-
Notifications
You must be signed in to change notification settings - Fork 27
/
dqn_net06.py
208 lines (186 loc) · 7.24 KB
/
dqn_net06.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
# -*- coding: utf-8 -*-
"""
Created on Sat Sep 15 11:24:43 2018
Q / gamma = 0
Baseline 2 feature:
Baseline 3 feature: 1.535 1.514 1.520 1.505
Using Experience_Replay
@author: mengxiaomao / Email: mengxiaomaomao@outlook.com
"""
import time
import numpy as np
import tensorflow as tf
from DQN import DNN, DQN
from Environment_CU import Env_cellular
from Experience_replay import ReplayBuffer
fd = 10
Ts = 20e-3
n_x = 5
n_y = 5
L = 2
C = 16
maxM = 4 # user number in one BS
min_dis = 0.01 #km
max_dis = 1. #km
max_p = 38. #dBm
p_n = -114. #dBm
power_num = 10 #action_num
def Train(sess, env, weight_file):
max_reward = 0
batch_size = 500
max_episode = 5000
buffer_size = 50000
Ns = 11
env.set_Ns(Ns)
dnn = DNN(env, weight_file, max_episode = max_episode)
dqn = DQN(sess, dnn)
tf.global_variables_initializer().run()
deque = ReplayBuffer(buffer_size)
interval = 100
st = time.time()
reward_hist = list()
for k in range(1, max_episode+1):
reward_dqn_list = list()
s_actor, _ = env.reset()
for i in range(int(Ns)-1):
a = dqn.predict_a(s_actor)
p, a = dqn.select_action(a, k)
s_actor_next, _, rate, r = env.step(p)
deque.add(s_actor, a, rate)
s_actor = s_actor_next
reward_dqn_list.append(r)
if deque.size() > batch_size:
batch_s, batch_a, batch_r = deque.sample_batch(batch_size)
dqn.train(batch_s, batch_a, batch_r)
reward_hist.append(np.mean(reward_dqn_list)) # bps/Hz per link
if k % interval == 0:
reward = np.mean(reward_hist[-interval:])
if reward > max_reward:
dnn.save_params()
max_reward = reward
print("Episode(train):%d DQN: %.3f Time cost: %.2fs"
%(k, reward, time.time()-st))
st = time.time()
return reward_hist
def Test(sess, env, weight_file):
max_episode = 100
Ns = 5e2+1
env.set_Ns(Ns)
dnn = DNN(env, weight_file)
dqn = DQN(sess, dnn)
tf.global_variables_initializer().run()
dqn.load_params()
st = time.time()
reward_hist = list()
for k in range(1, max_episode+1):
reward_dqn_list = list()
s_actor, _ = env.reset()
for i in range(int(Ns)-1):
p = dqn.predict_p(s_actor)
s_actor_next, _, _, r = env.step(p)
s_actor = s_actor_next
reward_dqn_list.append(r)
reward_hist.append(np.mean(reward_dqn_list)) # bps/Hz per link
print("Episode(test):%d DQN: %.3f Time cost: %.2fs"
%(k, reward_hist[-1], time.time()-st))
st = time.time()
print("Test average rate: %.3f" %(np.mean(reward_hist)))
return reward_hist
def Test_dqn_mem(weight_file, max_episode, Ns, fd, max_dis, maxM):
env = Env_cellular(fd, Ts, n_x, n_y, L, C, maxM, min_dis, max_dis, max_p, p_n, power_num)
tf.reset_default_graph()
with tf.Session() as sess:
env.set_Ns(Ns)
dnn = DNN(env, weight_file)
dqn = DQN(sess, dnn)
tf.global_variables_initializer().run()
dqn.load_params()
reward_hist = list()
for k in range(1, max_episode+1):
reward_dqn_list = list()
s_actor, _ = env.reset()
for i in range(int(Ns)-1):
p = dqn.predict_p(s_actor)
s_actor_next, _, _, r = env.step(p)
s_actor = s_actor_next
reward_dqn_list.append(r)
reward_hist.append(np.mean(reward_dqn_list)) # bps/Hz per link
return np.mean(reward_hist)
def Test_dqn_all(weight_file, max_episode, Ns, fd, max_dis, maxM):
env = Env_cellular(fd, Ts, n_x, n_y, L, C, maxM, min_dis, max_dis, max_p, p_n, power_num)
tf.reset_default_graph()
with tf.Session() as sess:
env.set_Ns(Ns)
dnn = DNN(env, weight_file)
dqn = DQN(sess, dnn)
tf.global_variables_initializer().run()
dqn.load_params()
reward_hist = list()
for k in range(1, max_episode+1):
reward_dqn_list = list()
s_actor, _ = env.reset()
for i in range(int(Ns)-1):
p = dqn.predict_p(s_actor)
s_actor_next, _, _, r = env.step(p)
s_actor = s_actor_next
reward_dqn_list.append(r)
reward_hist.append(np.mean(reward_dqn_list)) # bps/Hz per link
return np.mean(reward_hist)
def Test_dqn_time(weight_file, max_episode, Ns, fd, max_dis, maxM):
env = Env_cellular(fd, Ts, n_x, n_y, L, C, maxM, min_dis, max_dis, max_p, p_n, power_num)
tf.reset_default_graph()
with tf.Session() as sess:
env.set_Ns(Ns)
dnn = DNN(env, weight_file)
dqn = DQN(sess, dnn)
tf.global_variables_initializer().run()
dqn.load_params()
time_cost = 0
for k in range(1, max_episode+1):
s_actor, _ = env.reset()
for i in range(int(Ns)-1):
st = time.time()
p = dqn.predict_p(s_actor)
time_cost = time_cost + time.time() - st
s_actor_next, _, _, _ = env.step(p)
s_actor = s_actor_next
return time_cost
def Test_dqn_mem_quan(weight_file, max_episode, Ns, power_num, fd, max_dis, maxM):
env = Env_cellular(fd, Ts, n_x, n_y, L, C, maxM, min_dis, max_dis, max_p, p_n, power_num)
tf.reset_default_graph()
with tf.Session() as sess:
env.set_Ns(Ns)
dnn = DNN(env, weight_file)
dqn = DQN(sess, dnn)
tf.global_variables_initializer().run()
dqn.load_params()
reward_hist = list()
for k in range(1, max_episode+1):
reward_dqn_list = list()
s_actor, _ = env.reset()
for i in range(int(Ns)-1):
p = dqn.predict_p(s_actor)
s_actor_next, _, _, r = env.step(p)
s_actor = s_actor_next
reward_dqn_list.append(r)
reward_hist.append(np.mean(reward_dqn_list)) # bps/Hz per link
return np.mean(reward_hist)
def Train_dqn_mem(weight_file, fd, max_dis, maxM):
env = Env_cellular(fd, Ts, n_x, n_y, L, C, maxM, min_dis, max_dis, max_p, p_n, power_num)
tf.reset_default_graph()
with tf.Session() as sess:
Train(sess, env, weight_file)
def Train_dqn_mem_quan(weight_file, power_num, fd, max_dis, maxM):
env = Env_cellular(fd, Ts, n_x, n_y, L, C, maxM, min_dis, max_dis, max_p, p_n, power_num)
tf.reset_default_graph()
with tf.Session() as sess:
Train(sess, env, weight_file)
if __name__ == "__main__":
env = Env_cellular(fd, Ts, n_x, n_y, L, C, maxM, min_dis, max_dis, max_p, p_n, power_num)
weight_file = 'C:/Software/workshop/python/dqn_6.mat'
tf.reset_default_graph()
with tf.Session() as sess:
train_hist = Train(sess, env, weight_file)
tf.reset_default_graph()
with tf.Session() as sess:
test_hist = Test(sess, env, weight_file)