Default StackedEnsemble prediction fails when applied to a test dataset without response column

Description

based on Github ticket: https://github.com/h2oai/h2o-tutorials/issues/107

could reproduce using following code and the santander dataset:

After having previously modified scoring SE scoring so that it delegates correctly its logic to metalearner algo implementation (cf. #PUBDEV-6208), GLM (default metalearner algo) actually ignored the computeMetrics parameter, leading to an exception when the response column is missing.

Will write tests both for SE and for GLM as the root of the issue is there.

Assignee

Sebastien Poirier

Fix versions

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

Blocker
Configure