[ML] More efficient predictions and fix flaky test #2296
Merged
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
valeriy42
approved these changes
Jun 7, 2022
Comment on lines
+1560
to
+1563
| // We fix the tree topology penalty because its initialization is | ||
| // affected by the maxNumNewTrees. Changing the parameter ranges for | ||
| // trainIncremental means we can no longer be sure that the hold out | ||
| // loss is no larger when we _optionally_ allow adding extra trees. |
There was a problem hiding this comment.
Oh, that's a good catch 🚀
Co-authored-by: Valeriy Khakhutskyy <[email protected]>
Co-authored-by: Valeriy Khakhutskyy <[email protected]>
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Outdated suggestions cannot be applied.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.
This changes the way we set up the data frame for prediction since it doesn't need to cache loss derivatives. It also reworks the test for adding trees in incremental training. In particular, it switches measuring the accuracy on the hold out set to use the corrected loss. This is used for selecting the best model. It also prepares the hold data set more carefully to mix in out of domain data.
Closes #2271.