SofaGym/agents/hyperparameters/stable_baselines_params.yml at main · SofaDefrost/SofaGym · GitHub

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
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
# Documentation for training hyperparameters `init_kwargs` for each specific algorithm are found at the Stable Baselines3 documentation for each algorithm respectively
#eg. PPO: https://stable-baselines3.readthedocs.io/en/master/modules/ppo.html#parameters
# `learning_rate`, `clip_range`, `clip_range_vf``, and `delta_std` could be defined as linear schedules. eg. "learning_rate: lin_0.001"

# fit_kwargs:
#   - total_timesteps: The total number of steps taken by the agent during each training epoch.
#   - max_episode_steps: The maximum number of steps per episode.
#   - eval_freq: Evaluate the agent during training after every `eval_freq` steps.
#   - n_eval_episodes: The number of episodes to evaluate the agent during each evaluation callback.
#   - save_freq: Save a trained model checkpoint after every `save_freq` steps.
#   - video_length: Max length of recorded videos.


PPO:
  init_kwargs:
    policy: 'MlpPolicy'
    learning_rate: 0.0003
    n_steps: 100
    batch_size: 64
    n_epochs: 10
    gamma: 0.99
    gae_lambda: 0.95
    clip_range: 0.2
    clip_range_vf: null
    normalize_advantage: True
    ent_coef: 0.0
    vf_coef: 0.5
    max_grad_norm: 0.5
    use_sde: False
    sde_sample_freq: -1
    target_kl: null
    policy_kwargs: "dict(net_arch=dict(pi=[512, 512, 512], vf=[512, 512, 512]), optimizer_class=th.optim.Adam, ortho_init=True, activation_fn=nn.Tanh)"
    device: "auto"
  fit_kwargs:
    total_timesteps: 1000
    max_episode_steps: 200
    eval_freq: 100
    n_eval_episodes: 5
    save_freq: 100
    video_length: 200

A2C:
  init_kwargs:
    policy: 'MlpPolicy'
    learning_rate: 0.0007
    n_steps: 100
    gamma: 0.99
    gae_lambda: 1.0
    ent_coef: 0.0
    vf_coef: 0.5
    max_grad_norm: 0.5
    rms_prop_eps: 1.0e-05
    use_rms_prop: True
    use_sde: False
    sde_sample_freq: -1
    normalize_advantage: False
    policy_kwargs: "dict(net_arch=dict(pi=[512, 512, 512], vf=[512, 512, 512]))"
    device: "auto"
  fit_kwargs:
    total_timesteps: 1000
    max_episode_steps: 200
    eval_freq: 100
    n_eval_episodes: 5
    save_freq: 100
    video_length: 200

DQN:
  init_kwargs:
    policy: 'MlpPolicy'
    learning_rate: 0.0001
    buffer_size: 1000000
    learning_starts: 100
    batch_size: 32
    tau: 1.0
    gamma: 0.99
    train_freq: (4, 'step')
    gradient_steps: 1
    replay_buffer_class: null
    replay_buffer_kwargs: null
    optimize_memory_usage: False
    target_update_interval: 10000
    exploration_fraction: 0.1
    exploration_initial_eps: 1.0
    exploration_final_eps: 0.05
    max_grad_norm: 10
    policy_kwargs: "dict(net_arch=[512, 512, 512], optimizer_class=th.optim.Adam)"
    device: "auto"
  fit_kwargs:
    total_timesteps: 1000
    max_episode_steps: 200
    eval_freq: 100
    n_eval_episodes: 5
    save_freq: 100
    video_length: 200

SAC:
  init_kwargs:
    policy: 'MlpPolicy'
    learning_rate: 0.0003
    buffer_size: 1000000
    learning_starts: 100
    batch_size: 256
    tau: 0.005
    gamma: 0.99
    train_freq: (1, 'step')
    gradient_steps: 1
    action_noise: null
    replay_buffer_class: null
    replay_buffer_kwargs: null
    optimize_memory_usage: False
    ent_coef: 'auto'
    target_update_interval: 1
    target_entropy: 'auto'
    use_sde: False
    sde_sample_freq: -1
    use_sde_at_warmup: False
    policy_kwargs: "dict(net_arch=dict(pi=[512, 512, 512], qf=[512, 512, 512]), optimizer_class=th.optim.Adam)"
    device: "auto"
  fit_kwargs:
    total_timesteps: 1000
    max_episode_steps: 200
    eval_freq: 100
    n_eval_episodes: 5
    save_freq: 100
    video_length: 200

TD3:
  init_kwargs:
    policy: 'MlpPolicy'
    learning_rate: 0.001
    buffer_size: 1000000
    learning_starts: 100
    batch_size: 100
    tau: 0.005
    gamma: 0.99
    train_freq: (1, 'episode')
    gradient_steps: -1
    action_noise: null
    replay_buffer_class: null
    replay_buffer_kwargs: null
    optimize_memory_usage: False
    policy_delay: 2
    target_policy_noise: 0.2
    target_noise_clip: 0.5
    policy_kwargs: "dict(net_arch=dict(pi=[512, 512, 512], qf=[512, 512, 512]), optimizer_class=th.optim.Adam)"
    device: "auto"
  fit_kwargs:
    total_timesteps: 1000
    max_episode_steps: 200
    eval_freq: 100
    n_eval_episodes: 5
    save_freq: 100
    video_length: 200

DDPG:
  init_kwargs:
    policy: 'MlpPolicy'
    learning_rate: 0.001
    buffer_size: 1000000
    learning_starts: 100
    batch_size: 100
    tau: 0.005
    gamma: 0.99
    train_freq: (1, 'episode')
    gradient_steps: -1
    action_noise: null
    replay_buffer_class: null
    replay_buffer_kwargs: null
    optimize_memory_usage: False
    policy_kwargs: "dict(net_arch=dict(pi=[512, 512, 512], qf=[512, 512, 512]), optimizer_class=th.optim.Adam)"
    device: "auto"
  fit_kwargs:
    total_timesteps: 1000
    max_episode_steps: 200
    eval_freq: 100
    n_eval_episodes: 5
    save_freq: 100
    video_length: 200