Fix numerical instability in silhouette/distance computation#5
Merged
Merged
Conversation
cdist used the expansion formula (||x||²+||y||²-2xy) which suffers from catastrophic cancellation in float32 for nearby points. Switch to the direct formula sqrt(sum((x-y)²)) to match scib-metrics (JAX) behavior. Also fix silhouette denominator handling and mean_other mode to exactly match scib-metrics. Verified: all metrics now agree to <0.0002 on both 1k and 20k cell Tabula Sapiens benchmarks. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
cdisteuclidean used the expansion formula (||x||² + ||y||² - 2xy) which suffers from catastrophic cancellation in float32 for nearby points. Switched to the direct formulasqrt(sum((x-y)²))with sub-chunking over y to manage GPU memory, matching scib-metrics (JAX) behavior.(label_freqs - 1)[labels]instead ofcp.maximum(label_freqs[labels] - 1, 1)to match scib-metrics singleton handling.mean_othermode in_silhouette_reduceto usenan/nansuminstead of0.0/sum.Impact
Before this fix, benchmarking against scib-metrics showed large discrepancies:
All metrics now agree to <0.0002 on Tabula Sapiens data (1k and 20k cells).
Test plan
🤖 Generated with Claude Code