-
Notifications
You must be signed in to change notification settings - Fork 7
/
main.py
163 lines (147 loc) · 6.46 KB
/
main.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
import os
os.environ["CUDA_VISIBLE_DEVICES"] = '1'
import sys
import datasets
import model
APPROACH_LIST = ['PGD', 'IBP', 'FastLin', 'MILP', 'PercySDP', 'ZicoDual', 'CROWN', 'CROWN-IBP', 'LPAll' 'Diffai', 'RecurJac', 'FastLip']
dataset = 'mnist'
# source = 'test'
# selector = 'small.3'
source = 'fastlin'
selector = '2.20.reg'
# source = 'cnn_cert'
# selector = '3layer_fc_20'
skip = 500
norm = '2'
radii = 0.1
def pr(rad):
if dataset != 'mnist':
return f'{rad*255:.3}/255'
else:
return f'{rad:.3}'
if __name__ == '__main__':
ds = datasets.get_dataset(dataset, 'test')
print(dataset)
m = model.load_model(source, dataset, selector)
print(m)
from adaptor.basic_adaptor import PGDAdaptor, CWAdaptor
from adaptor.basic_adaptor import CleanAdaptor, FastLinIBPAdaptor, MILPAdaptor, PercySDPAdaptor, FazlybSDPAdaptor
from adaptor.lpdual_adaptor import ZicoDualAdaptor
from adaptor.crown_adaptor import FullCrownAdaptor, CrownIBPAdaptor
from adaptor.crown_adaptor import IBPAdaptor
from adaptor.recurjac_adaptor import FastLipAdaptor, RecurJacAdaptor, SpectralAdaptor
from adaptor.recurjac_adaptor import FastLinAdaptor
from adaptor.cnncert_adaptor import CNNCertAdaptor, FastLinSparseAdaptor, LPAllAdaptor
from adaptor.eran_adaptor import AI2Adaptor, DeepPolyAdaptor, RefineZonoAdaptor, KReluAdaptor
cln = CleanAdaptor(dataset, m)
pgd = PGDAdaptor(dataset, m)
cw = CWAdaptor(dataset, m)
# ibp = IBPAdaptor(dataset, m)
# fastlinibp = FastLinIBPAdaptor(dataset, m)
# milp = MILPAdaptor(dataset, m)
sdp = PercySDPAdaptor(dataset, m)
fazsdp = FazlybSDPAdaptor(dataset, m)
lpdual = ZicoDualAdaptor(dataset, m)
# fullcrown = FullCrownAdaptor(dataset, m)
# crownibp = CrownIBPAdaptor(dataset, m)
# fastlip = FastLipAdaptor(dataset, m)
# recurjac = RecurJacAdaptor(dataset, m)
# spectral = SpectralAdaptor(dataset, m)
# fastlin = FastLinAdaptor(dataset, m)
# cnncert = CNNCertAdaptor(dataset, m)
# fastlinsparse = FastLinSparseAdaptor(dataset, m)
# lpall = LPAllAdaptor(dataset, m)
ai2 = AI2Adaptor(dataset, m)
# deeppoly= DeepPolyAdaptor(dataset, m)
# refinezono = RefineZonoAdaptor(dataset, m)
# krelu = KReluAdaptorAdaptor(dataset, m)
for i in range(0, len(ds), skip):
X, y = ds[i]
cln_v = cln.verify(X, y, norm, 0.0)
# pgd_v = pgd.verify(X, y, norm, radii)
# pgd_radius = pgd.calc_radius(X, y, norm)
cw_v = cw.verify(X, y, norm, radii)
cw_radius = cw.calc_radius(X, y, norm)
# ibp_v = ibp.verify(X, y, norm, radii)
# ibp_radius = ibp.calc_radius(X, y, norm)
# fastlinibp_v = fastlinibp.verify(X, y, norm, radii)
# fastlinibp_radius = fastlinibp.calc_radius(X, y, norm)
# milp_v = milp.verify(X, y, norm, radii)
# milp_radius = milp.calc_radius(X, y, norm, eps=1e-2)
# sdp_v = sdp.verify(X, y, norm, radii)
# sdp_radius = sdp.calc_radius(X, y, norm)
# faz_v = fazsdp.verify(X, y, norm, radii)
# faz_radius = fazsdp.calc_radius(X, y, norm)
lpdual_v = lpdual.verify(X, y, norm, radii)
lpdual_radius = lpdual.calc_radius(X, y, norm)
# fullcrown_v = fullcrown.verify(X, y, norm, radii)
# fullcrown_radius = fullcrown.calc_radius(X, y, norm)
# crownibp_v = crownibp.verify(X, y, norm, radii)
# crownibp_radius = crownibp.calc_radius(X, y, norm)
# fastlip_v = fastlip.verify(X, y, norm, radii)
# fastlip_radius = fastlip.calc_radius(X, y, norm)
# recurjac_v = recurjac.verify(X, y, norm, radii)
# recurjac_radius = recurjac.calc_radius(X, y, norm)
# spectral_v = spectral.verify(X, y, norm, radii)
# spectral_radius = spectral.calc_radius(X, y, norm)
# fastlin_v = fastlin.verify(X, y, norm, radii)
# fastlin_radius = fastlin.calc_radius(X, y, norm)
# cnncert_v = cnncert.verify(X, y, norm, radii)
# cnncert_radius = cnncert.calc_radius(X, y, norm)
# fstlinsparse_v = fastlinsparse.verify(X, y, norm, radii)
# fstlinsparse_radius = fastlinsparse.calc_radius(X, y, norm)
# lpall_v = lpall.verify(X, y, norm, radii)
# lpall_radius = lpall.calc_radius(X, y, norm)
# ai2_v = ai2.verify(X, y, norm, radii)
# ai2_radius = ai2.calc_radius(X, y, norm)
# deeppoly_v = deeppoly.verify(X, y, norm, radii)
# deeppoly_radius = deeppoly.calc_radius(X, y, norm)
# refinezono_v = refinezono.verify(X, y, norm, radii)
# refinezono_radius = refinezono.calc_radius(X, y, norm)
# krelu_v = krelu.verify(X, y, norm, radii)
# krelu_radius = krelu.calc_radius(X, y, norm)
print(i, 'clean', cln_v,
# 'pgd', pgd_v,
# 'pgd_r', pr(pgd_radius),
'cw', cw_v,
'cw_r', pr(cw_radius),
# 'ibp', ibp_v,
# 'ibp_r', pr(ibp_radius),
# 'fastlinibp', fastlinibp_v,
# 'fastlinibp_r', pr(fastlinibp_radius),
# 'milp', milp_v,
# 'milp_r', pr(milp_radius),
# 'sdp', sdp_v,
# 'sdp_r', pr(sdp_radius),
# 'faz', faz_v,
# 'faz_r', pr(faz_radius),
'lpdual', lpdual_v,
'lpdual_r', pr(lpdual_radius),
# 'crown', fullcrown_v,
# 'crown_r', pr(fullcrown_radius),
# 'crownibp', crownibp_v,
# 'crownibp_r', pr(crownibp_radius),
# 'fastlip', fastlip_v,
# 'fastlip_r', pr(fastlip_radius),
# 'recurjac', recurjac_v,
# 'recurjac_r', pr(recurjac_radius),
# 'spectral', spectral_v,
# 'spectral_r', pr(spectral_radius),
# 'fastlin', fastlin_v,
# 'fastlin_r', pr(fastlin_radius),
# 'cnncert', cnncert_v,
# 'cnncert_r', pr(cnncert_radius),
# 'fstlinsparse', fstlinsparse_v,
# 'fstlinsparse_r', pr(fstlinsparse_radius),
# 'lpall', lpall_v,
# 'lpall_r', pr(lpall_radius),
# 'ai2', ai2_v,
# 'ai2_r', pr(ai2_radius),
# 'deeppoly', deeppoly_v,
# 'deeppoly_r', pr(deeppoly_radius),
# 'refinezono', refinezono_v,
# 'refinezono_r', pr(refinezono_radius),
# 'krelu', krelu_v,
# 'krelu_r', pr(krelu_radius),
file=sys.stderr)
# assert cln_v or not pgd_v