File size: 2,727 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
from copy import deepcopy
from easydict import EasyDict

spaceinvaders_ppg_config = dict(
    exp_name='spaceinvaders_ppg_seed0',
    env=dict(
        collector_env_num=16,
        evaluator_env_num=8,
        n_evaluator_episode=8,
        stop_value=10000000000,
        env_id='SpaceInvadersNoFrameskip-v4',
        #'ALE/SpaceInvaders-v5' is available. But special setting is needed after gym make.
        frame_stack=4,
        manager=dict(shared_memory=False, )
    ),
    policy=dict(
        cuda=True,
        model=dict(
            obs_shape=[4, 84, 84],
            action_shape=6,
            encoder_hidden_size_list=[32, 64, 64, 128],
            actor_head_hidden_size=128,
            critic_head_hidden_size=128,
            critic_head_layer_num=2,
        ),
        learn=dict(
            update_per_collect=24,
            batch_size=128,
            # (bool) Whether to normalize advantage. Default to False.
            adv_norm=False,
            learning_rate=0.0001,
            # (float) loss weight of the value network, the weight of policy network is set to 1
            value_weight=0.5,
            # (float) loss weight of the entropy regularization, the weight of policy network is set to 1
            entropy_weight=0.03,
            clip_ratio=0.1,
            epochs_aux=6,
            beta_weight=1,
            aux_freq=100
        ),
        collect=dict(
            # (int) collect n_sample data, train model n_iteration times
            n_sample=1024,
            # (float) the trade-off factor lambda to balance 1step td and mc
            gae_lambda=0.95,
            discount_factor=0.99,
        ),
        eval=dict(evaluator=dict(eval_freq=1000, )),
        other=dict(
            replay_buffer=dict(
                multi_buffer=True,
                policy=dict(
                    replay_buffer_size=100000,
                    max_use=3,
                ),
                value=dict(
                    replay_buffer_size=100000,
                    max_use=10,
                ),
            ),
        ),
    ),
)
spaceinvaders_ppg_config = EasyDict(spaceinvaders_ppg_config)
main_config = spaceinvaders_ppg_config
spaceinvaders_ppg_create_config = dict(
    env=dict(
        type='atari',
        import_names=['dizoo.atari.envs.atari_env'],
    ),
    env_manager=dict(type='subprocess'),
    policy=dict(type='ppg_offpolicy'),
)
spaceinvaders_ppg_create_config = EasyDict(spaceinvaders_ppg_create_config)
create_config = EasyDict(spaceinvaders_ppg_create_config)

if __name__ == '__main__':
    from dizoo.atari.entry.atari_ppg_main import main
    # PPG needs to use specific entry, you can run `dizoo/atari/entry/atari_ppg_main.py`
    main(main_config)