessing import FunctionTransformer
import pandas as pd
import numpy as np

X = pd.Series(data=[1, 2, 3], name='numbers')
transformer = FunctionTransformer(np.log, feature_names_out='one-to-one')
transformed_X = transformer.transform(X)
print(transformer.get_feature_names_out())  # 输出['numbers']

方案2：不升级版本，自定义特征名返回逻辑 #

如果无法升级版本，可以通过继承FunctionTransformer自定义get_feature_names_out方法：

from sklearn.preprocessing import FunctionTransformer
import pandas as pd
import numpy as np

class CustomFunctionTransformer(FunctionTransformer):
    def get_feature_names_out(self, input_features=None):
        # 直接返回输入的特征名，实现one-to-one的效果
        return input_features if input_features is not None else []

X = pd.Series(data=[1, 2, 3], name='numbers')
transformer = CustomFunctionTransformer(np.log)
transformed_X = transformer.transform(X)
print(transformer.get_feature_names_out(['numbers']))  # 输出['numbers']

内容的提问来源于stack exchange，提问作者saad