From aa2a0cb6da5c6689b9d4bd80187f2ae1bf5ec673 Mon Sep 17 00:00:00 2001 From: AnanthHari Date: Thu, 23 Jan 2020 03:42:52 -0500 Subject: [PATCH] Fixes empty `state` argument in compute_single_action method (#6894) * Fixes empty `state` parameter in compute_single_action method * Fixed style --- rllib/policy/policy.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/rllib/policy/policy.py b/rllib/policy/policy.py index 081d8f384..4d52d0398 100644 --- a/rllib/policy/policy.py +++ b/rllib/policy/policy.py @@ -128,6 +128,7 @@ class Policy(metaclass=ABCMeta): prev_reward_batch = None info_batch = None episodes = None + state_batch = None if prev_action is not None: prev_action_batch = [prev_action] if prev_reward is not None: @@ -136,9 +137,11 @@ class Policy(metaclass=ABCMeta): info_batch = [info] if episode is not None: episodes = [episode] + if state is not None: + state_batch = [[s] for s in state] [action], state_out, info = self.compute_actions( - [obs], [[s] for s in state], + [obs], state_batch, prev_action_batch=prev_action_batch, prev_reward_batch=prev_reward_batch, info_batch=info_batch,