java.lang.NoSuchMethodError since openjdk 1.8.0_252

Description

Since openjdk version 1.8.0_252 we are unable to launch an H2O cluster on EMR due to the following error (happens when we call h2o_context(sc) from R):

This works fine on one of our older clusters, where openjdk version is 1.8.0_242.

Other details:
H2O build version: 3.28.0.1 (yu)
Sparkling Water version: 3.28.0.1-1-2.4
Spark build version: 2.4.4
Scala version: 2.11

Activity

Show:
Krister Jaanhold
August 12, 2020, 8:34 AM
Edited

Sorry, was off for some time Sharing the snippet below:

Jakub Hava
August 6, 2020, 8:55 AM

Closing this issue due to inactivity

Jakub Hava
June 29, 2020, 2:35 PM

Thank you! Would it be possible to share the snipper were you configure your algorithm?

Also CC: and

Krister Jaanhold
June 26, 2020, 2:05 PM
Edited

We are using AUCPR with cross-validation and early stopping for tuning the number of trees for XGBoost. At some point, the interpolation of AUCPR got updated (was linear initially). However, after that change we’re seeing convergence way later (e.g. 2k trees instead of 500 trees). Furthermore – we did shadow mode testing on live data and noticed that models trained with the “old” way of calculating AUCPR actually perform way better. This also had implications on training speed (needed roughly 4x more time).

Jakub Hava
June 26, 2020, 12:39 PM

Thank you for your reply! Could you please share what kind of issue are you having on the latest version? Maybe we can help with that.

Thanks!

Assignee

Unassigned

Reporter

Krister Jaanhold

Labels

None

CustomerVisible

No

testcase 1

None

testcase 2

None

testcase 3

None

h2ostream link

None

Affected Spark version

None

AffectedContact

None

AffectedCustomers

None

AffectedPilots

None

AffectedOpenSource

None

Support Assessment

None

Customer Request Type

None

Support ticket URL

None

End date

None

Baseline start date

None

Baseline end date

None

Task progress

None

Task mode

None

ReleaseNotesHidden

None

Affects versions

Priority

Blocker