ding.example.dqn_her

ding.example.dqn_her

Full Source Code

../ding/example/dqn_her.py

1import gym 2from ditk import logging 3from ding.model import DQN 4from ding.policy import DQNPolicy 5from ding.reward_model import HerRewardModel 6from ding.envs import BaseEnvManagerV2 7from ding.data import DequeBuffer 8from ding.config import compile_config 9from ding.framework import task 10from ding.framework.context import OnlineRLContext 11from ding.framework.middleware import HERLearner, EpisodeCollector, interaction_evaluator, data_pusher, \ 12 eps_greedy_handler, CkptSaver 13from ding.utils import set_pkg_seed 14from dizoo.bitflip.envs import BitFlipEnv 15from dizoo.bitflip.config.bitflip_her_dqn_config import main_config, create_config 16 17 18def main(): 19 logging.getLogger().setLevel(logging.INFO) 20 cfg = compile_config(main_config, create_cfg=create_config, auto=True) 21 with task.start(async_mode=False, ctx=OnlineRLContext()): 22 collector_env = BaseEnvManagerV2( 23 env_fn=[lambda: BitFlipEnv(cfg.env) for _ in range(cfg.env.collector_env_num)], cfg=cfg.env.manager 24 ) 25 evaluator_env = BaseEnvManagerV2( 26 env_fn=[lambda: BitFlipEnv(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 = DQN(**cfg.policy.model) 32 buffer_ = DequeBuffer(size=cfg.policy.other.replay_buffer.replay_buffer_size) 33 policy = DQNPolicy(cfg.policy, model=model) 34 her_reward_model = HerRewardModel(cfg.policy.other.her, cfg.policy.cuda) 35 36 task.use(interaction_evaluator(cfg, policy.eval_mode, evaluator_env)) 37 task.use(eps_greedy_handler(cfg)) 38 task.use(EpisodeCollector(cfg, policy.collect_mode, collector_env)) 39 task.use(data_pusher(cfg, buffer_)) 40 task.use(HERLearner(cfg, policy.learn_mode, buffer_, her_reward_model)) 41 task.use(CkptSaver(policy, cfg.exp_name, train_freq=100)) 42 task.run() 43 44 45if __name__ == "__main__": 46 main()