ding.example.sac

ding.example.sac

Full Source Code

../ding/example/sac.py

1from ditk import logging 2from ding.model import ContinuousQAC 3from ding.policy import SACPolicy 4from ding.envs import BaseEnvManagerV2 5from ding.data import DequeBuffer 6from ding.config import compile_config 7from ding.framework import task, ding_init 8from ding.framework.context import OnlineRLContext 9from ding.framework.middleware import data_pusher, StepCollector, interaction_evaluator, \ 10 CkptSaver, OffPolicyLearner, termination_checker, online_logger 11from ding.utils import set_pkg_seed 12from dizoo.classic_control.pendulum.envs.pendulum_env import PendulumEnv 13from dizoo.classic_control.pendulum.config.pendulum_sac_config import main_config, create_config 14 15 16def main(): 17 logging.getLogger().setLevel(logging.INFO) 18 cfg = compile_config(main_config, create_cfg=create_config, auto=True) 19 ding_init(cfg) 20 with task.start(async_mode=False, ctx=OnlineRLContext()): 21 collector_env = BaseEnvManagerV2( 22 env_fn=[lambda: PendulumEnv(cfg.env) for _ in range(cfg.env.collector_env_num)], cfg=cfg.env.manager 23 ) 24 evaluator_env = BaseEnvManagerV2( 25 env_fn=[lambda: PendulumEnv(cfg.env) for _ in range(cfg.env.evaluator_env_num)], cfg=cfg.env.manager 26 ) 27 28 set_pkg_seed(cfg.seed, use_cuda=cfg.policy.cuda) 29 30 model = ContinuousQAC(**cfg.policy.model) 31 buffer_ = DequeBuffer(size=cfg.policy.other.replay_buffer.replay_buffer_size) 32 policy = SACPolicy(cfg.policy, model=model) 33 34 task.use(interaction_evaluator(cfg, policy.eval_mode, evaluator_env)) 35 task.use( 36 StepCollector(cfg, policy.collect_mode, collector_env, random_collect_size=cfg.policy.random_collect_size) 37 ) 38 task.use(data_pusher(cfg, buffer_)) 39 task.use(OffPolicyLearner(cfg, policy.learn_mode, buffer_)) 40 task.use(CkptSaver(policy, cfg.exp_name, train_freq=100)) 41 task.use(termination_checker(max_train_iter=10000)) 42 task.use(online_logger()) 43 task.run() 44 45 46if __name__ == "__main__": 47 main()