The SkLearn2PMML/JPMML-SkLearn stack should fail with an informative message if it's being used with a Scikit-Learn version that is newer than the "last stable Scikit-Learn version at the date of SkLearn2PMML package release". Should also run similar checks for child modules (eg. LightGBM, XGBoost, H2O.ai, StatsModels).
Alternatively, the SkLearn2PMML package could perform a "is there a newer version of myself available at PyPI?"-style check.
The first mechanism would be quick and work offline, whereas the second one would be more intrusive.
The SkLearn2PMML/JPMML-SkLearn stack should fail with an informative message if it's being used with a Scikit-Learn version that is newer than the "last stable Scikit-Learn version at the date of SkLearn2PMML package release". Should also run similar checks for child modules (eg. LightGBM, XGBoost, H2O.ai, StatsModels).
Alternatively, the SkLearn2PMML package could perform a "is there a newer version of myself available at PyPI?"-style check.
The first mechanism would be quick and work offline, whereas the second one would be more intrusive.