Broken Mojo when model trained with non-predictor column (e.g. fold_column) and a categorical encoding generating new columns (e.g. OHE)

Description

Currently some Mojo categoricalEncoding that create new columns (binary, onehot...) don't put the non-predictors columns (e.g. fold) to the end (wrong offset).
Contrary to what is done by the encoders during training, therefore the mojo fails if the model was trained with a fold column for example.

see test_pubdev_7775 in TargetEncoderPreprocessorTest (discovered there, but the failing test doesn’t use TE).

Assignee

New H2O Bugs

Fix versions

None

Reporter

Sebastien Poirier

Support ticket URL

None

Labels

None

Affected Spark version

None

Customer Request Type

None

Task progress

None

ReleaseNotesHidden

None

CustomerVisible

No

Components

Priority

Major
Configure