Baltor Get started

Library / Code module

Method-blind relabeling and behavioral equivalence

behavioral_equivalence_relabeling

Replace candidate names with seeded neutral labels kept in a separate key, and group candidates whose outputs on the same probe inputs are identical, with optional rounding. Use it to blind judges to method names and to stop counting behaviorally identical variants as separate results.

Files, names only

  • ATTRIBUTION.md 4.0 KB
  • LICENSE 1.0 KB
  • README.md 2.6 KB
  • atom_cli.py 3.0 KB
  • behavioral_equivalence_relabeling.py 5.2 KB
  • component.json 8.3 KB
  • numerics.py 28.0 KB
  • schema_lite.py 8.9 KB
  • test_behavioral_equivalence_relabeling.py 3.5 KB
  • test_package.py 12.0 KB

Full files in your harness.

What it takes and gives

Runs with
Python 3.10 or later
Entry points
blind_labels, equivalence_classes, blind_report
Known answers
5 cases its tests reproduce

Cases it covers

  • float noise does not split a class: Exact float comparison treats 0.1 + 0.2 and 0.3 as different behaviour.

Limits and scope

Limits
Equality on a finite probe set is evidence of equal behaviour on those probes only; candidates may differ elsewhere. Rounding boundaries can separate values that differ by less than the rounding step. Blind labels hide names, not tell-tale output styles a reader may still recognize.
Guarantees
Labels come from a seeded Fisher-Yates shuffle of the sorted names, so their order carries no information about the names; the same seed gives the same labels on every Python version. Classes are an equivalence relation: rounding to fixed decimals, never a tolerance, decides equality. Each class carries the SHA-256 of the canonical JSON of its (rounded) outputs.
Assumes
Every candidate was run on the same probe inputs in the same order. Equal outputs on the probes are taken as equal behaviour on those probes only.

Details

Kind
Code module
Licence
MIT
Declared effects
runs commands
Source
github.com/alisonjieli-png/loop-engine/tools/creative_originals/evaluation_atoms/items/behavioral_equivalence_relabeling
Size
10 files, 76.5 KB
Identity
library.supply.creative_originals.798fe8aba4715cb91897c3bd.cdb6fd11aab18c96
Version
cdb6fd11aab18c964e8638a3c70c5904139c68b442ef811724e08f175c6eff6a

The version is the SHA-256 of the item your harness downloads; every file in it names its own checksum.

Where it is shown

Related components

Something wrong with this item, its rights or its safety? Report it on the support page, or from your harness after a download.