[FIXED] Specifying a suffixing function in feature_names_out argument of the functiontransformer function in scikit-learn

Issue

I want to transform a series of variables 'x1', 'x2' with a squared function using FunctionTransformer and rename the feature names with a suffix 's2_'. For example, 'x1' would become 's2_x1' in the transformed data set. I have the following code:

import pandas as pd
from sklearn.preprocessing import FunctionTransformer
from sklearn import set_config

set_config(transform_output='pandas') # This ensures the transformed output is a dataframe

df = pd.DataFrame(
    {
        'x1' : [1, 2, 3],
        'x2' : [2, 3, 4]
    }
)

my_transformer = FunctionTransformer(lambda x: x**2,
                    feature_names_out= lambda x : [f's2_{c}' for c in x])

transformed_df = my_transformer.fit_transform(df)
transformed_df

However, the renaming did not work as expected. The columns names remained 'x1' and 'x2' How should I fix the code?

Solution

According to the FunctionTransformer docs, the feature_names_out argument

must take two positional arguments: this FunctionTransformer (self) and an array-like of input feature names (input_features).

In your case, you were missing the argument for the FunctionTransformer itself. Try this:

my_transformer = FunctionTransformer(lambda x: x ** 2,
                                     feature_names_out=lambda self, input_features: [f's2_{c}' for c in input_features])

Result:

>>> transformed_df
   s2_x1  s2_x2
0      1      4
1      4      9
2      9     16

Answered By - Georgi