Skip to content

Add materials discovery evaluation metrics#717

Open
janosh wants to merge 7 commits into
ddmms:mainfrom
janosh:feat/materials-discovery-evaluation
Open

Add materials discovery evaluation metrics#717
janosh wants to merge 7 commits into
ddmms:mainfrom
janosh:feat/materials-discovery-evaluation

Conversation

@janosh

@janosh janosh commented Jul 22, 2026

Copy link
Copy Markdown

Summary

  • port Matbench Discovery classification, regression, and subset metrics
  • add typed schemas and local CSV/JSONL artifact helpers
  • require explicit unrounded prevalence for leaderboard-compatible DAF scoring

Test plan

  • pytest -q tests/metrics/test_discovery_metrics.py (50 passed)
  • Ruff and diff checks

Port Matbench Discovery scoring and artifact schemas without coupling evaluation to model execution or bundled prediction data.
janosh added 3 commits July 23, 2026 07:47
Allow pytest framework filtering to select the imported Matbench Discovery suite.
Consolidate energy preparation and fold overlapping regression tests without changing scoring behavior.
Wrap subset metrics with a schema version and Matbench Discovery source metadata so serialized outputs can evolve safely.
@ElliottKasoar ElliottKasoar added new benchmark Proposals and suggestions for new benchmarks external Contribution from external framework labels Jul 23, 2026
janosh added 3 commits July 23, 2026 15:15
Note the schema and source version fields included in serialized evaluation output.
Route all public scoring and subset APIs through artifact masking and rounding, leaving raw helpers private.
Document parameters and return values required by the repository's numpydoc validation hook.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

external Contribution from external framework new benchmark Proposals and suggestions for new benchmarks

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants