[TE] When predicting for the row with unseen during training cat. level we need to do blending between posterior for NAs in training frame and prior of the response

Description

This bug is only related to an imputation for unseen categorical levels when blending is enabled and when there actually were missing values in the training frame on which TargetEncoderModel was trained.

Blending in TargetEncoderMojoModel model is done correctly but behaviour in original TargetEncoderModel is different and it would affect test in case we set inflection_point to be bigger than number of missing values in the training frame.

Environment

None

Status

Assignee

Andrey Spiridonov

Fix versions

Reporter

Andrey Spiridonov

Support ticket URL

None

Labels

None

Release Priority

None

Affected Spark version

None

Customer Request Type

None

Task progress

None

CustomerVisible

No

Priority

Major
Configure