Enable running MOJO spark pipeline without H2O init

[exalate-issue-sync]

Usecase: train a pipeline containing an h2o stage, store it to file. Then read it from a file in another script and use it for scoring without explicitly initialising H2O (like with similar calls: H2OConf(spark))

Without that an exception would be thrown:
java.lang.ClassNotFoundException: org.apache.spark.ml.h2o.models.H2OMOJOModel

[exalate-issue-sync]

Jakub Hava commented: After analyzing this, loading the classes can't be handled automatically. Therefore we can document this behavior and say that user needs to initialize the PySparkling in this case using the Initializer.load_sparkling_jar call

CC: [~accountid:5a73cf5fdca0242a1ca84b4b]

[DinukaH2O]

JIRA Issue Migration Info

Jira Issue: SW-817
Assignee: Jakub Hava
Reporter: Stefan Pacinda
State: Resolved
Fix Version: 2.1.28
Attachments: N/A
Development PRs: Available

Linked PRs from JIRA

#696
#697

[hasithjp]

JIRA Issue Migration Info Cont'd

Jira Issue Created Date: 2018-04-24T03:33:59.365-0700