fixed unpicklable bug in k step shaper (#87)

Co-authored-by: ysqyang <v-yangqi@microsoft.com>
microsoft · Sep 24, 2020 · a553cd3 · a553cd3
1 parent bad607e
commit a553cd3
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/maro/rl/shaping/k_step_experience_shaper.py b/maro/rl/shaping/k_step_experience_shaper.py
@@ -58,4 +58,4 @@ def __call__(self, trajectory, snapshot_list):
             agent_exp[KStepExperienceKeys.NEXT_ACTION.value].appendleft(next_transition["action"])
             agent_exp[KStepExperienceKeys.DISCOUNT.value].appendleft(self._reward_decay ** (min(self._steps, len(trajectory)-1-i)))
 
-        return experiences
+        return dict(experiences)