-
Notifications
You must be signed in to change notification settings - Fork 827
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
fix: Fix multiclass training with initial scores #1526
Conversation
/azp run |
Azure Pipelines successfully started running 1 pipeline(s). |
lightgbm/src/main/scala/com/microsoft/azure/synapse/ml/lightgbm/dataset/DatasetAggregator.scala
Show resolved
Hide resolved
Codecov Report
@@ Coverage Diff @@
## master #1526 +/- ##
==========================================
+ Coverage 84.36% 84.39% +0.02%
==========================================
Files 297 297
Lines 14912 14942 +30
Branches 718 728 +10
==========================================
+ Hits 12581 12610 +29
- Misses 2331 2332 +1
Continue to review full report at Codecov.
|
lightgbm/src/main/scala/com/microsoft/azure/synapse/ml/lightgbm/dataset/DatasetAggregator.scala
Outdated
Show resolved
Hide resolved
...src/test/scala/com/microsoft/azure/synapse/ml/lightgbm/split1/VerifyLightGBMClassifier.scala
Show resolved
Hide resolved
/azp run |
Azure Pipelines successfully started running 1 pipeline(s). |
lightgbm/src/main/scala/com/microsoft/azure/synapse/ml/lightgbm/dataset/DatasetAggregator.scala
Outdated
Show resolved
Hide resolved
…apseML into multiclass-init-score
/azp run |
Azure Pipelines successfully started running 1 pipeline(s). |
/azp run |
Azure Pipelines successfully started running 1 pipeline(s). |
/azp run |
Azure Pipelines successfully started running 1 pipeline(s). |
}) | ||
} | ||
|
||
test("Verify chunked array transpose complex") { |
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.
nice tests!
Summary
The use of initial scores with multiclass training was not working. There was 1 test, but it was marked as "ignore" due to being flaky. In actuality, it only passed sometimes out of pure luck since there were several bugs.
Tests
I turned the test back on, although it is still flaky for some reason. Could use some advice on this. However, having traced through all the code underneath, we are now appropriately sending multiclass initial scores to LightGBM. One column is sent for every class, and it is transposed from row to column format.
Dependency changes
If you needed to make any changes to dependencies of this project, please describe them here.
AB#1827802