-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathRandomProblems_BAI_algs_comparison.py
177 lines (134 loc) · 6.27 KB
/
RandomProblems_BAI_algs_comparison.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
import os
# Set working directory same as file
abspath = os.path.abspath(__file__)
dname = os.path.dirname(abspath)
os.chdir(dname)
import sys
sys.path.append(dname+'/Algorithms')
sys.path.append(dname+'/Environments')
import numpy as np
import matplotlib.pyplot as plt
import pickle5 as pickle
from SEQUCBE import SEQUCBE_run, SEQUCBE_NoLim_run, SEQUCBE
from UCBE import UCBE_run, UCBE
from SR import SR_run
# Random problem generation details
expName = 'Audibertea2010'
saveData = False
# Parameter c value
c = 1
Rp = 7
R_Rp = [[0.5] for _ in range(Rp)]
R_Rp[0] = R_Rp[0] + [0.4]*19
R_Rp[1] = R_Rp[1] + [0.42]*5 + [0.38]*14
R_Rp[2] = R_Rp[2] + [0.5-(0.37)**i for i in [2,3,4]]
R_Rp[3] = R_Rp[3] + [0.42,0.4,0.4,0.35,0.35]
R_Rp[4] = R_Rp[4] + [0.5-0.025*i for i in range(2,16)]
R_Rp[5] = R_Rp[5] + [0.48] + [0.37]*18
R_Rp[6] = R_Rp[6] + [0.45]*5 + [0.43]*14 + [0.38]*10
tau_Rp = [2000, 2000, 2000, 600, 4000, 6000, 6000]
# Number of tries for every random problem to estimate error percentage
Tries = 100
# Number of iterations to estimate error percentage standard deviation
Reps = 10
# Number of algorithms to be compared
Algs = 4
AlgsNames=['SEQ(UCBE)-LR','SEQ(UCBE)-LP','UCBE','SR+']
errorPercSEQUCBE_LR = [[None]*Reps for _ in range(Rp)]
errorPercSEQUCBE_LP = [[None]*Reps for _ in range(Rp)]
errorPercUCBE = [[None]*Reps for _ in range(Rp)]
errorPercSR = [[None]*Reps for _ in range(Rp)]
for kk in range(Reps):
# Create array to store results
bestArmSEQUCBE_LR = [[0]*Tries for _ in range(Rp)]
bestArmSEQUCBE_LP = [[0]*Tries for _ in range(Rp)]
bestArmUCBE = [[0]*Tries for _ in range(Rp)]
bestArmSR = [[0]*Tries for _ in range(Rp)]
NPullsSEQUCBE_LR = [[0]*Tries for _ in range(Rp)]
NPullsSEQUCBE_LP = [[0]*Tries for _ in range(Rp)]
NPullsUCBE = [[0]*Tries for _ in range(Rp)]
NPullsSR = [[0]*Tries for _ in range(Rp)]
indSEQUCBE_LR = [[0]*Tries for _ in range(Rp)]
indSEQUCBE_LP = [[0]*Tries for _ in range(Rp)]
indUCBE = [[0]*Tries for _ in range(Rp)]
indSR = [[0]*Tries for _ in range(Rp)]
finalTimeSEQUCBE_LP = [[0]*Tries for _ in range(Rp)]
for ii in range(Rp):
# Create random problem
#R = rng.random(size=n_arms)
R = np.array(R_Rp[ii])
tau = tau_Rp[ii]
n_arms = len(R)
for tr in range(Tries):
print('Rep: ', kk, ', R_p: ',ii,', try: ', tr)
# SEQ-UCBE LR
bestArmSEQUCBE_LR[ii][tr], NPullsSEQUCBE_LR[ii][tr], indSEQUCBE_LR[ii][tr] = SEQUCBE_NoLim_run(tau,n_arms,R,c)
# SEQ-UCBE LP
bestArmSEQUCBE_LP[ii][tr], NPullsSEQUCBE_LP[ii][tr], indSEQUCBE_LP[ii][tr], finalTimeSEQUCBE_LP[ii][tr] = SEQUCBE_run(tau,n_arms,R,c,tau)
# UCBE
bestArmUCBE[ii][tr], NPullsUCBE[ii][tr], indUCBE[ii][tr] = UCBE_run(tau,n_arms,R,c)
# SR
bestArmSR[ii][tr], NPullsSR[ii][tr], indSR[ii][tr] = SR_run(tau,n_arms,R)
##
bestArmTotal = []
bestArmTotal.append(bestArmSEQUCBE_LR)
bestArmTotal.append(bestArmSEQUCBE_LP)
bestArmTotal.append(bestArmUCBE)
bestArmTotal.append(bestArmSR)
NPullsTotal = []
NPullsTotal.append(NPullsSEQUCBE_LR)
NPullsTotal.append(NPullsSEQUCBE_LP)
NPullsTotal.append(NPullsUCBE)
NPullsTotal.append(NPullsSR)
indTotal = []
indTotal.append(indSEQUCBE_LR)
indTotal.append(indSEQUCBE_LP)
indTotal.append(indUCBE)
indTotal.append(indSR)
for r in range(Rp):
R = R_Rp[r]
arm_star = np.argmax(R)
# errorPercSEQUCBE_Rp = []
# errorPercUCBE_Rp = []
# errorPercSR_Rp = []
correctSEQUCBE_LR = [i == arm_star for i in bestArmSEQUCBE_LR[r]]
errorPercSEQUCBE_LR[r][kk] = (1 - sum(correctSEQUCBE_LR)/len(correctSEQUCBE_LR))
correctSEQUCBE_LP = [i == arm_star for i in bestArmSEQUCBE_LP[r]]
errorPercSEQUCBE_LP[r][kk] = (1 - sum(correctSEQUCBE_LP)/len(correctSEQUCBE_LP))
correctUCBE = [i == arm_star for i in bestArmUCBE[r]]
errorPercUCBE[r][kk] = (1 - sum(correctUCBE)/len(correctUCBE))
correctSR = [i == arm_star for i in bestArmSR[r]]
errorPercSR[r][kk] = (1 - sum(correctSR)/len(correctSR))
##
errorPercSEQUCBE_LR_Avg = np.mean(errorPercSEQUCBE_LR,1)
errorPercSEQUCBE_LR_Sd = np.std(errorPercSEQUCBE_LR,1)*1.96/np.sqrt(Reps)
errorPercSEQUCBE_LP_Avg = np.mean(errorPercSEQUCBE_LP,1)
errorPercSEQUCBE_LP_Sd = np.std(errorPercSEQUCBE_LP,1)*1.96/np.sqrt(Reps)
errorPercUCBE_Avg = np.mean(errorPercUCBE,1)
errorPercUCBE_Sd = np.std(errorPercUCBE,1)*1.96/np.sqrt(Reps)
errorPercSR_Avg = np.mean(errorPercSR,1)
errorPercSR_Sd = np.std(errorPercSR,1)*1.96/np.sqrt(Reps)
x = np.arange(Rp)
width = 0.2
fig, axs = plt.subplots(1, 1, figsize=(9, 6))
axs.bar(x - 1.5*width,errorPercSEQUCBE_LR_Avg,width,label=AlgsNames[0])
axs.errorbar(x - 1.5*width,errorPercSEQUCBE_LR_Avg,yerr=errorPercSEQUCBE_LR_Sd,c="black",capsize=5,fmt='none')
axs.bar(x - 0.5*width,errorPercSEQUCBE_LP_Avg,width,label=AlgsNames[1])
axs.errorbar(x - 0.5*width,errorPercSEQUCBE_LP_Avg,yerr=errorPercSEQUCBE_LP_Sd,c="black",capsize=5,fmt='none')
axs.bar(x + 0.5*width,errorPercUCBE_Avg,width,label=AlgsNames[2])
axs.errorbar(x + 0.5*width,errorPercUCBE_Avg,yerr=errorPercUCBE_Sd,c="black",capsize=5,fmt='none')
axs.bar(x + 1.5*width,errorPercSR_Avg,width,label=AlgsNames[3])
axs.errorbar(x + 1.5*width,errorPercSR_Avg,yerr=errorPercSR_Sd,c="black",capsize=5,fmt='none')
axs.set(ylabel='Probability of error [-]',xlabel='Experiment')
axs.legend()
if saveData:
plt.savefig('ErrorProb_BAI_comparison_'+expName+'.png', dpi=600,bbox_inches = 'tight')
if saveData:
with open('bestArmTotal_BAI_'+expName+'_rep'+str(kk)+'.pickle', 'wb') as f:
pickle.dump(bestArmTotal, f, pickle.HIGHEST_PROTOCOL)
with open('NPullsTotal_BAI_'+expName+'_rep'+str(kk)+'.pickle', 'wb') as f:
pickle.dump(NPullsTotal, f, pickle.HIGHEST_PROTOCOL)
with open('indTotal_BAI_'+expName+'_rep'+str(kk)+'.pickle', 'wb') as f:
pickle.dump(indTotal, f, pickle.HIGHEST_PROTOCOL)
with open('finalTimeSEQUCBE_LP_BAI_'+expName+'_rep'+str(kk)+'.pickle', 'wb') as f:
pickle.dump(finalTimeSEQUCBE_LP, f, pickle.HIGHEST_PROTOCOL)