cafaeval-protea

A speedup fork of cafaeval (CAFA-evaluator-PK) with a bit-exact parity harness against pristine upstream.

At a glance

End-to-end speedup versus upstream CAFA-evaluator-PK 16a6a6d on a real GOA-derived benchmark corpus, measured at th_step=0.01 (the CAFA default), n_cpu=1:

Variant	Upstream	Fork (B1–B7)	Speedup
NK/LK	92.96 s	4.08 s	22.8×
PK	418.53 s	10.33 s	40.5×

Every optimisation in this fork is gated by a parity harness that runs the fork against a frozen snapshot of unmodified upstream on three synthetic corpora, under rtol=1e-6, atol=1e-9. See Parity harness.

Guide

• Installation
  • Runtime dependencies
  • Development
  • Environment variables
• Quickstart
  • CLI
  • Python API
  • What is different from upstream

Technical

• Performance
  • Headline
  • Phase-by-phase breakdown
  • Phase B7 drilldown
  • Micro-benchmarks
  • What is left on the table
• Parity harness
  • Why it exists
  • Tolerance phases
  • Branches covered
  • Semantic divergence on PK
  • Self-parity (sparse vs dense)
  • Running the harness
  • Re-freezing the oracle
  • Synthetic corpora
• Architecture
  • Sparse confusion matrix kernel
  • Sparse push-up propagation
  • Vectorised prediction parser
  • toi_is_full short-circuit
  • Triples passthrough API
  • Logging

API reference

• cafaeval.evaluation
  • solidify_prediction()
  • compute_f()
  • compute_s()
  • compute_confusion_matrix()
  • compute_confusion_matrix_exclude()
  • compute_confusion_matrix_sparse()
  • compute_confusion_matrix_exclude_sparse()
  • compute_metrics()
  • normalize()
  • evaluate_prediction()
  • cafa_eval()
  • write_results()
• cafaeval.graph
  • Graph
  • Prediction
  • GroundTruth
  • propagate()
• cafaeval.parser
  • obo_parser()
  • update_toi()
  • gt_parser()
  • gt_exclude_parser()
  • pred_parser()
  • ia_parser()

Attribution

This fork descends from

1. BioComputingUP/CAFA-evaluator (original upstream, Piovesan et al.)

2. claradepaolis/CAFA-evaluator-PK at commit 16a6a6d (partial knowledge variant, de Paolis Kaluza et al.)

3. T0chka/CAFA-evaluator-PK-speedup (cherry-picked Phase A speedups, Dolgorukova)

See NOTICE for the full chain and CHANGES.md for the per-phase modification log, as required by GPLv3 §5.a.

Indices

• Index

• Module Index