WHen the option "reproducible" is not used in h2o.deeplearning the results are worse than when it is used

Description

When the option reproducible is not used in h2o.deeplearning the results are worse than when reproducible=TRUE.
The errors when doing a cross validation with 5 folds are 30% worse in average than when only one processor is used.

Assignee

New H2O Bugs

Fix versions

None

Reporter

jairo Rocha

Support ticket URL

None

Affected Spark version

None

Customer Request Type

None

Task progress

None

ReleaseNotesHidden

None

AffectedContact

CustomerVisible

No

Components

Priority

Major
Configure