forked from CSharpYDS/edge-computing-Q-learning
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdeep_q_learning_v2.py
183 lines (172 loc) · 7.29 KB
/
deep_q_learning_v2.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
from torch import FloatTensor
from brain import *
from env import *
from schedule_policy import *
from brain_v2 import *
import time
import matplotlib.pyplot as plt
from matplotlib.pyplot import MultipleLocator
from tqdm import tqdm
def parseState(x,y,z):
x = np.concatenate([x,y,z])
x = x.reshape((1, (2*N_SERVER+1)*N_JOB))
x = torch.from_numpy(np.array(x)).type(FloatTensor).to(device)
return x
def deepQLearning_v2(job_sequence, time1 = None):
agent = Agent_v2()
cost_min, cost_min1 = 10000000, 10000000
wrong = 0
ret_history = []
loss_arr = []
loss_idx = []
if time1 == None:
time1 = time.strftime("%Y-%m-%d_%H:%M:%S", time.localtime())
final_server_state = [[[],[]] for i in range(N_SERVER)]
for episode in tqdm(range(1200)):
S = Servers()
cost ,cost1 = 0, 0
history = []
idx = 0
state, state_next = State(), State()
tensor_state, tensor_state_next = None, None
server_id = -1
reward = 0
loss = 0
server_state = [[[],[]] for i in range(N_SERVER)]
for t in range(0, 100000):
if idx >= len(job_sequence) and S.done():
state_next.updateState(S, t)
agent.memorize(tensor_state, action.to(device), tensor_state_next, reward)
x = agent.update_q_function()
loss = loss + x
break
while True:
if idx < len(job_sequence) and job_sequence[idx].depart_time == t:
job = job_sequence[idx]
idx+=1
one_hot = np.zeros((1, N_JOB))
one_hot[0][job.job_id] += 1
if server_id!=-1:
state_next_trans, state_next_queue, _ = state_next.updateState(S, t)
tensor_state_next = parseState(state_next_trans, state_next_queue, one_hot)
agent.memorize(tensor_state, action.to(device), tensor_state_next, reward)
x = agent.update_q_function()
loss = loss + x
# new round state
state_trans, state_queue, _ = state.updateState(S, t)
tensor_state = parseState(state_trans, state_queue, one_hot)
# 2.1 获取action
action = agent.get_action(tensor_state, episode)
server_id = action.cpu().item() # get a number
action = action.cuda()
# 2.2 执行action
job.arrive_time = job.arriveTime(server_id)
S1 = Servers()
S1.clone(S)
reward = S1.getAddedCost(job, server_id, job.arrive_time)
S.servers[server_id].server.append(job)
reward = torch.from_numpy(np.array([-reward])).type(torch.FloatTensor).to(device)
else:
break
S, cost,cost1, history, server_state = SJFPolicy(S, t, cost, cost1,history, server_state) # 直接更新cost
if(episode % 2 == 0):
agent.update_target_q_function()
if cost_min > cost and cost_min1 > cost1:
ret_history = history
final_server_state = server_state
# print(cost, cost1)
# print(final_server_state)
cost_min = min(cost_min, cost)
cost_min1 = min(cost_min1, cost1)
loss_arr.append(loss.cpu().item())
loss_idx.append(episode)
if judge(history) == False or len(history) != len(job_sequence):
wrong += 1
print(history)
plt.figure()
plt.plot(loss_idx, loss_arr,'r', label='loss')
plt.savefig('loss_result/temp1/' + time1 + str(".png"))
plt.close()
# plt.show()
plt.figure()
plt.subplot(151)
plt.plot(final_server_state[0][1], final_server_state[0][0],'r', label='loss')
plt.subplot(152)
plt.plot(final_server_state[1][1], final_server_state[1][0],'r', label='loss')
plt.subplot(153)
plt.plot(final_server_state[2][1], final_server_state[2][0],'r', label='loss')
plt.subplot(154)
plt.plot(final_server_state[3][1], final_server_state[3][0],'r', label='loss')
plt.subplot(155)
plt.plot(final_server_state[4][1], final_server_state[4][0],'r', label='loss')
plt.savefig('loss_result/temp_server1/' + time1 + str(".png"))
plt.close()
torch.save(agent.brain.main_q_network, PATH)
print("wrong DQN1", wrong)
return cost_min, cost_min1, ret_history, loss_arr
def eval_v2(job_sequence, time1 = None):
agent = Agent_v2()
cost_min, cost_min1 = 10000000, 10000000
wrong = 0
ret_history = []
loss_arr = []
loss_idx = []
if time1 == None:
time1 = time.strftime("%Y-%m-%d_%H:%M:%S", time.localtime())
final_server_state = [[[],[]] for i in range(N_SERVER)]
for episode in (range(1)):
S = Servers()
cost ,cost1 = 0, 0
history = []
idx = 0
state, state_next = State(), State()
tensor_state, tensor_state_next = None, None
server_id = -1
reward = 0
loss = 0
server_state = [[[],[]] for i in range(N_SERVER)]
for t in range(0, 100000):
if idx >= len(job_sequence) and S.done():
break
while True:
if idx < len(job_sequence) and job_sequence[idx].depart_time == t:
job = job_sequence[idx]
idx+=1
one_hot = np.zeros((1, N_JOB))
one_hot[0][job.job_id] += 1
# new round state
state_trans, state_queue, _ = state.updateState(S, t)
tensor_state = parseState(state_trans, state_queue, one_hot)
# 2.1 获取action
action = agent.get_action(tensor_state, 0)
server_id = action.cpu().item() # get a number
action = action.cuda()
# 2.2 执行action
job.arrive_time = job.arriveTime(server_id)
S.servers[server_id].server.append(job)
else:
break
S, cost,cost1, history, server_state = SJFPolicy(S, t, cost, cost1,history, server_state) # 直接更新cost
if cost_min > cost and cost_min1 > cost1:
ret_history = history
final_server_state = server_state
cost_min = min(cost_min, cost)
cost_min1 = min(cost_min1, cost1)
# plt.show()
plt.figure()
plt.subplot(151)
plt.plot(final_server_state[0][1], final_server_state[0][0],'r', label='loss')
plt.subplot(152)
plt.plot(final_server_state[1][1], final_server_state[1][0],'r', label='loss')
plt.subplot(153)
plt.plot(final_server_state[2][1], final_server_state[2][0],'r', label='loss')
plt.subplot(154)
plt.plot(final_server_state[3][1], final_server_state[3][0],'r', label='loss')
plt.subplot(155)
plt.plot(final_server_state[4][1], final_server_state[4][0],'r', label='loss')
plt.savefig('loss_result/temp_server1/' + time1 + str(".png"))
plt.close()
if judge(history) == False or len(history) != len(job_sequence):
wrong += 1
print(history)
return cost_min, cost_min1, ret_history, loss_arr