fix(lxy): fix completed value inf bug when zero exists in action_mask…

… in gumbel muzero (#178)
opendilab · Dec 29, 2023 · 7953c54 · 7953c54
1 parent 6af174b
commit 7953c54
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/lzero/policy/gumbel_muzero.py b/lzero/policy/gumbel_muzero.py
@@ -575,6 +575,7 @@ def _forward_collect(
                 # entire action set.
                 valid_value = np.where(action_mask[i] == 1.0, improved_policy_probs, 0.0)
                 action = np.argmax([v for v in valid_value])
+                roots_completed_value = np.where(action_mask[i] == 1.0, roots_completed_value, 0.0)
 
                 output[env_id] = {
                     'action': action,