This repository was archived by the owner on Oct 4, 2026. It is now read-only.
Repository navigation
Preserve warm-start baselines across boosting rounds - #11
Merged
Merged
Conversation
This file contains hidden or 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
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem and change
Training uses
warm_startin round one but drops it from subsequent rounds. With a nonzero baseline, later trees therefore optimize derivatives at the wrong score.Include the baseline alongside accumulated tree predictions on every round, in both full-data and row-sampled training. For sampled rows, select the corresponding baseline entries. The fitted model continues to return only the learned correction; the docstring now explains how to add the external baseline for full predictions.
Add regression tests with row-varying baselines: an independent weighted logistic Newton oracle and least-squares equivalence to fitting residual targets, covering 1/2/4 rounds, full and 50% sampled data, observation weights, regularization, and preservation of the input baseline.
Addresses the warm-start finding in #8. Merge this fix before the retirement documentation in #10.
Validation
0d83f261e0a32f506ecf1cc144ff2a939c67d1ae, the original baseline-0.7 comparison now has maximum probability differences of 4.21e-8 after one round and 5.40e-8 after two rounds (previously 0.03527 after two).git diff --check.