Fix KDTree pruning to test against the splitting node#343
Merged
Conversation
The backtracking step decided whether to descend into the other subtree by measuring the distance to the child node point instead of the splitting hyperplane of the current node. Since the child lies somewhere inside the subtree, this bound overestimates the true minimum distance and prunes subtrees that still contain closer points, returning wrong neighbors. On random 12x2 datasets with k=3, more than half of the queries disagreed with brute-force search. Closes elixir-nx#342
josevalim
approved these changes
Jul 20, 2026
Contributor
|
💚 💙 💜 💛 ❤️ |
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Closes #342.
The backtracking step in
predict_ndecided whether to visit the sibling subtree by measuring the distance from the query to the child node's point along the split axis, instead of to the splitting hyperplane of the current node. Since the child lies inside the subtree, this overestimates the minimum possible distance and prunes subtrees that still contain closer points. On random 12x2 datasets with k=3, more than half of the queries disagreed with brute-force search.The fix uses
indices[node], the point defining the cut, in both pruning tests.Checked against
BruteKNNacross dimensions, k, and metrics (Minkowski p=1,2,3 and Chebyshev), and against scikit-learn's kd_tree. Added two regression tests comparing with brute force.