File size: 2,335 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
from easydict import EasyDict

collector_env_num = 1
evaluator_env_num = 1
halfcheetah_ppo_config = dict(
    exp_name='halfcheetah_onppo_seed0',
    env=dict(
        env_id='HalfCheetah-v3',
        norm_obs=dict(use_norm=False, ),
        norm_reward=dict(use_norm=False, ),
        collector_env_num=collector_env_num,
        evaluator_env_num=evaluator_env_num,
        n_evaluator_episode=1,
        stop_value=12000,
    ),
    policy=dict(
        cuda=True,
        recompute_adv=True,
        action_space='continuous',
        model=dict(
            action_space='continuous',
            obs_shape=17,
            action_shape=6,
        ),
        learn=dict(
            epoch_per_collect=10,
            update_per_collect=1,
            batch_size=320,
            learning_rate=3e-4,
            value_weight=0.5,
            entropy_weight=0.001,
            clip_ratio=0.2,
            adv_norm=True,
            value_norm=True,
            # for onppo, when we recompute adv, we need the key done in data to split traj, so we must
            # use ignore_done=False here,
            # but when we add key traj_flag in data as the backup for key done, we could choose to use ignore_done=True
            # for halfcheetah, the length=1000
            ignore_done=True,
            grad_clip_type='clip_norm',
            grad_clip_value=0.5,
        ),
        collect=dict(
            collector_env_num=collector_env_num,
            n_sample=3200,
            unroll_len=1,
            discount_factor=0.99,
            gae_lambda=0.95,
        ),
        eval=dict(evaluator=dict(eval_freq=500, )),
    ),
)
halfcheetah_ppo_config = EasyDict(halfcheetah_ppo_config)
main_config = halfcheetah_ppo_config

halfcheetah_ppo_create_config = dict(
    env=dict(
        type='mujoco',
        import_names=['dizoo.mujoco.envs.mujoco_env'],
    ),
    env_manager=dict(type='subprocess'),
    # env_manager=dict(type='subprocess'),
    policy=dict(type='ppo', ),
)
halfcheetah_ppo_create_config = EasyDict(halfcheetah_ppo_create_config)
create_config = halfcheetah_ppo_create_config

if __name__ == "__main__":
    # or you can enter `ding -m serial_onpolicy -c halfcheetah_onppo_config.py -s 0`
    from ding.entry import serial_pipeline_onpolicy
    serial_pipeline_onpolicy((main_config, create_config), seed=0)