Skip to content

ding.example.dqn_nstep_gymnasium

ding.example.dqn_nstep_gymnasium

Full Source Code

../ding/example/dqn_nstep_gymnasium.py

1import gymnasium as gym 2from ditk import logging 3from ding.model import DQN 4from ding.policy import DQNPolicy 5from ding.envs import DingEnvWrapper, BaseEnvManagerV2 6from ding.data import DequeBuffer 7from ding.config import compile_config 8from ding.framework import task 9from ding.framework.context import OnlineRLContext 10from ding.framework.middleware import OffPolicyLearner, StepCollector, interaction_evaluator, data_pusher, \ 11 eps_greedy_handler, CkptSaver, nstep_reward_enhancer, final_ctx_saver 12from ding.utils import set_pkg_seed 13from dizoo.classic_control.cartpole.config.cartpole_dqn_config import main_config, create_config 14 15 16def main(): 17 logging.getLogger().setLevel(logging.INFO) 18 main_config.exp_name = 'cartpole_dqn_nstep_gymnasium' 19 main_config.policy.nstep = 3 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: DingEnvWrapper(gym.make("CartPole-v0")) for _ in range(cfg.env.collector_env_num)], 24 cfg=cfg.env.manager 25 ) 26 evaluator_env = BaseEnvManagerV2( 27 env_fn=[lambda: DingEnvWrapper(gym.make("CartPole-v0")) for _ in range(cfg.env.evaluator_env_num)], 28 cfg=cfg.env.manager 29 ) 30 31 set_pkg_seed(cfg.seed, use_cuda=cfg.policy.cuda) 32 33 model = DQN(**cfg.policy.model) 34 buffer_ = DequeBuffer(size=cfg.policy.other.replay_buffer.replay_buffer_size) 35 policy = DQNPolicy(cfg.policy, model=model) 36 37 task.use(interaction_evaluator(cfg, policy.eval_mode, evaluator_env)) 38 task.use(eps_greedy_handler(cfg)) 39 task.use(StepCollector(cfg, policy.collect_mode, collector_env)) 40 task.use(nstep_reward_enhancer(cfg)) 41 task.use(data_pusher(cfg, buffer_)) 42 task.use(OffPolicyLearner(cfg, policy.learn_mode, buffer_)) 43 task.use(CkptSaver(policy, cfg.exp_name, train_freq=100)) 44 task.use(final_ctx_saver(cfg.exp_name)) 45 task.run() 46 47 48if __name__ == "__main__": 49 main()