pyfracval.experimental.candidate_policies

Archived CCA candidate-pair ordering policies.

leaf_soft (leaf-pairs-first ordering) and leaf_score/leaf_hybrid (heuristic-scored ordering) were benchmarked against the unranked (shuffled) baseline ordering at N=512 in the hard regime (see docs/source/experiments.md): both land on the same success rate as baseline. How you search for a contact pair doesn’t matter much when the real question is whether a valid contact pair exists at all at the enforced gamma_pc.

Kept reachable via cca_candidate_policy for anyone who wants to try a different scoring heuristic later. Leaf classification and scoring themselves (_candidate_leaf_class/_candidate_score in cca/candidates.py) stay in production code since the per-attempt telemetry they feed is diagnostic instrumentation used regardless of policy, not part of what was benchmarked here.

Module Contents

pyfracval.experimental.candidate_policies.reorder_candidates_by_policy(candidate_policy, candidate_indices, leaf_mask_1, leaf_mask_2, coords1, radii1, cm1, coords2, radii2, cm2, gamma_pc, algorithm_config, candidate_leaf_class_fn, candidate_score_fn)

Reorder shuffled candidate pairs according to an archived policy.