ding.example.mappo

ding.example.mappo

Full Source Code

../ding/example/mappo.py

1import gym 2from ditk import logging 3from ding.model import MAVAC 4from ding.policy import PPOPolicy 5from ding.envs import DingEnvWrapper, SubprocessEnvManagerV2 6from ding.data import DequeBuffer 7from ding.config import compile_config 8from ding.framework import task, ding_init 9from ding.framework.context import OnlineRLContext 10from ding.framework.middleware import multistep_trainer, StepCollector, interaction_evaluator, CkptSaver, \ 11 gae_estimator, online_logger, termination_checker 12from ding.utils import set_pkg_seed 13from dizoo.petting_zoo.config.ptz_simple_spread_mappo_config import main_config, create_config 14from dizoo.petting_zoo.envs.petting_zoo_simple_spread_env import PettingZooEnv 15 16 17def main(): 18 logging.getLogger().setLevel(logging.INFO) 19 cfg = compile_config(main_config, create_cfg=create_config, auto=True) 20 ding_init(cfg) 21 with task.start(async_mode=False, ctx=OnlineRLContext()): 22 collector_env = SubprocessEnvManagerV2( 23 env_fn=[lambda: PettingZooEnv(cfg.env) for _ in range(cfg.env.collector_env_num)], cfg=cfg.env.manager 24 ) 25 evaluator_env = SubprocessEnvManagerV2( 26 env_fn=[lambda: PettingZooEnv(cfg.env) for _ in range(cfg.env.evaluator_env_num)], cfg=cfg.env.manager 27 ) 28 29 set_pkg_seed(cfg.seed, use_cuda=cfg.policy.cuda) 30 31 model = MAVAC(**cfg.policy.model) 32 policy = PPOPolicy(cfg.policy, model=model) 33 34 task.use(interaction_evaluator(cfg, policy.eval_mode, evaluator_env)) 35 task.use(StepCollector(cfg, policy.collect_mode, collector_env)) 36 task.use(gae_estimator(cfg, policy.collect_mode)) 37 task.use(multistep_trainer(policy.learn_mode, log_freq=100)) 38 task.use(CkptSaver(policy, cfg.exp_name, train_freq=1000)) 39 task.use(online_logger(train_show_freq=10)) 40 task.use(termination_checker(max_env_step=int(1e6))) 41 task.run() 42 43 44if __name__ == "__main__": 45 main()