import pandas as pd
import sklearn.preprocessing
from sklearn_pandas import DataFrameMapper

df = pd.DataFrame({'col1': [0, 0, 1, 1, 2, 3, 0], 'col2': [0, 0, 1, 1, 2, 3, 0]})
mapper = DataFrameMapper([
    (['col1', 'col2'], sklearn.preprocessing.StandardScaler()),
    (['col1', 'col2'], sklearn.preprocessing.OneHotEncoder()),
], df_out=True)
print(mapper.fit_transform(df).columns)

from sklearn_pandas import DataFrameMapper
from sklearn.preprocessing import PolynomialFeatures  

df = pd.DataFrame({'col1': [0, 0, 1, 1, 2, 3, 0], 'col2': [0, 0, 1, 1, 2, 3, 0]})
poly = PolynomialFeatures(degree=2, include_bias=False)
mapper = DataFrameMapper([
    (['col1', 'col2'], poly),
], df_out=True)

print(mapper.fit_transform(df).columns)
>>> FutureWarning: Function get_feature_names is deprecated; get_feature_names is deprecated in 1.0 and will be removed in 1.2. Please use get_feature_names_out instead.
  warnings.warn(msg, category=FutureWarning)
>>> ['col1_col2_x0', 'col1_col2_x1', 'col1_col2_x0^2', 'col1_col2_x0 x1', 'col1_col2_x1^2']

print(poly.get_feature_names_out(['col1', 'col2']))
>>> ['col1' 'col2' 'col1^2' 'col1 col2' 'col2^2']

def _fix_column_names(df: pd.DataFrame, mapper: DataFrameMapper) -> pd.DataFrame:
    for columns, transformer, kwargs in mapper.built_features:
        if (isinstance(transformer, OneHotEncoder)
                or (isinstance(transformer, Pipeline) and any(isinstance(t, OneHotEncoder) for t in transformer))):
            assert isinstance(columns, Iterable) and not isinstance(columns, str)

            new_names = transformer.get_feature_names_out(columns)

            old_name_prefix = kwargs.get("alias", "_".join(str(c) for c in columns))
            old_names = [f"{old_name_prefix}_{i}" for i in range(len(new_names))]

            df = df.rename(columns=dict(zip(old_names, new_names)))
        elif isinstance(transformer, Pipeline) and isinstance(transformer[0], MultiLabelBinarizer):
            # The way sklearn-pandas infers the names is by iterating the transformers and getting the names and trying
            # to get the features names that are available from the last one that has them. Then, it checks if their
            # length matches the output number of features. However, if the binarizer is followed by feature selection,
            # this process fails as the previous condition is not met. So we handle it manually here.
            assert isinstance(columns, str)

            # `MultiLabelBinarizer` doesn't implement `get_feature_names_out`.
            new_names = [f"{columns}_{c}" for c in transformer[0].classes_]

            # We slice as an iterator and not by passing a slice to `__getitem__` because if the transformer is of type
            # `TransformerPipeline` then it fails.
            for t in itertools.islice(transformer, 1, None):
                new_names = t.get_feature_names_out(new_names)

            old_name_prefix = kwargs.get("alias", columns)
            old_names = [f"{old_name_prefix}_{i}" for i in range(len(new_names))]

            df = df.rename(columns=dict(zip(old_names, new_names)))

    return df