[ci] [R-package] add unit tests on monotone constraints

[jameslamb]

Contributes to #4345, see #4346 (comment).

This PR proposes adding unit tests on monotone constraints to the R package, since there currently are not any.

This test is basically a translation of the corresponding Python test into R, but simplified by leaving out details about interaction constraints. The goal of this PR is to have at least basic coverage of monotone constraints from the R package.

LightGBM/tests/python_package_test/test_engine.py

Line 1256 in bab58d0

def test_monotone_constraints(test_with_interaction_constraints):

[jameslamb]

/gha run r-valgrind

Workflow R valgrind tests has been triggered! 🚀
https://github.com/microsoft/LightGBM/actions/runs/920224591

Status: success ✔️.

[jameslamb]

/gha run r-valgrind

Workflow R valgrind tests has been triggered! 🚀
https://github.com/microsoft/LightGBM/actions/runs/921982989

Status: failure ❌.

[jameslamb]

@StrikerRUS I wanted to test the results of the valgrind job when removing testthat::skip() calls from the R package's tests, to compare it to the results I'm seeing in #4353. I'm doing that here on this unrelated PR, with #4352 (comment), because I couldn't get the optional checks working on my fork.

Could you please tell me what secrets I need to set up on my fork to be able to run the triggering_comments jobs there, like you did in #3740 (comment)?

I tried setting a secret SECRETS_WORKFLOW in the settings on my fork to a GitHub personal access token and to the client token for a GitHub oauth app. In both cases, when I left a comment to trigger valgrind (jameslamb#43) the jobs failed with this error:

{
"message": "Bad credentials",
"documentation_url": "https://docs.github.com/rest"
}

Thank you!

[jameslamb]

Ok I've reverted the temporary change from 2714777, and this is now ready for review again.

[jameslamb]

/gha run r-valgrind

Workflow R valgrind tests has been triggered! 🚀
https://github.com/microsoft/LightGBM/actions/runs/922466698

Status: success ✔️.

[StrikerRUS]

@jameslamb

Could you please tell me what secrets I need to set up on my fork

You need to set one secret named WORKFLOW with your PAT with at least repo:status and workflow scopes (selecting workflow auto-chooses the whole repo section unfortunately).

After that don't forget to replace all microsoft/LightGBM occurrences with jameslamb/LightGBM in repo's code, e.g. here

LightGBM/.github/workflows/r_solaris.yml

Line 29 in 24ac920

repository: microsoft/LightGBM

LightGBM/.ci/trigger_dispatch_run.sh

Line 51 in 24ac920

"${GITHUB_API_URL}/repos/microsoft/LightGBM/dispatches"

[jameslamb]

aha!!

After that don't forget to replace all microsoft/LightGBM occurrences with jameslamb/LightGBM in repo's code, e.g. here

This was the part I was missing. I'll try that, thank you very much.

[StrikerRUS]

Generally LGTM, but I leaved some comments that highlight the difference with the original test. I think that it's OK to have not two identical tests for R and Python.

Also, I have one comment about how to speed up the test.

[jameslamb]

Tests started failing after a15596a, which exposed a few more places where these tests differed from the Python equivalents.

More changes in 4950ed3 to make these tests consistent with the Python ones.

1. Replace seq_len(n) / n with code that produces an output identical to numpy.linspace()
2. train for 100 rounds, not 10

This is ready for another review.

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.