Follow-up from #15.
When predict_method="predict_proba" is used with a River classifier, the output ch axis is labeled with np.arange(n_classes) because River models don't expose classes_. However, predict_proba_many returns a DataFrame whose columns are the actual class labels, so if the classes aren't 0..n-1 the output channels are mislabeled.
Suggested fix: capture df.columns in the predict_proba_many branch of SklearnModelProcessor._process and use those values for the ch CoordinateAxis instead of falling back to np.arange.
Follow-up from #15.
When
predict_method="predict_proba"is used with a River classifier, the outputchaxis is labeled withnp.arange(n_classes)because River models don't exposeclasses_. However,predict_proba_manyreturns a DataFrame whose columns are the actual class labels, so if the classes aren't0..n-1the output channels are mislabeled.Suggested fix: capture
df.columnsin thepredict_proba_manybranch ofSklearnModelProcessor._processand use those values for thechCoordinateAxisinstead of falling back tonp.arange.