t5lab.render_mvprop¶
Render a trained MVProp-controller policy to a GIF — for visual audit (Bijan's requirement).
Same un-jitted rollout as ctde_v0.render, but loads the frozen distilled planner and threads it
into ppo._goal_to_move(..., mvplanner=) so the moves are the LEARNED planner's, matching the
trained pipeline exactly. Colours agents by role when the role picker is on.
PYTHONPATH=.:../../../FiedlerValueEstimation $PY -m t5lab.render_mvprop --run-dir runs/mvprop/open_32_seed0 --out gifs/mvprop_open32.gif --seed 3