File size: 2,098 Bytes
079c32c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
import gym
from ditk import logging
from ding.model.template.qac_dist import QACDIST
from ding.policy import D4PGPolicy
from ding.envs import DingEnvWrapper, BaseEnvManagerV2
from ding.data import DequeBuffer
from ding.data.buffer.middleware import PriorityExperienceReplay
from ding.config import compile_config
from ding.framework import task
from ding.framework.context import OnlineRLContext
from ding.framework.middleware import OffPolicyLearner, StepCollector, interaction_evaluator, data_pusher, \
    CkptSaver, nstep_reward_enhancer
from ding.utils import set_pkg_seed
from dizoo.classic_control.pendulum.envs.pendulum_env import PendulumEnv
from dizoo.classic_control.pendulum.config.pendulum_d4pg_config import main_config, create_config


def main():
    logging.getLogger().setLevel(logging.INFO)
    cfg = compile_config(main_config, create_cfg=create_config, auto=True)
    with task.start(async_mode=False, ctx=OnlineRLContext()):
        collector_env = BaseEnvManagerV2(
            env_fn=[lambda: PendulumEnv(cfg.env) for _ in range(cfg.env.collector_env_num)], cfg=cfg.env.manager
        )
        evaluator_env = BaseEnvManagerV2(
            env_fn=[lambda: PendulumEnv(cfg.env) for _ in range(cfg.env.evaluator_env_num)], cfg=cfg.env.manager
        )

        set_pkg_seed(cfg.seed, use_cuda=cfg.policy.cuda)

        model = QACDIST(**cfg.policy.model)
        buffer_ = DequeBuffer(size=cfg.policy.other.replay_buffer.replay_buffer_size)
        buffer_.use(PriorityExperienceReplay(buffer_, IS_weight=True))
        policy = D4PGPolicy(cfg.policy, model=model)

        task.use(interaction_evaluator(cfg, policy.eval_mode, evaluator_env))
        task.use(
            StepCollector(cfg, policy.collect_mode, collector_env, random_collect_size=cfg.policy.random_collect_size)
        )
        task.use(nstep_reward_enhancer(cfg))
        task.use(data_pusher(cfg, buffer_))
        task.use(OffPolicyLearner(cfg, policy.learn_mode, buffer_))
        task.use(CkptSaver(policy, cfg.exp_name, train_freq=100))
        task.run()


if __name__ == "__main__":
    main()