[REVIEW] reenable lightgbm test with a lower (1%) proba accuracy threshold #2799
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…ns. Hence, worth reenabling
|
Please update the changelog in order to start CI tests. View the gpuCI docs here. |
hcho3
approved these changes
Sep 4, 2020
levsnv
changed the title
dantegd
approved these changes
Sep 8, 2020
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
as explained by @hcho3 , lightgbm uses float64 in thresholds. FIL rounds them to float32, which slightly (and unpredictably) alters the predictions.
Since the tests are deterministic, it takes a code [version] change to change the conditions and trigger the discrepancy.
By lowering the proba (not prediction!) accuracy threshold to 1%, we can pass the test until FIL adds float64 threshold support.
This should be a realistic requirement, especially given that the discrepancy shrinks with a more realistic number of input columns, as less information is derived from the exact value of each.