Testing light gbm bad allocation #6968
Conversation
ghost
assigned 
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #6968 +/- ##
=======================================
Coverage 68.80% 68.81%
=======================================
Files 1249 1249
Lines 249686 249686
Branches 25485 25485
=======================================
+ Hits 171806 171814 +8
+ Misses 71286 71282 -4
+ Partials 6594 6590 -4
Flags with carried forward coverage won't be shown. Click here to find out more.
|
|
/azp run |
|
Azure Pipelines successfully started running 2 pipeline(s). |
|
/azp run |
|
Azure Pipelines successfully started running 2 pipeline(s). |
|
/azp run |
|
Azure Pipelines successfully started running 2 pipeline(s). |
|
/azp run |
|
Azure Pipelines successfully started running 2 pipeline(s). |
| [CollectionDefinition(nameof(NoParallelizationDefinition), DisableParallelization = true)] | ||
| public class NoParallelizationDefinition { } | ||
|
|
||
| [Collection(nameof(NoParallelizationDefinition))] |
There was a problem hiding this comment.
Is there something specific about these LightGBM tests that differs from others?
https://github.com/search?q=repo%3Adotnet%2Fmachinelearning%20LightGBMFact&type=code
There was a problem hiding this comment.
Honestly I'm not sure. I just put this tag around the test I saw failing and it seems to have worked. So its possible there is? Or its possible it just has to do with test ordering? I'm not sure.
There was a problem hiding this comment.
It would be good to at comment that we've observed concurrency issues in LightGBM when running these in parallel. Consider keeping an issue open to track that and link to the issue since we still don't understand why it's failing.
|
Just saw this in 3.0 branch Code Coverage build. Maybe we should backport this test reliability fix? |
|
/backport to release/3.0 |
|
Started backporting to release/3.0: https://github.com/dotnet/machinelearning/actions/runs/8272862595 |
Changes some of the LightGBM tests to run sequentially to try and fix the bad allocation issue.
This is in regards to issue #6961. I'll leave that issue open so we can continue to monitor after this goes in.