From 08d70296cf03ca920444c09bea16de55d3539353 Mon Sep 17 00:00:00 2001
From: Erik Nygren <erik.nygren@sbb.ch>
Date: Sun, 1 Sep 2019 11:19:41 -0400
Subject: [PATCH] only store observations in buffer when you are allowed to
take an action!
---
torch_training/multi_agent_training.py | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/torch_training/multi_agent_training.py b/torch_training/multi_agent_training.py
index 4822704..ba30d46 100644
--- a/torch_training/multi_agent_training.py
+++ b/torch_training/multi_agent_training.py
@@ -175,6 +175,10 @@ def main(argv):
# Action
for a in range(env.get_num_agents()):
+ if env.agents[a].speed_data['position_fraction'] == 0.:
+ register_action_state[a] = True
+ else:
+ register_action_state[a] = False
action = agent.act(agent_obs[a], eps=eps)
action_prob[action] += 1
action_dict.update({a: action})
@@ -192,7 +196,7 @@ def main(argv):
final_obs[a] = agent_obs[a].copy()
final_obs_next[a] = agent_next_obs[a].copy()
final_action_dict.update({a: action_dict[a]})
- if not done[a]:
+ if not done[a] and register_action_state[a]:
agent.step(agent_obs[a], action_dict[a], all_rewards[a], agent_next_obs[a], done[a])
score += all_rewards[a] / env.get_num_agents()
--
GitLab