-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathFirm.py
283 lines (235 loc) · 12.3 KB
/
Firm.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
import random
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
import time
from sklearn.preprocessing import StandardScaler
from torch.nn import BatchNorm1d
import datetime
from RLenv import Environment
import logging
import os
dir_path = os.path.dirname(os.path.realpath(__file__))
handler = logging.FileHandler(os.path.join(dir_path, "log/firm_wkchange.log"))
handler.setFormatter(logging.Formatter('%(asctime)s %(message)s'))
logger = logging.getLogger("Debugging logger")
logger.setLevel(logging.DEBUG)
logger.addHandler(handler)
##################### hyper parameters ####################
MAX_EPISODES = 50
MAX_EP_STEPS = 10
LR_A = 0.001 # learning rate for actor
LR_C = 0.001 # learning rate for critic
GAMMA = 0.95 # reward discount
TAU = 0.005 # soft replacement
MEMORY_CAPACITY = 1500
BATCH_SIZE = 32
benchmark='search'
users=20
directory=f"/home/user/code/faas-resource/firm/{benchmark}/"
############################### DDPG ####################################
class ANet(nn.Module): # ae(s)=a
def __init__(self, s_dim, a_dim):
super(ANet, self).__init__()
self.fc1 = nn.Linear(s_dim, 64)
self.fc1.weight.data.normal_(0, 0.1) # initialization
self.fc2 = nn.Linear(64, 64)
self.fc2.weight.data.normal_(0, 0.1)
self.out = nn.Linear(64, a_dim)
self.bn1 = nn.BatchNorm1d(64)
self.bn2 = nn.BatchNorm1d(64)
self.bn3 = nn.BatchNorm1d(a_dim)
self.out.weight.data.normal_(0, 0.1)
def forward(self, x):
x = self.fc1(x)
x = F.relu(x)
x = self.bn1(x)
x = self.fc2(x)
x = F.relu(x)
x = self.bn2(x)
x = self.out(x)
x = torch.sigmoid(x)
return x
class CNet(nn.Module): # ae(s)=a
def __init__(self, s_dim, a_dim):
super(CNet, self).__init__()
self.fcs = nn.Linear(s_dim, 64)
self.fcs.weight.data.normal_(0, 0.1) # initialization
self.fca = nn.Linear(a_dim, 64)
self.fca.weight.data.normal_(0, 0.1) # initialization
self.fc = nn.Linear(64, 64)
self.fc.weight.data.normal_(0, 0.1) # initialization
self.out = nn.Linear(64, 1)
self.out.weight.data.normal_(0, 0.1) # initialization
self.bn1 = nn.BatchNorm1d(64)
def forward(self, s, a):
x = self.fcs(s)
y = self.fca(a)
net = F.relu(x + y)
net = self.fc(net)
net = self.bn1(net)
actions_value = self.out(net)
return actions_value
class DDPG(object):
def __init__(self, a_dim, s_dim):
self.a_dim, self.s_dim = a_dim, s_dim
# self.memory = np.zeros((MEMORY_CAPACITY, s_dim * 2 + a_dim + 2), dtype=np.float32)
self.pointer = 0
# self.sess = tf.Session()
self.Actor_eval = ANet(s_dim, a_dim)
self.Actor_target = ANet(s_dim, a_dim)
self.Critic_eval = CNet(s_dim, a_dim)
self.Critic_target = CNet(s_dim, a_dim)
self.ctrain = torch.optim.Adam(self.Critic_eval.parameters(), lr=LR_C)
self.atrain = torch.optim.Adam(self.Actor_eval.parameters(), lr=LR_A)
self.loss_td = nn.MSELoss()
self.device='cuda' if torch.cuda.is_available() else 'cpu'
# self.scaler = scaler
self.function_number=12
self.simulator=Environment(self.function_number)
self.a_dmin=3
self.s_dim=10
# self.buffer_a=np.loadtxt('/home/user/code/faas-resource/GAT-MF/experience/image1/ddpg/action_buffer.csv',delimiter=",", dtype=np.float32)
# self.buffer_r=np.loadtxt('/home/user/code/faas-resource/GAT-MF/experience/image1/ddpg/reward_buffer.csv',delimiter=",", dtype=np.float32).reshape(self.n,1)
# wk='all'
# self.buffer_s_origin=np.loadtxt('/home/user/code/faas-resource/model/sequence/10/gatmf/state_all.csv',delimiter=",", dtype=np.float32)
# scaled_buffer_state=self.scale_state(self.buffer_s_origin)
# self.buffer_s=scaled_buffer_state
self.buffer_s_origin=np.loadtxt(f'/home/user/code/faas-resource/GAT-MF/firm/{benchmark}/nextstate_search{users}.csv',delimiter=",", dtype=np.float32)
# self.buffer_s1_origin=np.loadtxt('/home/user/code/faas-resource/GAT-MF/experience/image1/ddpg/state_next_buffer.csv',delimiter=",", dtype=np.float32)
# scaled_buffer_state_next=self.scale_state(self.buffer_s1_origin)
# self.buffer_s1=scaled_buffer_state_next
def scale_state(self,x):
standardscaler = StandardScaler()
scaler=standardscaler.fit(self.buffer_s_origin) # scaler-> mean,var
x = scaler.transform(x)
return x
def choose_action(self, s): # s []
s = normalization(s, 1, self.scaler) # s [[]] for normalize
s = torch.FloatTensor(s)
self.Actor_eval.eval() # when BN or Dropout in testing ################
# s = torch.unsqueeze(torch.FloatTensor(s), 0)
act = self.Actor_eval(s)[0].detach() # ae（s）
self.Actor_eval.train() ###############
return act
def learn(self,round):
for iter in range(round):
if iter%200==0:
print(f'train {iter} iters')
for (target_param,param) in zip(self.Actor_target.parameters(),self.Actor_eval.parameters()):
target_param.data.copy_(
target_param.data*(1-TAU)+param.data*TAU
)
for (target_param,param) in zip(self.Critic_target.parameters(),self.Critic_eval.parameters()):
target_param.data.copy_(
target_param.data*(1-TAU)+param.data*TAU
)
# indices = np.random.choice(self.pointer, size=BATCH_SIZE)
# indices=[]
# for i in range(303):
# indices.append(i)
batch_index=np.random.randint(self.n,size=64)
bs=torch.FloatTensor(self.buffer_s[batch_index]).to(self.device)
ba=torch.FloatTensor(self.buffer_a[batch_index]).to(self.device)
bs_=torch.FloatTensor(self.buffer_s1[batch_index]).to(self.device) #s'
br=torch.FloatTensor(self.buffer_r[batch_index]).to(self.device)
a = self.Actor_eval(bs)
q = self.Critic_eval(bs, a)
loss_a = -torch.mean(q)
self.atrain.zero_grad()
loss_a.backward()
self.atrain.step()
a_ = self.Actor_target(bs_)
q_ = self.Critic_target(bs_, a_)
q_target = br + GAMMA * q_ # q_target = 负的
q_v = self.Critic_eval(bs, ba)
td_error = self.loss_td(q_target, q_v)
# print('critic td_error=',td_error)
self.ctrain.zero_grad()
td_error.backward()
self.ctrain.step()
def serving(self,users,benchmark):
agent_number=12
current_state=np.array([0.8,0.8,0.64062411,18.15,88.54,9.54,7409.67,4.3832,7.54225,20,0.358202992,0.2,0.306629904,20.41,278.64,24.7,7362.66,0.2418,0.44351,20,0.28994719,0.8,0.750424885,2.53,13.8,5.62,1219.59,0.0126,0.02663,20,0.8,0.69306447,0.2,21.96,43.32,9.18,12527.45,0.0058,0.0103,20,0.214844162,0.506682197,0.2,12.19,160.31,11.68,8671.59,0.05732,0.10466,20,0.784428164,0.599292552,0.490237864,4.99,3.15,7.67,2748.8,0.0081,0.0121,20,0.8,0.339434288,0.8,5.65,0.43,12.06,2842.8,0.0114,0.01645,20,0.529000031,0.725408583,0.49978863,5.95,8.18,8.24,5242.74,0.0318,0.04761,20,0.2,0.766296912,0.571430969,2.7,15.7,5.85,1163.21,0.0115,0.01938,20,0.438182836,0.73599337,0.213756483,6.63,2.91,6.86,3693.97,0.0075,0.0106,20,0.8,0.2,0.2,23.3,77.65,27.65,14996.27,0.0066,0.0158,20,0.606030909,0.372541165,0.60626224,6.61,12.19,12.16,3661.73,0.17625,0.33414,20])
current_state=self.scale_state(current_state.reshape(self.function_number,self.s_dim))
for j in range(15):
action=[]
for x in range(self.function_number):
state=current_state[x]
state=torch.FloatTensor(state).to(self.device)
self.Actor_eval.eval() # when BN or Dropout in testing
a = self.Actor_eval(state.unsqueeze(0))
# action = np.clip(np.random.normal(a, 0.1), 0.1, 0.9)
action.append(a.detach()[0])
arrays = [t.numpy() for t in action]
action_bar = np.concatenate(arrays, axis=0)
print("origin_action=",action_bar)
a = np.clip(np.random.normal(action_bar, 0.05), 0.2, 0.8)
print("action=",a)
reward,next_state,avg,p95,throughput,price=self.simulator.step(a,users,benchmark,False)
print_info = f'--The {j} iteration,{benchmark}-{users} price: {price}$,concurrency: {users}action: {a}, avg_e2e_latency: {avg} s,p95:{p95} throughput: {throughput},reward:{reward}'
current_state=next_state
current_state=np.array([current_state])
current_state=self.scale_state(current_state.reshape(self.function_number,self.s_dim))
print(print_info)
print(current_state)
logger.info(print_info)
def changewk(self,test_trace):
current_state=np.array([0.4,0.825621662,0.623262245,23.84,252.34,0.4,29845.24,5.56959,13.53646,25,0.883504416,0.882439015,0.4,31.41,80,0.6,51739.39,1.17778,1.97851,25,0.4,0.4,0.4,36.59,404.84,4.11,17563.2,11.67873,17.66522,25,0.4,0.4,0.4,23.93,248.45,0.98,13672.75,1.47443,3.27221,25,0.700099189,0.4,0.76061169,33.21,178.49,0.96,7307.37,3.02348,4.37693,25,0.4,0.4,0.4,42.21,488.15,5.99,34812.97,8.52047,19.31171,25])
current_state=self.scale_state(current_state.reshape(self.function_number,self.s_dim))
tracelen=87
for step in range(tracelen):
users=int(test_trace[step])
next_wk=int(test_trace[step+1])
print(f'the {step} iterations, the workload is {users},next_workload is {next_wk}')
action=[]
for x in range(self.function_number): # share networks among agents
state=current_state[x]
state=torch.FloatTensor(state).to(self.device)
self.Actor_eval.eval() # when BN or Dropout in testing ################
a = self.Actor_eval(state.unsqueeze(0))
# action = np.clip(np.random.normal(a, 0.1), 0.1, 0.9)
action.append(a.detach()[0])
arrays = [t.numpy() for t in action]
action_bar = np.concatenate(arrays, axis=0)
a = np.clip(np.random.normal(action_bar, 0.05), 0.3, 0.8)
print("action=",a)
reward,next_state,avg,p95,throughput,price=self.simulator.step(a,users) #apply进环境
print_info = f'--The {step} iteration,{benchmark}-{users},next-workload:{next_wk},price: {price}$,concurrency: {users}action: {a}, avg_e2e_latency: {avg} s, throughput: {throughput},reward:{reward}'
current_state=next_state
current_state=np.array([current_state])
current_state=self.scale_state(current_state.reshape(self.function_number,self.s_dim))
print(print_info)
print(current_state)
logger.info(print_info)
def load_model(self, episode):
model_name_c = "Critic" + str(episode) + ".pt"
model_name_a = "Actor" + str(episode) + ".pt"
self.Critic_target = torch.load(directory + model_name_c)
self.Critic_eval = torch.load(directory + model_name_c)
self.Actor_target = torch.load(directory + model_name_a)
self.Actor_eval= torch.load(directory + model_name_a)
def save_model(self, episode):
model_name_c = "Critic" + str(episode) + ".pt"
model_name_a = "Actor" + str(episode) + ".pt"
torch.save(self.Critic_eval, directory + model_name_c)
torch.save(self.Actor_eval, directory + model_name_a)
############################### training ####################################
def normalization(x,typeid,scaler):
# x:FloatTensor to array normalize then back to FloatTensor
if typeid==1: # choose action (s)
x=np.array([x])
x=scaler.transform(x)
else: # batch normal(s)
x=np.array(x)
x = scaler.transform(x)
x=torch.FloatTensor(x)
return x
if __name__=='__main__':
s_dim = 10
a_dim = 3
ddpg = DDPG(a_dim, s_dim)
ddpg.load_model('search-wk20-5k')
ddpg.serving(users,benchmark)
# ddpg.changewk()