mlxtend
mlxtend copied to clipboard
stacking, named base models
It would be nice if the base models in the stacking ensembles explicitly required names, similar to sklearn's Pipeline named_steps and ColumnTransformer. When all my base models are themselves pipelines, digging into the nest to find actual names becomes tedious. (Especially in the context of set_params and grid searches, where the default behavior requires things like pipeline-1__...
: http://rasbt.github.io/mlxtend/user_guide/classifier/StackingCVClassifier/#example-3-stacked-cv-classification-and-gridsearch )
Maybe it's too late to try to change the API to require names as in ColumnTransformer, but perhaps as an optional parameter?
I agree that this can be helpful, especially for large sets of classifiers. I think having an optional way for naming them would be nice.