Closed desilinguist closed 4 years ago
scikit-learn 0.23.1 introduces a new attribute for (almost) all estimators called n_features_in_
which keeps tracks of how many features were actually passed to fit()
for that estimator. Since this attribute is not present in older SKLL (and underlying scikit-learn) models, this is a breaking change.
Although, we can try to introduce a workaround by trying to dynamically set this value when loading the model file but it seems pretty hacky and unnecessary. This means, that SKLL v2.5 will be a breaking release that will not work with models trained using previous SKLL versions. All older models will need to be retrained.
See full release notes for v0.23.1 here.