from approaches.reinforced.reward_function import UnstableReward
if __name__ == "__main__":
ray.init(
address='auto'
) # address = None when running locally. address = 'auto' when running on aws.]
obs_state_processor = SimpleObsStateProcessor(
pathogen_sorting_strategy=infected_population_sorting_per_city)
act_state_processor = SimpleActStateProcessor(
sort_pathogens=obs_state_processor.sort_pathogens)
# Notice that trial_max will only work for stochastic policies
register_env(
"ic20env", lambda _: SimplifiedIC20Environment(obs_state_processor,
act_state_processor,
UnstableReward(),
trial_max=10))
ten_gig = 10737418240
trainer = A2CTrainer(
env="ic20env",
config=merge_dicts(
DEFAULT_CONFIG,
{
# -- Specific parameters
'num_gpus': 0,
'num_workers': 15,
"num_envs_per_worker": 1,
"num_cpus_per_worker": 1,
"memory_per_worker": ten_gig,
'gamma': 0.99,
from ray.tune.util import merge_dicts
from approaches.reinforced.action_state_processor import SimpleActStateProcessor
from approaches.reinforced.constants import DEFAULT_CONFIG
from approaches.reinforced.environment import SimplifiedIC20Environment, CHECKPOINT_FILE
from approaches.reinforced.observation_state_processor import SimpleObsStateProcessor, infected_population_sorting_per_city
from approaches.reinforced.reward_function import UnstableReward
if __name__ == "__main__":
ray.init(address='auto') # address = None when running locally. address = 'auto' when running on aws.]
obs_state_processor = SimpleObsStateProcessor(pathogen_sorting_strategy=infected_population_sorting_per_city)
act_state_processor = SimpleActStateProcessor(sort_pathogens=obs_state_processor.sort_pathogens)
# Notice that trial_max will only work for stochastic policies
register_env("ic20env",
lambda _: SimplifiedIC20Environment(obs_state_processor, act_state_processor, UnstableReward(),
trial_max=10))
ten_gig = 10737418240
trainer = PPOTrainer(
env="ic20env",
config=merge_dicts(DEFAULT_CONFIG, {
# -- Rollout-Worker
'num_gpus': 1,
'num_workers': 15,
"num_envs_per_worker": 1,
"num_cpus_per_worker": 0.5,
"memory_per_worker": ten_gig,
# -- Specific parameters
"use_gae": True,