ctde_v0.run_frontier¶
run_frontier
¶
Frontier sweep — trace the coverage<->connectivity Pareto by sweeping the COVERAGE reward weight (the cov:conn trade-off dial), SOFT mechanism only, NO action_mask.
~6 points x N seeds @32^2/10, OPEN terrain, cover_r=0. Each point = one --w-coverage setting against a fixed soft connectivity penalty; raising w-coverage tilts the balance toward coverage (and should shed connectivity), tracing the frontier. Report RAW coverage + REAL connectivity per point -> the empirical Pareto curve (NOT '% of optimal' — that yardstick is retired). Anchors already on the curve: slamattn ~43/73 (conn-dominant), old frontier_attn ~98/32 (cov-dominant).
PYTHONPATH=.:../../../FiedlerValueEstimation ~/ZymeraLab/.venv/bin/python ctde_v0/run_frontier.py --out runs/frontier --seeds 4 --iters 2000 --jobs 1