diff --git a/RLLib_training/RailEnvRLLibWrapper.py b/RLLib_training/RailEnvRLLibWrapper.py
index ad504e271f8758f6f445917c1963b9da316b9ba3..5ab92a48cf815c1cca17eac993c8ff528aabd32c 100644
--- a/RLLib_training/RailEnvRLLibWrapper.py
+++ b/RLLib_training/RailEnvRLLibWrapper.py
@@ -19,13 +19,14 @@ class RailEnvRLLibWrapper(MultiAgentEnv):
vector_index = config.vector_index
else:
vector_index = 1
- self.rail_generator = config["rail_generator"](nr_start_goal=config['number_of_agents'], min_dist=5,
- nr_extra=30, seed=config['seed'] * (1+vector_index))
+ #self.rail_generator = config["rail_generator"](nr_start_goal=config['number_of_agents'], min_dist=5,
+ # nr_extra=30, seed=config['seed'] * (1+vector_index))
set_seed(config['seed'] * (1+vector_index))
- self.env = RailEnv(width=config["width"], height=config["height"], rail_generator=self.rail_generator,
+ #self.env = RailEnv(width=config["width"], height=config["height"],
+ self.env = RailEnv(width=10, height=20,
number_of_agents=config["number_of_agents"], obs_builder_object=config['obs_builder'])
- self.env.load('./baselines/torch_training/railway/complex_scene.pkl')
+ self.env.load('/mount/SDC/flatland/baselines/torch_training/railway/complex_scene.pkl')
self.width = self.env.width
self.height = self.env.height
@@ -45,7 +46,6 @@ class RailEnvRLLibWrapper(MultiAgentEnv):
self.agents = self.env.agents
self.agents_static = self.env.agents_static
self.dev_obs_dict = self.env.dev_obs_dict
-
return obs
def step(self, action_dict):
diff --git a/RLLib_training/custom_preprocessors.py b/RLLib_training/custom_preprocessors.py
index cc58a0d51188c756fb7c37e0f10ada403ecf804e..1c3fa0898582a6f9d093dbcac787d70805b2e0b6 100644
--- a/RLLib_training/custom_preprocessors.py
+++ b/RLLib_training/custom_preprocessors.py
@@ -50,10 +50,10 @@ def norm_obs_clip(obs, clip_min=-1, clip_max=1):
class CustomPreprocessor(Preprocessor):
def _init_shape(self, obs_space, options):
- return (105,)
+ return (111,)
def transform(self, observation):
- if len(observation) == 105:
+ if len(observation) == 111:
return norm_obs_clip(observation)
else:
return observation
diff --git a/RLLib_training/experiment_configs/CustomModels.py b/RLLib_training/experiment_configs/CustomModels.py
deleted file mode 100644
index e69de29bb2d1d6434b8b29ae775ad8c2e48c5391..0000000000000000000000000000000000000000
diff --git a/RLLib_training/experiment_configs/n_agents_experiment/config.gin b/RLLib_training/experiment_configs/n_agents_experiment/config.gin
index 31eedbe868622e21bda76b80c85f6596db028820..025eab9130086188892e406edbfb477215f8a8cd 100644
--- a/RLLib_training/experiment_configs/n_agents_experiment/config.gin
+++ b/RLLib_training/experiment_configs/n_agents_experiment/config.gin
@@ -1,14 +1,19 @@
-run_experiment.name = "n_agents_results"
+run_experiment.name = "observation_benchmark_results"
run_experiment.num_iterations = 1002
-run_experiment.save_every = 200
-run_experiment.hidden_sizes = [32, 32]
+run_experiment.save_every = 100
+run_experiment.hidden_sizes = [32,32]
run_experiment.map_width = 20
run_experiment.map_height = 20
-run_experiment.n_agents = {"grid_search": [1]}#, 2, 5, 10]}
-run_experiment.policy_folder_name = "ppo_policy_{config[n_agents]}_agents"
+run_experiment.n_agents = {"grid_search": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]}
+run_experiment.policy_folder_name = "ppo_policy_{config[obs_builder].__class__.__name__}_entropy_coeff_{config[entropy_coeff]}_{config[n_agents]}_agents_"
run_experiment.horizon = 50
run_experiment.seed = 123
+run_experiment.entropy_coeff = {"grid_search": [1e-3, 1e-2, 0]}
+
+run_experiment.obs_builder = {"grid_search": [@TreeObsForRailEnv()]}# [@TreeObsForRailEnv(), @GlobalObsForRailEnv() ]}
+TreeObsForRailEnv.max_depth = 2
+LocalObsForRailEnv.view_radius = 5
diff --git a/RLLib_training/experiment_configs/observation_benchmark_loaded_env/config.gin b/RLLib_training/experiment_configs/observation_benchmark_loaded_env/config.gin
index 03aae996a12bf69717599cf734a39b6780bbfc72..64ff1c981dc9d068dee3a089bc8cb77c834d9e63 100644
--- a/RLLib_training/experiment_configs/observation_benchmark_loaded_env/config.gin
+++ b/RLLib_training/experiment_configs/observation_benchmark_loaded_env/config.gin
@@ -1,7 +1,7 @@
run_experiment.name = "observation_benchmark_loaded_env_results"
run_experiment.num_iterations = 1002
run_experiment.save_every = 50
-run_experiment.hidden_sizes = 32
+run_experiment.hidden_sizes = [32, 32]
run_experiment.map_width = 20
run_experiment.map_height = 20
@@ -10,9 +10,10 @@ run_experiment.policy_folder_name = "ppo_policy_{config[obs_builder].__class__._
run_experiment.horizon = 50
run_experiment.seed = 123
+run_experiment.conv_model = False
run_experiment.entropy_coeff = 1e-2
-run_experiment.obs_builder = {"grid_search": [@LocalObsForRailEnv(), @TreeObsForRailEnv(), @GlobalObsForRailEnv(), @GlobalObsForRailEnvDirectionDependent]}
+run_experiment.obs_builder = @TreeObsForRailEnv()#{"grid_search": [@LocalObsForRailEnv(), @TreeObsForRailEnv(), @GlobalObsForRailEnv(), @GlobalObsForRailEnvDirectionDependent()]}
TreeObsForRailEnv.max_depth = 2
LocalObsForRailEnv.view_radius = 5
diff --git a/RLLib_training/train_experiment.py b/RLLib_training/train_experiment.py
index e7085b48819028f1ee3e9eee2f9cf8945ff0c870..d58e9bf0e02ee270e97672a42dd26384c68d7b4e 100644
--- a/RLLib_training/train_experiment.py
+++ b/RLLib_training/train_experiment.py
@@ -52,6 +52,10 @@ def train(config, reporter):
set_seed(config['seed'], config['seed'], config['seed'])
+ config['map_width']= 20
+ config['map_height']= 10
+ config['n_agents'] = 8
+
# Example configuration to generate a random rail
env_config = {"width": config['map_width'],
"height": config['map_height'],
@@ -62,7 +66,7 @@ def train(config, reporter):
# Observation space and action space definitions
if isinstance(config["obs_builder"], TreeObsForRailEnv):
- obs_space = gym.spaces.Box(low=-float('inf'), high=float('inf'), shape=(105,))
+ obs_space = gym.spaces.Box(low=-float('inf'), high=float('inf'), shape=(111,))
preprocessor = "tree_obs_prep"
elif isinstance(config["obs_builder"], GlobalObsForRailEnv):
@@ -191,6 +195,6 @@ def run_experiment(name, num_iterations, n_agents, hidden_sizes, save_every,
if __name__ == '__main__':
gin.external_configurable(tune.grid_search)
- dir = '/home/guillaume/EPFL/Master_Thesis/flatland/baselines/RLLib_training/experiment_configs/conv_model_test' # To Modify
+ dir = '/mount/SDC/flatland/baselines/RLLib_training/experiment_configs/observation_benchmark_loaded_env' # To Modify
gin.parse_config_file(dir + '/config.gin')
run_experiment(local_dir=dir)