From 1c1e6dc8706447fc9ab3bad1b4e2fd7ab01f87d8 Mon Sep 17 00:00:00 2001
From: Guillaume Mollard <guillaume@iccluster091.iccluster.epfl.ch>
Date: Tue, 14 May 2019 16:35:29 +0200
Subject: [PATCH] bug with multiple environments
---
RailEnvRLLibWrapper.py | 15 +++++++--------
train.py | 15 +++++++++++----
2 files changed, 18 insertions(+), 12 deletions(-)
diff --git a/RailEnvRLLibWrapper.py b/RailEnvRLLibWrapper.py
index 3649434..1b537a6 100644
--- a/RailEnvRLLibWrapper.py
+++ b/RailEnvRLLibWrapper.py
@@ -13,8 +13,8 @@ class RailEnvRLLibWrapper(RailEnv, MultiAgentEnv):
number_of_agents=1,
obs_builder_object=TreeObsForRailEnv(max_depth=2)):
- super(RailEnvRLLibWrapper, self).__init__(width, height, rail_generator,
- number_of_agents, obs_builder_object)
+ super(RailEnvRLLibWrapper, self).__init__(width=width, height=height, rail_generator=rail_generator,
+ number_of_agents=number_of_agents, obs_builder_object=obs_builder_object)
def reset(self, regen_rail=True, replace_agents=True):
self.agents_done = []
@@ -32,17 +32,16 @@ class RailEnvRLLibWrapper(RailEnv, MultiAgentEnv):
if agent != '__all__':
o[agent] = obs[agent]
r[agent] = rewards[agent]
-
+
d[agent] = dones[agent]
- # obs.pop(agent_done)
- # rewards.pop(agent_done)
- # dones.pop(agent_done)
for agent, done in dones.items():
if done and agent != '__all__':
self.agents_done.append(agent)
-
+
+ #print(obs)
+ #return obs, rewards, dones, infos
return o, r, d, infos
-
+
def get_agent_handles(self):
return super(RailEnvRLLibWrapper, self).get_agent_handles()
diff --git a/train.py b/train.py
index 71f214c..686ae5d 100644
--- a/train.py
+++ b/train.py
@@ -64,9 +64,9 @@ def train(config):
rail_generator=random_rail_generator(cell_type_relative_proportion=transition_probability),
number_of_agents=1)
"""
- env = RailEnv(width=50,
- height=50,
- rail_generator=complex_rail_generator(nr_start_goal=50, min_dist=5, max_dist=99999, seed=0),
+ env = RailEnvRLLibWrapper(width=20,
+ height=20,
+ rail_generator=complex_rail_generator(nr_start_goal=5, min_dist=5, max_dist=99999, seed=0),
number_of_agents=5)
"""
env = RailEnv(width=20,
@@ -94,11 +94,18 @@ def train(config):
return f"ppo_policy"
agent_config = ppo.DEFAULT_CONFIG.copy()
- agent_config['model'] = {"fcnet_hiddens": [32, 32], "custom_preprocessor": "my_prep"}
+ agent_config['model'] = {"fcnet_hiddens": [32, 32]}#, "custom_preprocessor": "my_prep"}
agent_config['multiagent'] = {"policy_graphs": policy_graphs,
"policy_mapping_fn": policy_mapping_fn,
"policies_to_train": list(policy_graphs.keys())}
agent_config["horizon"] = 50
+ #agent_config["num_workers"] = 0
+ #agent_config["num_cpus_per_worker"] = 40
+ #agent_config["num_gpus"] = 2.0
+ # agent_config["num_gpus_per_worker"] = 2.0
+ agent_config["num_cpus_for_driver"] = 5
+ agent_config["num_envs_per_worker"] = 15
+ #agent_config["batch_mode"] = "complete_episodes"
ppo_trainer = PPOAgent(env=f"railenv", config=agent_config)
--
GitLab