forked from Xiaoyinggit/ConUCB
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Simulation_mult_process_pool_new.py
283 lines (234 loc) · 10.3 KB
/
Simulation_mult_process_pool_new.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
##input
# -- alg: name of algorithm
# -- input_files: input folder
# -- out_files: output_folder
import os
import json
import argparse
import Arm
import SupArm
import conf
import os
import numpy as np
from LinUCB import LinUCB
from Con_UCB import Con_UCB
import User
import random
import datetime
from multiprocessing import Pool
thread_num=10
class simulateExp:
def __init__(self, users, arms, suparms,out_folder, pool_size,batchSize=50,noise=None, suparm_noise=None,test_iter=1000, alias='time', dim=50):
self.users=users
self.all_arms=arms
self.suparms=suparms
self.out_folder=out_folder
self.noise=noise
self.suparm_noise=suparm_noise
self.batchSize=batchSize
self.poolArticleSize=pool_size
self.test_iter=test_iter
self.alias=alias
self.dim=dim
def getReward(self,u, arm):
return np.dot(u.theta.T, arm.fv)
def regulateArticlePool(self):
# Randomly generate articles
all_index=range(0,len(self.all_arms))
selected_pool_index=np.random.choice(all_index,self.poolArticleSize, replace=False)
self.armPool={}
for si in selected_pool_index:
self.armPool[si]=self.all_arms[si]
if len(self.armPool) !=self.poolArticleSize:
raise AssertionError
def getSuparmReward(self, u, suparm):
return np.dot(u.theta.T,suparm.fv)
def getOptimalReward(self,u, article_pool):
maxReward=float('-inf')
best_article=None
for x,x_o in article_pool.items():
reward=self.getReward(u,x_o)
if reward> maxReward:
best_article=x_o
maxReward=reward
if best_article==None:
raise AssertionError
return maxReward, best_article
def getL2Diff(self, x, y):
return np.linalg.norm(x-y) # L2 norm
def getX(self,arms):
X_t=np.zeros((len(arms), self.dim))
i=0
for aid, ainfo in arms.items():
X_t[i,:]=ainfo.fv.T
i+=1
return X_t
def getAddiBudget(self, cur_bt, iter_):
left_budget=-1
if iter_==0:
left_budget=cur_bt(iter_)
else:
left_budget=cur_bt(iter_)-cur_bt(iter_-1)
if left_budget>0:
return int(left_budget)
else:
return -1
def simulationPerUser(self, u,test_iter):
process_id=os.getpid()
print('[simulationPerUser] uid: %d, process_id: %d'%(u.uid, process_id))
user_regret={}
theta_diff={}
user_regret_file=os.path.join(self.out_folder,'users_regret/%d.txt'%u.uid)
debug_fw = None
with open(user_regret_file,'w') as fw:
for iter_ in range(0, test_iter):
Addi_budget=self.getAddiBudget(algorithms['Arm-Con'].bt, iter_)
print('[simulationPerUser] uid: %d, iter: %d, addi_budget: %d'%(u.uid, iter_, Addi_budget))
self.regulateArticlePool()
cur_iter_noise=self.noise()
cur_iter_suparm_noise=self.suparm_noise()
Optimal_Reward, OptimalArticle=self.getOptimalReward(u,self.armPool)
try:
tmp=user_regret[iter_]
tmp=theta_diff[iter_]
raise AssertionError
except:
user_regret[iter_]={}
theta_diff[iter_]={}
for algname,alg in algorithms.items():
pickedArticle=None
if algname in ["ConUCB", "Var-RS", "Var-MRC", "Var-LCR"]:
tmp_budget=Addi_budget
X_t=None
if Addi_budget>0:
X_t=self.getX(self.armPool)
while tmp_budget>0:
pickedsuparm=alg.decide_suparms(self.suparms,u.uid, np.linalg.norm(u.theta),arms=self.armPool, X_t=X_t, debug_fw=debug_fw)
reward=self.getSuparmReward(u,pickedsuparm)+cur_iter_suparm_noise
alg.updateSuparmParameters(pickedsuparm,reward,u.uid)
tmp_budget-=1
if Addi_budget>0:
alg.increaseSuparmTimes(u.uid)
if algname=='Arm-Con':
lh_budget=Addi_budget
while lh_budget>0:
pickedAddiArm=alg.decide(self.armPool, u.uid, np.linalg.norm(u.theta),debug_fw=debug_fw, best_arm=OptimalArticle.id)
reward=self.getReward(u,pickedAddiArm)+cur_iter_noise
alg.updateParameters(pickedAddiArm,reward, u.uid)
lh_budget-=1
pickedArticle=None
if algname=='Random':
pickedIndex=np.random.choice(list(self.armPool.keys()), 1, replace=False)
pickedArticle=self.armPool[pickedIndex[0]]
else:
pickedArticle=alg.decide(self.armPool, u.uid, np.linalg.norm(u.theta),debug_fw=debug_fw, best_arm=OptimalArticle.id)
if pickedArticle==None:
raise AssertionError
reward=self.getReward(u,pickedArticle)+cur_iter_noise
if algname!='Random':
alg.updateParameters(pickedArticle,reward, u.uid)
# calculate regret
regret=Optimal_Reward+cur_iter_noise-reward
user_regret[iter_][algname]=regret
#calculate theta_dff
iter_theta_diff=-1
if algname!='Random':
iter_theta_diff=self.getL2Diff(u.theta,alg.getTheta(u.uid))
theta_diff[iter_][algname]=iter_theta_diff
fw.write('iter:%d\talgname:%s\tregret:%f\ttheta_diff:%f\n'%(iter_,algname,regret,iter_theta_diff))
return user_regret, theta_diff
def runAlgorithms(self, algorithms):
if self.alias=='time':
self.starttime=datetime.datetime.now()
timeRun=self.starttime.strftime('_%m_%d_%H_%M')
self.alias=timeRun
out_regret_file=os.path.join(self.out_folder, "AccRegret"+self.alias+'.csv')
out_theta_file=os.path.join(self.out_folder, "AccTheta"+self.alias+'.csv')
AlgRegret={}
AlgThetaD={}
BatchCumulateRegret={}
BatchAvgThetaD={}
for algname, alg in algorithms.items():
AlgRegret[algname]=[]
BatchCumulateRegret[algname]=[]
AlgThetaD[algname]=[]
BatchAvgThetaD[algname]=[]
with open(out_regret_file,'w') as fw:
fw.write('Time(Iteration)\t')
fw.write(','.join([str(alg_name) for alg_name in algorithms.keys()]))
fw.write('\n')
with open(out_theta_file,'w') as fw:
fw.write('Time(Iteration)\t')
fw.write(','.join([str(alg_name) for alg_name in algorithms.keys()]))
fw.write('\n')
#simulation
print('[runAlgorithms] Training iterations: %d'%self.test_iter)
pool= Pool(processes=thread_num)
results=[]
for uid, user in self.users.items():
result=pool.apply_async(self.simulationPerUser,(user,self.test_iter))
results.append(result)
pool.close()
pool.join()
all_user_regret=[]
all_theta_diff=[]
for result in results:
tmp_regret,tmp_theta_diff=result.get()
all_user_regret.append(tmp_regret)
all_theta_diff.append(tmp_theta_diff)
for iter_ in range(self.test_iter):
for ure in all_user_regret:
for algname, reg in ure[iter_].items():
AlgRegret[algname].append(reg)
for ud in all_theta_diff:
for algname, thetad in ud[iter_].items():
AlgThetaD[algname].append(thetad)
if iter_%self.batchSize==0:
for alg_name in algorithms.keys():
BatchCumulateRegret[alg_name].append(sum(AlgRegret[alg_name]))
BatchAvgThetaD[alg_name].append(sum(AlgThetaD[alg_name]))
AlgThetaD[alg_name]=[]
with open(out_regret_file,'a+') as f:
f.write(str(iter_)+'\t')
f.write(','.join([str(BatchCumulateRegret[alg_name][-1]) for alg_name in algorithms.keys()]))
f.write('\n')
with open(out_theta_file,'a+') as f:
f.write(str(iter_)+'\t')
f.write(','.join([str(BatchAvgThetaD[alg_name][-1]) for alg_name in algorithms.keys()]))
f.write('\n')
finalRegret={}
for alg_name in algorithms.keys():
finalRegret[alg_name]=BatchCumulateRegret[alg_name][:-1]
return finalRegret
def noise():
return np.random.normal(scale=conf.armNoiseScale)
def suparm_noise():
return random.gauss(mu=0,sigma=conf.suparmNoiseScale)
if __name__=='__main__':
parser=argparse.ArgumentParser(description='')
parser.add_argument('--in_folder',dest='in_folder',help='input the folder containing input files')
parser.add_argument('--out_folder',dest='out_folder', help='input the folder to output')
parser.add_argument('--poolSize',dest='poolSize',type=int, help='poolSize of each iteration')
parser.add_argument('--seedIndex',dest='seedIndex',type=int, help='seedIndex')
args=parser.parse_args()
#load arms
np.random.seed(conf.seeds_set[args.seedIndex])
random.seed(conf.seeds_set[args.seedIndex])
AM=Arm.ArmManager(args.in_folder)
AM.loadArms()
print('[main] Finish loading arms: %d'%AM.n_arms)
#load Suparms
SAM=SupArm.SupArmManager(args.in_folder,AM)
SAM.loadArmSuparmRelation()
print('[main] Finish loading suparms')
#load User
UM=User.UserManager(args.in_folder)
UM.loadUser()
print('[main] Finishing loading users: %d'%UM.n_user)
simExperiment=simulateExp(UM.users,AM.arms, SAM.suparms,args.out_folder,args.poolSize,conf.batch_size,noise,suparm_noise,conf.test_iter, alias="time", dim=AM.dim)
algorithms={}
algorithms['Random']=None
algorithms['LinUCB']=LinUCB(AM.dim, conf.linucb_para)
algorithms['Arm-Con']=LinUCB(AM.dim, conf.linucb_para,bt=conf.bt)
algorithms['ConUCB']=Con_UCB(AM.dim, conf.conucb_para,'optimal_greedy', bt=conf.bt)
simExperiment.runAlgorithms(algorithms)