stochastic_muzero: Fix wrong chance values being populated in batch (#…

…275)
opendilab · Sep 19, 2024 · db9e017 · db9e017
1 parent dd89d1c
commit db9e017
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/lzero/mcts/buffer/game_buffer_stochastic_muzero.py b/lzero/mcts/buffer/game_buffer_stochastic_muzero.py
@@ -85,7 +85,7 @@ def _make_batch(self, batch_size: int, reanalyze_ratio: float) -> Tuple[Any]:
             ]
             if self._cfg.use_ture_chance_label_in_chance_encoder:
                 chances_tmp += [
-                    np.random.randint(0, game.action_space_size)
+                    np.random.randint(0, self._cfg.model.chance_space_size)
                     for _ in range(self._cfg.num_unroll_steps - len(chances_tmp))
                 ]
             # obtain the input observations