I think we can have get_feature_names_out always return "isotonicregression0".

Seems fine.

Is there a reason why you don't use the _ClassNamePrefixFeaturesOutMixin machinery and set self._n_features_out = 1 ?

Is there a reason why you don't use the _ClassNamePrefixFeaturesOutMixin machinery and set self._n_features_out = 1 ?

I wanted to adjust the docstring for IsotonicRegression.get_feature_names_out to show that it ignores input_features. IsotonicRegression does not define feature_names_in_, which means input_features are ignored.

The alternative is to add support for feature_names_in_ and n_features_in_ for IsotonicRegression and then use the mixin:

1. For the 2d case, feature_names_in_ and n_features_in_ are defined as usual. This means get_feature_names_out will actually validate input_features
2. For the 1d case, the semantics are not clear. Maybe n_features_in_ = 1 and feature_names_in_ is undefined or None. get_feature_names_out will not validate input_features.

Thanks for the clarification. I don't really know what we should expect in the 1d case either. Maybe it's better to stick to your current workaround for now and come back to implementing n_features_in_ when we have a clear expectation.