-
Notifications
You must be signed in to change notification settings - Fork 5
/
D_settings.py
284 lines (279 loc) · 13.7 KB
/
D_settings.py
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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
import argparse
from common import preprocess_graph
import numpy as np
def parse_arguments():
parser = argparse.ArgumentParser("Deep RL for Cooperative Multi-Agent Control with Discrete Action Space")
parser.add_argument("--env-name", type=str, default="navigation6v6")
parser.add_argument("--agent-name", type=str, default="IDQN", help="IDQN, VDN, QMIX, NCC_VDN, NCC_QMIX, etc.")
parser.add_argument("--head-count", type=int, default=4, help="number of heads in DGN, etc.")
parser.add_argument("--hidden-layer-count", type=int, default=2, help="number of hidden layers")
parser.add_argument("--results-dir", type=str, default="./results/")
args = parser.parse_args()
args.exp_name = args.env_name + "-" + args.agent_name
if args.env_name in ["navigation2v2", "navigation3v3", "navigation4v4", "navigation6v6", "navigation10v10"]:
if args.env_name == "navigation2v2":
args.agent_count = 2
args.landmark_count = 2
args.observation_dim_list = [8, 8]
args.action_dim_list = [5, 5] # stop, up, right, down, left
args.adj = [[0, 1],
[1, 0]] # this ADJ does not contain self-connections
if args.agent_name in ['Contrastive_VDN', 'Contrastive_QMIX']: # specially designed for Contrastive_*
raise ValueError('navigation2v2 is unsuitable for testing Contrastive_NCC/QMIX models ...')
elif args.env_name == "navigation3v3":
args.agent_count = 3
args.landmark_count = 3
args.observation_dim_list = [12, 12, 12]
args.action_dim_list = [5, 5, 5]
args.adj = [[0, 1, 1],
[1, 0, 1],
[1, 1, 0]]
if args.agent_name in ['Contrastive_VDN', 'Contrastive_QMIX']: # specially designed for Contrastive_*
raise ValueError('navigation3v3 is unsuitable for testing Contrastive_NCC/QMIX models ...')
elif args.env_name == "navigation4v4":
args.agent_count = 4
args.landmark_count = 4
args.observation_dim_list = [16, 16, 16, 16]
args.action_dim_list = [5, 5, 5, 5]
args.adj = [[0, 1, 1, 1],
[1, 0, 1, 1],
[1, 1, 0, 1],
[1, 1, 1, 0]]
if args.agent_name in ['Contrastive_VDN', 'Contrastive_QMIX']: # specially designed for Contrastive_NCC
args.adj = [[0, 1, 0, 0],
[1, 0, 0, 0],
[0, 0, 0, 1],
[0, 0, 1, 0]]
elif args.env_name == "navigation6v6":
args.agent_count = 6
args.landmark_count = 6
args.observation_dim_list = [24, 24, 24, 24, 24, 24]
args.action_dim_list = [5, 5, 5, 5, 5, 5]
args.adj = [[0, 1, 1, 1, 1, 1],
[1, 0, 1, 1, 1, 1],
[1, 1, 0, 1, 1, 1],
[1, 1, 1, 0, 1, 1],
[1, 1, 1, 1, 0, 1],
[1, 1, 1, 1, 1, 0]]
if args.agent_name in ['Contrastive_VDN', 'Contrastive_QMIX']: # specially designed for Contrastive_*
args.adj = [[0, 1, 1, 0, 0, 0],
[1, 0, 1, 0, 0, 0],
[1, 1, 0, 0, 0, 0],
[0, 0, 0, 0, 1, 1],
[0, 0, 0, 1, 0, 1],
[0, 0, 0, 1, 1, 0]]
elif args.env_name == "navigation10v10":
args.agent_count = 10
args.landmark_count = 10
args.observation_dim_list = [40, 40, 40, 40, 40, 40, 40, 40, 40, 40]
args.action_dim_list = [5, 5, 5, 5, 5, 5, 5, 5, 5, 5]
args.adj = [[0, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 0, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 0, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 0, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 0, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 0, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 0, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 0]]
if args.agent_name in ['Contrastive_VDN', 'Contrastive_QMIX']: # specially designed for Contrastive_*
args.adj = [[0, 1, 1, 1, 1, 0, 0, 0, 0, 0],
[1, 0, 1, 1, 1, 0, 0, 0, 0, 0],
[1, 1, 0, 1, 1, 0, 0, 0, 0, 0],
[1, 1, 1, 0, 1, 0, 0, 0, 0, 0],
[1, 1, 1, 1, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 1, 1, 1, 1],
[0, 0, 0, 0, 0, 1, 0, 1, 1, 1],
[0, 0, 0, 0, 0, 1, 1, 0, 1, 1],
[0, 0, 0, 0, 0, 1, 1, 1, 0, 1],
[0, 0, 0, 0, 0, 1, 1, 1, 1, 0]]
args.env_rename = "CooperativeNavigation"
args.adj_norm = preprocess_graph(args.adj)
args.env_bound = 10
args.env_dim = 2 # do not change this, since it is co-related with observation_dim_list
args.action_effective_step = 1
#
args.buffer_size = 100000
args.batch_size = 64
args.exp_count = 5
args.hidden_dim = 32
args.lr = 1e-3
args.clipped_norm_value = 10.0
args.seed = 0
#
args.episode_count = 2000
args.epsilon = 1.0
args.epsilon_delta = 0.001
args.epsilon_end = 0.0
args.episode_for_updating_T = 20
args.max_episode_len = 10
args.gamma = 0.9
#
args.alpha_L2 = 0.1 # the weight of L2-loss in NCC/Pikachu
args.alpha_KL = 0.1 # the weight of KL-loss in NCC/Pikachu
args.alpha_PCA = 0.1 # the weight of PCA-loss in Pikachu
args.alpha_PCA_end = 0.0 # 0.1 or 0.0, where 0.1 means fixed alpha_PCA
args.alpha_PCA_delta = (args.alpha_PCA - args.alpha_PCA_end) / args.episode_count
args.alpha_CON = 0.1 # the weight of contrastive-loss
elif args.env_name in ["simple", "simple_adversary", "simple_crypto", "simple_push", "simple_reference", "simple_speaker_listener", "simple_spread", "simple_tag", "simple_world_comm"]:
from MPE import make_env
args.env = make_env.make_env(args.env_name, benchmark=False)
# print("********************8")
args.env_rename = "MPE"
args.agent_count = len(args.env.world.agents)
args.landmark_count = len(args.env.world.landmarks)
print('args.agent_count ==>', args.agent_count)
print('args.landmark_count ==>', args.landmark_count)
args.observation_dim_list = [i.shape[0] for i in args.env.observation_space]
if args.env_name == "simple_reference":
raise ValueError('something wrong for the [simple_reference] environment ...')
# print("args.env.action_spaces", args.env.action_space)
args.action_dim_list = [i.num_discrete_space for i in args.env.action_space]
elif args.env_name == "simple_world_comm":
raise ValueError('something wrong for the [simple_reference] environment ...')
# print("args.env.action_spaces", args.env.action_space)
args.action_dim_list = []
args.action_dim_list.append(args.env.action_space[0].num_discrete_space)
args.action_dim_list.append(args.env.action_space[1].n)
args.action_dim_list.append(args.env.action_space[2].n)
args.action_dim_list.append(args.env.action_space[3].n)
args.action_dim_list.append(args.env.action_space[4].n)
args.action_dim_list.append(args.env.action_space[5].n)
else:
args.action_dim_list = [i.n for i in args.env.action_space]
args.adj = np.ones((args.agent_count, args.agent_count), dtype=int) - np.eye(args.agent_count, dtype=int)
args.adj_norm = preprocess_graph(args.adj)
args.env_bound = 10
args.env_dim = 2 # do not change this, since it is co-related with observation_dim_list
args.action_effective_step = 1
#
args.buffer_size = 100000
args.batch_size = 64
args.exp_count = 5
args.hidden_dim = 32
args.lr = 1e-3
args.clipped_norm_value = 10.0
args.seed = 0
#
args.episode_count = 2000
args.epsilon = 1.0
args.epsilon_delta = 0.001
args.epsilon_end = 0.0
args.episode_for_updating_T = 20
args.max_episode_len = 10
args.gamma = 0.9
#
args.alpha_L2 = 0.1 # the weight of L2-loss in NCC/Pikachu
args.alpha_KL = 0.1 # the weight of KL-loss in NCC/Pikachu
args.alpha_PCA = 0.1 # the weight of PCA-loss in Pikachu
args.alpha_PCA_end = 0.0 # 0.1 or 0.0, where 0.1 means fixed alpha_PCA
args.alpha_PCA_delta = (args.alpha_PCA - args.alpha_PCA_end) / args.episode_count
args.alpha_CON = 0.1 # the weight of contrastive-loss
elif args.env_name in ["2s3z", "3s5z", "1c3s5z", "3m"]:
if args.env_name == '2s3z':
args.n_agents = 5
args.n_actions = 11
args.obs_shape = 80
args.state_shape = 120
args.episode_limit = 120
#
args.agent_count = args.n_agents
args.observation_dim_list = [args.obs_shape for _ in range(args.agent_count)]
args.action_dim_list = [args.n_actions for _ in range(args.agent_count)]
args.adj = [[0, 1, 1, 1, 1],
[1, 0, 1, 1, 1],
[1, 1, 0, 1, 1],
[1, 1, 1, 0, 1],
[1, 1, 1, 1, 0]]
elif args.env_name == '3s5z':
args.n_agents = 8
args.n_actions = 14
args.obs_shape = 128
args.state_shape = 216
args.episode_limit = 150
#
args.agent_count = args.n_agents
args.observation_dim_list = [args.obs_shape for _ in range(args.agent_count)]
args.action_dim_list = [args.n_actions for _ in range(args.agent_count)]
args.adj = [[0, 1, 1, 1, 1, 1, 1, 1],
[1, 0, 1, 1, 1, 1, 1, 1],
[1, 1, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 0, 1, 1, 1, 1],
[1, 1, 1, 1, 0, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 1, 1],
[1, 1, 1, 1, 1, 1, 0, 1],
[1, 1, 1, 1, 1, 1, 1, 0]]
elif args.env_name == '1c3s5z':
args.n_agents = 9
args.n_actions = 15
args.obs_shape = 162
args.state_shape = 270
args.episode_limit = 180
#
args.agent_count = args.n_agents
args.observation_dim_list = [args.obs_shape for _ in range(args.agent_count)]
args.action_dim_list = [args.n_actions for _ in range(args.agent_count)]
args.adj = [[0, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 0, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 0, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 0, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 0, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 0, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 0]]
elif args.env_name == '3m':
args.n_agents = 3
args.n_actions = 9
args.obs_shape = 30
args.state_shape = 48
args.episode_limit = 60
#
args.agent_count = args.n_agents
args.observation_dim_list = [args.obs_shape for _ in range(args.agent_count)]
args.action_dim_list = [args.n_actions for _ in range(args.agent_count)]
args.adj = [[0, 1, 1],
[1, 0, 1],
[1, 1, 0]]
args.env_rename = "StarCraftII"
args.adj_norm = preprocess_graph(args.adj)
args.difficulty = '7' # the difficulty of the game, 7==VeryHard
args.game_version = 'latest'
args.step_mul = 8 # how many steps to make an action
args.replay_dir = '/home/noahrl/workspace/sc2_replay/' # MUST be an absolute path to save the replay
args.evaluate_episode = 20 # number of episode to evaluate the agent
args.evaluate_cycle = 100 # every evaluate_cycle episodes to evaluate the agent
args.last_action = False # whether to use the last action to choose action
#
args.buffer_size = int(5e3)
args.batch_size = 32
args.hidden_dim = 32
args.hidden_dim_rnn = 64
args.lr = 5e-4
args.clipped_norm_value = 10.0
args.seed = 0
#
args.episode_count = 10000
args.epsilon = 1.0
args.epsilon_end = 0.05
args.epsilon_anneal_scale = 'step' # step or episode
anneal_steps = 50000
args.epsilon_delta = (args.epsilon - args.epsilon_end) / anneal_steps
#
args.exp_count = 5
args.train_steps_one_episode = 1
args.episode_for_saving_model = 2000 # how often to save the model
args.episode_for_updating_T = 200 # how often to update the target_net
args.target_update_cycle = args.episode_for_updating_T
args.max_episode_len = 1000
args.gamma = 0.99
#
args.alpha_L2 = 0.1 # the weight of L2-loss in NCC/QWEIGHT
args.alpha_KL = 0.1 # the weight of KL-loss in NCC/QWEIGHT
else:
raise ValueError("args.env_name is not defined! ...")
return args
if __name__ == '__main__':
args = parse_arguments()
print(args)