Skip to content

fix(coef_train): align native-surrogate default lr to the M10 corrected barrier #106

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
transfix merged 1 commit into from
Sep 28, 2026
Merged

fix(coef_train): align native-surrogate default lr to the M10 corrected barrier #106

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion grl_snam/tools/coef_train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -365,7 +365,11 @@ def train_native(out, *, grid=96, steps=400, rollout="surrogate", use_cuda=False
story = shrunk(STORIES["city"], n=grid, max_steps=100)
meta = story.meta()
if lr is None:
lr = 1e-5 if str(rollout).lower() == "bicycle" else 2e-4
# Mirror the C++ coef_trainer defaults (libcvc cvc/nav/coef_train.h). M10:
# the corrected (steeper) IPC barrier made the old surrogate lr 2e-4 overshoot
# and DEGRADE the policy (city seed-0 trained reach 71%->26%); retuned to 5e-5
# (1e-4..1e-5 all recover; 5e-5 sits mid-band). The bicycle keeps its own 1e-5.
lr = 1e-5 if str(rollout).lower() == "bicycle" else 5e-5
return nav_native.train_coef_mlp(
story.truth_grid().astype(np.uint8),
out,
Expand Down
Loading