ding.example.dqn_frozen_lake

ding.example.dqn_frozen_lake

Full Source Code

../ding/example/dqn_frozen_lake.py

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