-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconformal_baseline.py
171 lines (154 loc) · 6.95 KB
/
conformal_baseline.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
import argparse
from used_knowledge.used_knowledge import get_knowledge
import numpy as np
import pandas as pd
import os.path
from os import path
from dataset.dataset_conformal import GetDataset
import random
import torch
import sys
from arc.classification import ProbabilityAccumulator as ProbAccum
from dataset.dataset import DataMain
from dataset.dataset import Cifar10
from model.model import NEURAL
from model.model_single import NEURAL_single
from tqdm import tqdm
from scipy.stats.mstats import mquantiles
import arc
from conformal_attack import conformal_attack
from model.resnet import resnet56
from RSCP import Score_Functions
from RSCP import utils
def conformal_baseline(args):
print(f'Loading dataset')
if args.dataset=='GTSRB':
data = DataMain(batch_size=args.batch_size)
data.data_set_up(istrain=False)
data.greeting()
num_label, num_channel = 12, 3
print(f'Loading models')
model = NEURAL_single(n_class=num_label, n_channel=num_channel)
path = os.path.join(args.model_path, 'main/model_%.2f.pt' % (args.sigma))
model.load_state_dict(torch.load(path))
model = model.cuda()
model.eval()
elif args.dataset=='cifar10':
data = Cifar10(batch_size=args.batch_size)
data.data_set_up(istrain=False)
data.greeting()
num_label, num_channel = 10, 3
print(f'Loading models')
model = resnet56()
path = f'pretrained_models/cifar10/model_single_cifar10_{args.sigma}'
model= torch.load(path)
model = model.cuda()
model.n_class = num_label
model.eval()
else:
print(f'Dataset {args.dataset} is not implemented!')
sys.exit(1)
if args.method_conformal == 'split_conformal':
if args.calibrate and not args.inference:
y_hat = torch.zeros((len(data.y_val), num_label)).cuda()
label = data.y_val.cuda()
X, GT = data.sequential_val_batch()
cur = 0
while X is not None:
X = X.cuda()
# X = X + torch.randn_like(X).cuda() * sigma
Y = model(X)
this_batch_size = len(Y)
y_hat[cur:cur + this_batch_size, :] = Y.sigmoid()
cur = cur + this_batch_size
X, GT = data.sequential_val_batch()
for i in range(y_hat.shape[0]):
sum_ = y_hat[i].sum()
y_hat[i] /= sum_
y_hat = y_hat.detach().cpu()
label = label.cpu()
print(f'Accuracy: {(torch.argmax(y_hat,dim=1)==label).sum()/len(label)}')
# conformal calibration
n2 = y_hat.shape[0]
if args.score_type=='aps':
grey_box = ProbAccum(y_hat)
rng = np.random.default_rng(args.seed)
epsilon = rng.uniform(low=0.0, high=1.0, size=n2)
alpha_max = grey_box.calibrate_scores(label, epsilon=epsilon)
scores = args.alpha - alpha_max
level_adjusted = (1.0 - args.alpha) * (1.0 + 1.0 / float(n2))
alpha_correction = mquantiles(scores, prob=level_adjusted)
alpha_calibrated = args.alpha - alpha_correction
elif args.score_type=='hps':
scores = torch.zeros(n2)
for i in range(n2):
scores[i] = 1 - y_hat[i][label[i]]
# scores = args.alpha - scores
level_adjusted = (1.0 - args.alpha) * (1.0 + 1.0 / float(n2))
alpha_correction = mquantiles(scores, prob=level_adjusted)
alpha_calibrated = alpha_correction
print(f'alpha_calibrated: {alpha_calibrated}')
torch.save(alpha_calibrated, 'log/alpha_calibrated_model_single')
print(f'saving alpha_calibrated at log/alpha_calibrated_model_single')
elif not args.calibrate and args.inference:
alpha_calibrated = torch.load('log/alpha_calibrated_model_single')
P_test = torch.zeros((len(data.y_test), num_label)).cuda()
label_test = data.y_test
if args.attack_type=='physical_attack':
data.X_test = np.load('./data/physical_attack/stop_sign_adv_X_test.npy')
data.y_test = np.load('./data/physical_attack/stop_sign_label.npy')
data.X_test = np.array([data.pre_process_image(data.X_test[i]) for i in range(len(data.X_test))],dtype=np.float32)
data.y_test = np.array(data.y_test, dtype=np.long)
data.X_test = torch.FloatTensor(data.X_test)
data.X_test = data.X_test.permute(0, 3, 1, 2)
data.y_test = torch.LongTensor(data.y_test)
cur = 0
X, GT = data.sequential_test_batch()
while X is not None:
X = X.cuda()
# X = X + torch.randn_like(X).cuda() * sigma
GT = GT.cuda()
if args.attack_type=='pgd':
X = conformal_attack(X, GT, model, max_norm=args.max_norm)
elif args.attack_type=='smoothadv':
X = conformal_attack(X, GT, model, max_norm=args.max_norm, smoothadv=True)
Y = model(X)
this_batch_size = len(Y)
P_test[cur:cur + this_batch_size, :] = Y.sigmoid()
cur = cur + this_batch_size
X, GT = data.sequential_test_batch()
for i in range(P_test.shape[0]):
sum_ = P_test[i].sum()
P_test[i] /= sum_
P_test = P_test.detach().cpu()
if args.score_type == 'aps':
rng = np.random.default_rng(args.seed)
epsilon = rng.uniform(low=0.0, high=1.0, size=len(P_test))
grey_box_test = ProbAccum(P_test)
S_hat = grey_box_test.predict_sets(alpha_calibrated, epsilon=epsilon, allow_empty=False)
elif args.score_type == 'hps':
S_hat = []
n2 = len(P_test)
for i in range(n2):
tmp = []
for j in range(num_label):
if 1 - P_test[i][j].item() <= alpha_calibrated:
tmp.append(j)
S_hat.append(tmp)
# evaluation
total_size = 0
marginal_coverage = 0
for i, l in enumerate(S_hat):
total_size += len(l)
if label_test[i] in l:
marginal_coverage += 1
avg_size = total_size / len(S_hat)
marginal_coverage = marginal_coverage / len(S_hat)
print(f'marginal coverage with attack type {args.attack_type}: {marginal_coverage}')
print(f'avg_size with attack type {args.attack_type}: {avg_size}')
else:
print(f'args.calibrate={args.calibrate}, args.inference={args.inference} are not satisfactory!')
sys.exit(1)
else:
print(f'Conformal method {args.method_conformal} is not implemented!')
sys.exit(1)