Repository navigation
feat(bench): cluster-split retrospective benchmark with cluster-aware bootstrap CI - #111
Merged
Merged
Conversation
… bootstrap CI Add cluster-split benchmark that honestly accounts for near-duplicate contamination in the AMP positive set. 33 of 95 benchmark AMPs fall into 14 near-duplicate clusters (sim >= 0.70): magainin-1/2/3, protegrin-1/2/3, tachyplesin-I/II/polyphemusin-I, indolicidin/analog/lys-analog, etc. The standard bootstrap CI (0.717-0.8423) treats near-duplicate sequences as independent samples, underestimating variance. The cluster-aware bootstrap resamples clusters (not sequences), producing an honest CI (0.7061-0.8526). The representative-only AUROC (one AMP per cluster = 76 AMPs vs 96 decoys) is 0.7607 (CI: 0.6854-0.8301), confirming the signal is not entirely driven by near-duplicate redundancy but with less margin than the standard benchmark suggested. New CLI command: bench cluster-split New tests: 31 tests in test_cluster_split_benchmark.py (1421 total, 7 skipped) Updated docs: METRICS_CURRENT.md, BENCHMARK_CARD.md, ROADMAP.md, ARCHITECTURE.md, BENCHMARKING.md Key finding: pipeline signal survives near-duplicate de-inflation. Cluster-aware CI lower bound (0.7061) stays above 0.65 gate. Representative CI lower bound (0.6854) dips below 0.70 — honest limitation.
The CI workflow runs 'make demo' which uses 'python -m openamp_foundry.cli' but the cli package had no __main__.py, causing 'cannot be directly executed' errors. This is a pre-existing issue exposed by CI on this PR.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
The standard benchmark AUROC (0.7832, CI: 0.72-0.84) treats all 95 AMPs as independent samples. 33 of 95 AMPs are in 14 near-duplicate clusters (sim >= 0.70), including magainin-1/2/3, protegrin-1/2/3, tachyplesin-I/II/polyphemusin-I, and indolicidin/analog/lys-analog. The standard bootstrap CI underestimates variance by treating near-identical sequences as independent draws.
This PR adds a cluster-split benchmark that honestly accounts for this contamination.
Key Finding
Pipeline signal survives near-duplicate de-inflation. The cluster-aware CI lower bound (0.7061) stays above the 0.65 gate. The representative-only CI lower bound (0.6854) crossing below 0.70 is an honest limitation.
Changes