-
Notifications
You must be signed in to change notification settings - Fork 1.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[CI] Explicitly set eval batch size in determinism tests, introduce a new integration test group, and exclude slow tests. #3590
Conversation
Unit Test Results 6 files ±0 6 suites ±0 44m 5s ⏱️ - 36m 49s Results for commit c6309d3. ± Comparison against base commit 6931fe4. This pull request removes 3 tests.
♻️ This comment has been updated with latest results. |
@@ -5,6 +5,7 @@ | |||
from tests.integration_tests.utils import generate_data_as_dataframe | |||
|
|||
|
|||
@pytest.mark.slow |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I wonder what makes this test particularly slow, perhaps we can modify to use 1 feature and just train for 5 steps instead of an entire epoch
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Generally LGTM except for one comment!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! +1 to @arnavgarg1's comment about the hyperopt tests
Make Ludwig CI consistently green again.
With the changes in this PR, CI time is cut down from 1.5 hours (with timeouts) to 40 minutes.
tests/ludwig/models/test_training_determinism.py::test_training_determinism_local_backend
:CI speedup: New integration test group and marking tests as slow
Added a new integration test group E to further parallelize integration tests.
Marking tests as slow: The purpose of on-PR-CI is to give us a timely sense of whether a change is safe to land. The slowest tests (largely hyperopt+ray), in my opinion, provide more limited utility and not worth the on-PR-CI slowdown.
NOTE: Slow tests are still run when a PR is merged to master (PR authors are notified).
Tests marked as slow: