-
Notifications
You must be signed in to change notification settings - Fork 17
/
main_train.py
283 lines (252 loc) · 10.8 KB
/
main_train.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
# debug field
import os
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
import argparse
# get argument from user
parser = argparse.ArgumentParser()
parser.add_argument('--drl', type = str, required = True, default='td3', help="which drl algo would you like to choose ['ddpg', 'td3']")
parser.add_argument('--reward', type = str, required = True, default='see', help="which reward would you like to implement ['ssr', 'see']")
parser.add_argument('--seeds', type = int, required = False, default=None, nargs='+', help="what seed(s) would you like to use for DRL 1 and 2, please provide in one or two int")
parser.add_argument('--ep-num', type = int, required = False, default=300, help="how many episodes do you want to train your DRL")
parser.add_argument('--trained-uav', default=False, action='store_true', help='use trained uav instead of retraining')
args = parser.parse_args()
DRL_ALGO = args.drl
REWARD_DESIGN = args.reward
SEEDS = args.seeds
EPISODE_NUM = args.ep_num
TRAINED_UAV = args.trained_uav
# process the argument
assert DRL_ALGO in ['ddpg', 'td3'], "drl must be ['ddpg', 'td3']"
assert REWARD_DESIGN in ['ssr', 'see'], "reward must be ['ssr', 'see']"
if SEEDS is not None:
assert len(SEEDS) in [1, 2] and isinstance(SEEDS[0], int) and isinstance(SEEDS[-1], int), "seeds must be a list of 1 or 2 integer"
if DRL_ALGO == 'td3':
from td3 import Agent
elif DRL_ALGO == 'ddpg':
from ddpg import Agent
import ddpg
from env import MiniSystem
import numpy as np
import math
import time
import torch
# 1 init system model
episode_num = EPISODE_NUM # recommend to be 300
episode_cnt = 0
step_num = 100
project_name = f'trained_uav/{DRL_ALGO}_{REWARD_DESIGN}' if TRAINED_UAV else f'scratch/{DRL_ALGO}_{REWARD_DESIGN}'
system = MiniSystem(
user_num=2,
RIS_ant_num=4,
UAV_ant_num=4,
if_dir_link=1,
if_with_RIS=True,
if_move_users=True,
if_movements=True,
reverse_x_y=(False, False),
if_UAV_pos_state = True,
reward_design = REWARD_DESIGN,
project_name = project_name,
step_num = step_num
)
if_Theta_fixed = False
if_G_fixed = False
if_BS = False
if_robust = True
# 2 init RL Agent
agent_1_param_dic = {}
agent_1_param_dic["alpha"] = 0.0001
agent_1_param_dic["beta"] = 0.001
agent_1_param_dic["input_dims"] = system.get_system_state_dim()
agent_1_param_dic["tau"] = 0.001
agent_1_param_dic["batch_size"] = 64
agent_1_param_dic["n_actions"] = system.get_system_action_dim() - 2
agent_1_param_dic["action_noise_factor"] = 0.1
agent_1_param_dic["memory_max_size"] = int(5/5 * episode_num * step_num) #/2
agent_1_param_dic["agent_name"] = "G_and_Phi"
agent_1_param_dic["layer1_size"] = 800
agent_1_param_dic["layer2_size"] = 600
agent_1_param_dic["layer3_size"] = 512
agent_1_param_dic["layer4_size"] = 256
agent_2_param_dic = {}
agent_2_param_dic["alpha"] = 0.0001
agent_2_param_dic["beta"] = 0.001
agent_2_param_dic["input_dims"] = 3
agent_2_param_dic["tau"] = 0.001
agent_2_param_dic["batch_size"] = 64
agent_2_param_dic["n_actions"] = 2
agent_2_param_dic["action_noise_factor"] = 0.5
agent_2_param_dic["memory_max_size"] = int(5/5 * episode_num * step_num) #/2
agent_2_param_dic["agent_name"] = "UAV"
agent_2_param_dic["layer1_size"] = 400
agent_2_param_dic["layer2_size"] = 300
agent_2_param_dic["layer3_size"] = 256
agent_2_param_dic["layer4_size"] = 128
if SEEDS is not None:
torch.manual_seed(SEEDS[0]) # 1
torch.cuda.manual_seed_all(SEEDS[0]) # 1
agent_1 = Agent(
alpha = agent_1_param_dic["alpha"],
beta = agent_1_param_dic["beta"],
input_dims = [agent_1_param_dic["input_dims"]],
tau = agent_1_param_dic["tau"],
env = system,
batch_size = agent_1_param_dic["batch_size"],
layer1_size=agent_1_param_dic["layer1_size"],
layer2_size=agent_1_param_dic["layer2_size"],
layer3_size=agent_1_param_dic["layer3_size"],
layer4_size=agent_1_param_dic["layer4_size"],
n_actions = agent_1_param_dic["n_actions"],
max_size = agent_1_param_dic["memory_max_size"],
agent_name= agent_1_param_dic["agent_name"]
)
if SEEDS is not None:
torch.manual_seed(SEEDS[-1]) # 2
torch.cuda.manual_seed_all(SEEDS[-1]) # 2
agent_2 = Agent(
alpha = agent_2_param_dic["alpha"],
beta = agent_2_param_dic["beta"],
input_dims = [agent_2_param_dic["input_dims"]],
tau = agent_2_param_dic["tau"],
env = system,
batch_size = agent_2_param_dic["batch_size"],
layer1_size=agent_2_param_dic["layer1_size"],
layer2_size=agent_2_param_dic["layer2_size"],
layer3_size=agent_2_param_dic["layer3_size"],
layer4_size=agent_2_param_dic["layer4_size"],
n_actions = agent_2_param_dic["n_actions"],
max_size = agent_2_param_dic["memory_max_size"],
agent_name= agent_2_param_dic["agent_name"]
)
if TRAINED_UAV:
benchmark = f'data/storage/benchmark/{DRL_ALGO}_{REWARD_DESIGN}_benchmark'
if DRL_ALGO == 'td3':
agent_2.load_models(
load_file_actor = benchmark + '/Actor_UAV_TD3',
load_file_critic_1 = benchmark + '/Critic_1_UAV_TD3',
load_file_critic_2 = benchmark + '/Critic_2_UAV_TD3'
)
elif DRL_ALGO == 'ddpg':
agent_2.load_models(
load_file_actor = benchmark + '/Actor_UAV_ddpg',
load_file_critic = benchmark + '/Critic_UAV_ddpg'
)
meta_dic = {}
print("***********************system information******************************")
print("folder_name: "+str(system.data_manager.store_path))
meta_dic['folder_name'] = system.data_manager.store_path
print("user_num: "+str(system.user_num))
meta_dic['user_num'] = system.user_num
print("if_dir: "+str(system.if_dir_link))
meta_dic['if_dir_link'] = system.if_dir_link
print("if_with_RIS: "+str(system.if_with_RIS))
meta_dic['if_with_RIS'] = system.if_with_RIS
print("if_user_m: "+str(system.if_move_users))
meta_dic['if_move_users'] = system.if_move_users
print("RIS_ant_num: "+str(system.RIS.ant_num))
meta_dic['system_RIS_ant_num'] = system.RIS.ant_num
print("UAV_ant_num: "+str(system.UAV.ant_num))
meta_dic['system_UAV_ant_num'] = system.UAV.ant_num
print("if_movements: "+str(system.if_movements))
meta_dic['system_if_movements'] = system.if_movements
print("reverse_x_y: "+str(system.reverse_x_y))
meta_dic['system_reverse_x_y'] = system.reverse_x_y
print("if_UAV_pos_state:"+str(system.if_UAV_pos_state))
meta_dic['if_UAV_pos_state'] = system.if_UAV_pos_state
print("ep_num: "+str(episode_num))
meta_dic['episode_num'] = episode_num
print("step_num: "+str(step_num))
meta_dic['step_num'] = step_num
print("***********************agent_1 information******************************")
tplt = "{0:{2}^20}\t{1:{2}^20}"
for i in agent_1_param_dic:
parm = agent_1_param_dic[i]
print(tplt.format(i, parm, chr(12288)))
meta_dic["agent_1"] = agent_1_param_dic
print("***********************agent_2 information******************************")
for i in agent_2_param_dic:
parm = agent_2_param_dic[i]
print(tplt.format(i, parm, chr(12288)))
meta_dic["agent_2"] = agent_2_param_dic
system.data_manager.save_meta_data(meta_dic)
print("***********************traning information******************************")
while episode_cnt < episode_num:
# 1 reset the whole system
system.reset()
step_cnt = 0
score_per_ep = 0
# 2 get the initial state
if if_robust:
tmp = system.observe()
#z = np.random.multivariate_normal(np.zeros(2), 0.5*np.eye(2), size=len(tmp)).view(np.complex128)
z = np.random.normal(size=len(tmp))
observersion_1 = list(
np.array(tmp) + 0.6 *1e-7* z
)
else:
observersion_1 = system.observe()
observersion_2 = list(system.UAV.coordinate)
if episode_cnt == 80:
print("break point")
while step_cnt < step_num:
# 1 count num of step in one episode
step_cnt += 1
# judge if pause the whole system
if not system.render_obj.pause:
# 2 choose action acoording to current state
action_1 = agent_1.choose_action(observersion_1, greedy=agent_1_param_dic["action_noise_factor"] * math.pow((1-episode_cnt / episode_num), 2))
action_2 = agent_2.choose_action(observersion_2, greedy=agent_2_param_dic["action_noise_factor"]* math.pow((1-episode_cnt / episode_num), 2))
if if_BS:
action_2[0]=0
action_2[1]=0
if if_Theta_fixed:
action_1[0+2 * system.UAV.ant_num * system.user_num:] = len(action_1[0+2 * system.UAV.ant_num * system.user_num:])*[0]
if if_G_fixed:
action_1[0:0+2 * system.UAV.ant_num * system.user_num]=np.array([-0.0313, -0.9838, 0.3210, 1.0, -0.9786, -0.1448, 0.3518, 0.5813, -1.0, -0.2803, -0.4616, -0.6352, -0.1449, 0.7040, 0.4090, -0.8521]) * math.pow(episode_cnt / episode_num, 2) * 0.7
#action_1[0:0+2 * system.UAV.ant_num * system.user_num]=len(action_1[0:0+2 * system.UAV.ant_num * system.user_num])*[0.5]
# 3 get newstate, reward
if system.if_with_RIS:
new_state_1, reward, done, info = system.step(
action_0=action_2[0],
action_1=action_2[1],
G=action_1[0:0+2 * system.UAV.ant_num * system.user_num],
Phi=action_1[0+2 * system.UAV.ant_num * system.user_num:],
set_pos_x=action_2[0],
set_pos_y=action_2[1]
)
new_state_2 = list(system.UAV.coordinate)
else:
new_state_1, reward, done, info = system.step(
action_0=action_2[0],
action_1=action_2[1],
G=action_1[0:0+2 * system.UAV.ant_num * system.user_num],
set_pos_x=action_2[0],
set_pos_y=action_2[1]
)
new_state_2 = list(system.UAV.coordinate)
score_per_ep += reward
# 4 store state pair into mem pool
agent_1.remember(observersion_1, action_1, reward, new_state_1, int(done))
agent_2.remember(observersion_2, action_2, reward, new_state_2, int(done))
# 5 update DDPG net
agent_1.learn()
if not TRAINED_UAV:
agent_2.learn()
#system.render_obj.render(0.001) # no rendering for faster
observersion_1 = new_state_1
observersion_2 = new_state_2
if done == True:
break
else:
#system.render_obj.render_pause() # no rendering for faster
time.sleep(0.001) #time.sleep(1)
system.data_manager.save_file(episode_cnt=episode_cnt)
system.reset()
print("ep_num: "+str(episode_cnt)+" ep_score: "+str(score_per_ep))
episode_cnt +=1
if episode_cnt % 10 == 0:
agent_1.save_models()
agent_2.save_models()
# save the last model
agent_1.save_models()
agent_2.save_models()