File size: 3,743 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
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
import os
from easydict import EasyDict
from tensorboardX import SummaryWriter

from ding.config import compile_config
from ding.worker import BaseLearner, SampleSerialCollector, InteractionSerialEvaluator, AdvancedReplayBuffer
from ding.envs.env_manager.envpool_env_manager import PoolEnvManager
from ding.policy import DQNPolicy
from ding.model import DQN
from ding.utils import set_pkg_seed
from ding.rl_utils import get_epsilon_greedy_fn
from dizoo.atari.config.serial import pong_dqn_envpool_config


def main(cfg, seed=0, max_iterations=int(1e10)):
    cfg.exp_name = 'atari_dqn_envpool'
    cfg = compile_config(
        cfg,
        PoolEnvManager,
        DQNPolicy,
        BaseLearner,
        SampleSerialCollector,
        InteractionSerialEvaluator,
        AdvancedReplayBuffer,
        save_cfg=True
    )
    collector_env_cfg = EasyDict(
        {
            'env_id': cfg.env.env_id,
            'env_num': cfg.env.collector_env_num,
            'batch_size': cfg.env.collector_batch_size,
            # env wrappers
            'episodic_life': True,  # collector: True
            'reward_clip': True,  # collector: True
            'gray_scale': cfg.env.get('gray_scale', True),
            'stack_num': cfg.env.get('stack_num', 4),
            'frame_skip': cfg.env.get('frame_skip', 4),
        }
    )
    collector_env = PoolEnvManager(collector_env_cfg)
    evaluator_env_cfg = EasyDict(
        {
            'env_id': cfg.env.env_id,
            'env_num': cfg.env.evaluator_env_num,
            'batch_size': cfg.env.evaluator_batch_size,
            # env wrappers
            'episodic_life': False,  # evaluator: False
            'reward_clip': False,  # evaluator: False
            'gray_scale': cfg.env.get('gray_scale', True),
            'stack_num': cfg.env.get('stack_num', 4),
            'frame_skip': cfg.env.get('frame_skip', 4),
        }
    )
    evaluator_env = PoolEnvManager(evaluator_env_cfg)
    collector_env.seed(seed)
    evaluator_env.seed(seed)
    set_pkg_seed(seed, use_cuda=cfg.policy.cuda)

    model = DQN(**cfg.policy.model)
    policy = DQNPolicy(cfg.policy, model=model)
    tb_logger = SummaryWriter(os.path.join('./{}/log/'.format(cfg.exp_name), 'serial'))
    learner = BaseLearner(cfg.policy.learn.learner, policy.learn_mode, tb_logger, exp_name=cfg.exp_name)
    collector = SampleSerialCollector(
        cfg.policy.collect.collector, collector_env, policy.collect_mode, tb_logger, exp_name=cfg.exp_name
    )
    evaluator = InteractionSerialEvaluator(
        cfg.policy.eval.evaluator, evaluator_env, policy.eval_mode, tb_logger, exp_name=cfg.exp_name
    )
    replay_buffer = AdvancedReplayBuffer(
        cfg.policy.other.replay_buffer, tb_logger, exp_name=cfg.exp_name, instance_name='replay_buffer'
    )
    eps_cfg = cfg.policy.other.eps
    epsilon_greedy = get_epsilon_greedy_fn(eps_cfg.start, eps_cfg.end, eps_cfg.decay, eps_cfg.type)

    while True:
        if evaluator.should_eval(learner.train_iter):
            stop, reward = evaluator.eval(learner.save_checkpoint, learner.train_iter, collector.envstep)
            if stop:
                break
        eps = epsilon_greedy(collector.envstep)
        new_data = collector.collect(train_iter=learner.train_iter, policy_kwargs={'eps': eps})
        replay_buffer.push(new_data, cur_collector_envstep=collector.envstep)
        for i in range(cfg.policy.learn.update_per_collect):
            batch_size = learner.policy.get_attribute('batch_size')
            train_data = replay_buffer.sample(batch_size, learner.train_iter)
            if train_data is not None:
                learner.train(train_data, collector.envstep)


if __name__ == "__main__":
    main(EasyDict(pong_dqn_envpool_config))