data: deposit canonical identity-cluster artifacts (recovered from permea-h100) - #3
Open
Permea-lab-admin wants to merge 1 commit into
Open
Permea-lab-admin wants to merge 1 commit into
Permea-lab-admin wants to merge 1 commit into
Conversation
The eight cluster assignment TSVs used to produce every Condition-B number in the paper were never committed and were not part of the v1.0.1 Zenodo deposit. They existed in exactly one place: a working directory on permea-h100. This deposits them, with checksums and provenance. Why this matters: regeneration does not substitute for the artifact. MMseqs2 18-8cc5c on the same input at tau=0.6 yields 2918 clusters, not the 2913 used for the paper, and therefore different Condition-B numbers. The MMseqs2 version used at generation time was not recorded. A checksum of a file nobody kept proves nothing, so the files themselves are now archived. Also records the exact environment for bit-exact reproduction (x86_64, python 3.10.12, sklearn 1.7.2, numpy 2.2.6, scipy 1.15.3). The published numbers are bit-exact only on that platform; on arm64 ROC-AUC reproduces exactly but PR-AUC and MCC deviate by ~0.001 through float ordering and the F1-max threshold grid. Adds: results/clusters/*.tsv 8 artifacts, 2959 rows each results/clusters/MANIFEST.sha256 results/clusters/README.md provenance, identity definition, environment Co-Authored-By: Claude Opus 4.8 (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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Deposits the eight identity-cluster assignment TSVs used to produce every Condition-B number in the paper, with checksums and provenance.
They were never committed and were not in the v1.0.1 Zenodo deposit. They existed in exactly one place — a working directory on
permea-h100. This PR gets them into the archive.Why this is urgent
Regeneration does not substitute for the artifact. MMseqs2
18-8cc5con the same input at τ=0.6 produces 2918 clusters, not the 2913 used for the paper — a different partition, and therefore different Condition-B numbers. The MMseqs2 version used at generation time was never recorded.A checksum of a file nobody kept proves nothing. The files themselves are now archived.
Contents
clusters_thr0.6.tsvclusters_thr0.{4,5,7}.tsvclusters_align_thr0.{4,5,6,7}.tsvPlus
MANIFEST.sha256and aREADME.mdrecording the identity definition (matches / shorter sequence, global, gaps free), the MMseqs2 flags, and the provenance gap.Each file covers exactly the 2959
sequence_ids of the canonical dataset — verified.Environment for bit-exact reproduction
Now recorded, because it was not before:
The published numbers are bit-exact only on that platform. On arm64 (Apple Silicon), ROC-AUC reproduces exactly, but PR-AUC and MCC deviate by ~0.001: floating-point ordering differs, and the F1-maximising threshold search then lands on a different grid point. A follow-up PR will correct the README's "reproduce every number on a laptop" claim, which is not true on arm64.
Scope
Additive only. No existing file is modified; no result changes. The harness reads these from
results/exactly as before.🤖 Generated with Claude Code