Both LightGBM and XGBoost treat NaN as missing values. This is in conflict with (J)PMML conventions, which treat NaN as invalid values.
The solutiion would be to generate MiningField@invalidValueTreatment="asMissing" attributes, ehich would cause the model to promote all invalid values (including NaN) to missing values.
See https://github.com/jpmml/jpmml-lightgbm/issues/33
Both LightGBM and XGBoost treat
NaN
as missing values. This is in conflict with (J)PMML conventions, which treatNaN
as invalid values.The solutiion would be to generate
MiningField@invalidValueTreatment="asMissing"
attributes, ehich would cause the model to promote all invalid values (includingNaN
) to missing values.