def run():
expr_config = Efficient_DQN_Config()
expr_config.get_agent_config()
# make sure double q update is turn on
expr_config.agent_config.policy_hyper_params.use_double_q_update = True
sys_args = expr_config.agent_config.sys_args
policy_hyper_params = expr_config.agent_config.policy_hyper_params
policy_network_cfg = expr_config.agent_config.policy_network_cfg
policy_optim_cfg = expr_config.agent_config.policy_optim_cfg
files_to_save = [
os.path.abspath(sys.modules[DQN_Agent.__module__].__file__)
]
env, expr_manager = init_experiment(
env_name=sys_args.env,
model_name=model_name,
env_hyper_params=expr_config.agent_config,
random_seed=sys_args.seed,
files_to_save=files_to_save,
config_to_save=expr_config.agent_config,
experiment_log_dir=experiment_log_dir,
use_wandb=sys_args.use_wandb,
virtual_display=sys_args.virtual_display)
# create agent
dqn_agent = DQN_Agent(env,
sys_args,
policy_hyper_params,
policy_network_cfg,
policy_optim_cfg,
logger=expr_manager)
# create agent trainer
agent_trainer = Train_Value_Based_Agent(dqn_agent, env)
if sys_args.test:
agent_trainer.test()
else:
agent_trainer.train()
def run():
expr_config = SDRL_Config()
expr_config.get_agent_config()
sys_args = expr_config.agent_config.sys_args
policy_hyper_params = expr_config.agent_config.policy_hyper_params
policy_network_cfg = expr_config.agent_config.policy_network_cfg
policy_optim_cfg = expr_config.agent_config.policy_optim_cfg
files_to_save = [os.path.abspath(sys.modules[SDRL_Agent.__module__].__file__)]
env, expr_manager = init_experiment(env_name=sys_args.env,
model_name=model_name,
env_hyper_params=expr_config.agent_config,
random_seed=sys_args.seed,
files_to_save=files_to_save, config_to_save=expr_config.agent_config,
experiment_log_dir=experiment_log_dir,
use_wandb=sys_args.use_wandb, virtual_display=sys_args.virtual_display)
sdrl_agent = SDRL_Agent(env, sys_args, policy_hyper_params,
policy_network_cfg, policy_optim_cfg, logger=expr_manager)
if sys_args.test:
sdrl_agent.test()
else:
sdrl_agent.train()