fix failed tests when data.table is enabled.

[exalate-issue-sync]

If data.table is enabled, make sure the following tests pass:

runit_GBM_quantile.R,
runit_GLM_libR_airlines.R,
runit_GLM_libR_prostate.R,
runit_GLM_link_functions_vpow.R,
runit_kmeans_scoring.R,
runit_RF_bigcat.R,
runit_RF_smallcat.R,
runit_Rdoc_set_time_zone.R,
runit_nidhi_part.R,
runit_hex_1841_asdate_datemanipulation.R,
runit_pubdev_svm_bug.R,
runit_pubdev_svm_bug2.R,
runit_empty_strings.R,
runit_logical_and_or.R,
runit_string_distance.R,
runit_string_manipulations.R,
runit_weighted_quantile.R,
runit_merge_random_data.R,
runit_lstrip.R,
runit_rstrip.R,
runit_mean_axis.R,
runit_round_signif.R,
runit_sum_axis.R

[exalate-issue-sync]

Erin LeDell commented: Hey [~accountid:557058:393936ef-8683-427b-babb-14ffad4bb6d7] we’ve been trying to get data.table support in H2O for the past few years (so that as.h2o() uses fwrite under the hood) and there are a number of tests that fail when we turn it on by default. [~accountid:557058:1f01b471-f37b-40af-bae9-a18b38e24549] is going to make another PR to turn it on by default and we need your help to debug the tests that are not working (they are mostly data.table errors). Thanks!

[exalate-issue-sync]

Wendy commented: Matt/Erin:

The Jenkin tests actually look pretty good. I will paste the new failed tests here if they appear. Give me a few to gather the information.

Thanks, Wendy

[exalate-issue-sync]

Wendy commented: Matt:

I found this two tests failing off the PR here:

• [+r_suite.runit_GBM_weight_gamma.R+|http://mr-0xc1:8080/job/h2o-3-pipeline/job/PR-4265/16/testReport/junit/(root)/r_suite/runit_GBM_weight_gamma_R/]
• [+r_suite.runit_GLM_weights2.R+|http://mr-0xc1:8080/job/h2o-3-pipeline/job/PR-4265/16/testReport/junit/(root)/r_suite/runit_GLM_weights2_R/]

You can get to the error messages from here:

[http://mr-0xc1:8080/job/h2o-3-pipeline/job/PR-4265/16/#showFailuresLink|http://mr-0xc1:8080/job/h2o-3-pipeline/job/PR-4265/16/#showFailuresLink]

[exalate-issue-sync]

Matt Dowle commented: Thanks Wendy & Erin. I traced it through.

Yes: edge case bug in data.table. Now fixed in data.table 1.12.9. [https://github.com/Rdatatable/data.table/pull/4238|https://github.com/Rdatatable/data.table/pull/4238|smart-link]

I filed a one-line workaround to h2o-3: [https://github.com//pull/4308|https://github.com//pull/4308|smart-link].

Please merge PR 4308 to h2o-3 and rerun.

[hasithjp]

JIRA Issue Migration Info

Jira Issue: PUBDEV-4823
Assignee: Wendy
Reporter: Wendy
State: Closed
Fix Version: N/A
Attachments: N/A
Development PRs: Available

Linked PRs from JIRA

#4308
#1471