-
Notifications
You must be signed in to change notification settings - Fork 13
/
Copy pathmix_and_match_slab.py
125 lines (106 loc) · 5.23 KB
/
mix_and_match_slab.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
from __future__ import division
from __future__ import print_function
from __future__ import absolute_import
from __future__ import unicode_literals
import os
import sys
import argparse
import json
import shutil
from collections import defaultdict
import numpy as np
import pandas as pd
from sklearn import linear_model, preprocessing, cluster, metrics, svm, model_selection
import matplotlib.pyplot as plt
import seaborn as sns
import scipy.linalg as slin
import scipy.sparse.linalg as sparselin
import scipy.sparse as sparse
import scipy.io as sio
import IPython
import data_utils as data
import datasets
import defenses
import defense_testers
import upper_bounds
from upper_bounds import hinge_loss, hinge_grad
for dataset_name in ['dogfish', 'mnist_17']:
epsilons = datasets.DATASET_EPSILONS[dataset_name]
norm_sq_constraint = datasets.DATASET_NORM_SQ_CONSTRAINTS[dataset_name]
bounds_v2_path = os.path.join(
datasets.DATA_FOLDER,
dataset_name,
'%s_slab_normc-%s_bounds_v2.npz' % (dataset_name, norm_sq_constraint))
bounds_v3_path = os.path.join(
datasets.DATA_FOLDER,
dataset_name,
'%s_slab_normc-%s_bounds_v3.npz' % (dataset_name, norm_sq_constraint))
f_v2 = np.load(bounds_v2_path)
f_v3 = np.load(bounds_v3_path)
assert np.all(epsilons == f_v2['epsilons'])
assert np.all(epsilons == f_v3['epsilons'])
assert f_v2['percentile'] == f_v3['percentile']
assert f_v2['percentile'] == 70
percentile = f_v2['percentile']
# Initialize with v3
lower_test_losses = f_v3['lower_test_losses']
lower_total_train_losses = f_v3['lower_total_train_losses']
lower_test_acc = f_v3['lower_test_acc']
lower_good_train_acc = f_v3['lower_good_train_acc']
lower_bad_train_acc = f_v3['lower_bad_train_acc']
lower_overall_train_acc = f_v3['lower_overall_train_acc']
lower_avg_good_train_losses = f_v3['lower_avg_good_train_losses']
lower_avg_bad_train_losses = f_v3['lower_avg_bad_train_losses']
lower_params_norm_sq = f_v3['lower_params_norm_sq']
lower_weight_decays = f_v3['lower_weight_decays']
upper_total_losses = f_v3['upper_total_losses']
upper_bad_losses = f_v3['upper_bad_losses']
upper_good_losses = f_v3['upper_good_losses']
upper_good_acc = f_v3['upper_good_acc']
upper_bad_acc = f_v3['upper_bad_acc']
upper_params_norm_sq = f_v3['upper_params_norm_sq']
for epsilon_idx, epsilon in enumerate(epsilons):
# Take lower upper bound
if upper_total_losses[epsilon_idx] > f_v2['upper_total_losses'][epsilon_idx]:
upper_total_losses[epsilon_idx] = f_v2['upper_total_losses'][epsilon_idx]
upper_bad_losses[epsilon_idx] = f_v2['upper_bad_losses'][epsilon_idx]
upper_good_losses[epsilon_idx] = f_v2['upper_good_losses'][epsilon_idx]
upper_good_acc[epsilon_idx] = f_v2['upper_good_acc'][epsilon_idx]
upper_bad_acc[epsilon_idx] = f_v2['upper_bad_acc'][epsilon_idx]
upper_params_norm_sq[epsilon_idx] = f_v2['upper_params_norm_sq'][epsilon_idx]
# Take higher lower bound based on lower_avg_good_train_losses
if lower_avg_good_train_losses[epsilon_idx] < f_v2['lower_avg_good_train_losses'][epsilon_idx]:
lower_test_losses[epsilon_idx] = f_v2['lower_test_losses'][epsilon_idx]
lower_total_train_losses[epsilon_idx] = f_v2['lower_total_train_losses'][epsilon_idx]
lower_test_acc[epsilon_idx] = f_v2['lower_test_acc'][epsilon_idx]
lower_good_train_acc[epsilon_idx] = f_v2['lower_good_train_acc'][epsilon_idx]
lower_bad_train_acc[epsilon_idx] = f_v2['lower_bad_train_acc'][epsilon_idx]
lower_overall_train_acc[epsilon_idx] = f_v2['lower_overall_train_acc'][epsilon_idx]
lower_avg_good_train_losses[epsilon_idx] = f_v2['lower_avg_good_train_losses'][epsilon_idx]
lower_avg_bad_train_losses[epsilon_idx] = f_v2['lower_avg_bad_train_losses'][epsilon_idx]
lower_params_norm_sq[epsilon_idx] = f_v2['lower_params_norm_sq'][epsilon_idx]
lower_weight_decays[epsilon_idx] = f_v2['lower_weight_decays'][epsilon_idx]
save_path = datasets.get_slab_bounds_path(dataset_name, norm_sq_constraint)
np.savez(
save_path,
percentile=percentile,
epsilons=epsilons,
upper_total_losses=upper_total_losses,
upper_good_losses=upper_good_losses,
upper_bad_losses=upper_bad_losses,
# upper_reg_losses=upper_reg_losses,
upper_good_acc=upper_good_acc,
upper_bad_acc=upper_bad_acc,
upper_params_norm_sq=upper_params_norm_sq,
lower_total_train_losses=lower_total_train_losses,
lower_avg_good_train_losses=lower_avg_good_train_losses,
lower_avg_bad_train_losses=lower_avg_bad_train_losses,
lower_test_losses=lower_test_losses,
# lower_reg_losses=lower_reg_losses,
lower_overall_train_acc=lower_overall_train_acc,
lower_good_train_acc=lower_good_train_acc,
lower_bad_train_acc=lower_bad_train_acc,
lower_test_acc=lower_test_acc,
lower_params_norm_sq=lower_params_norm_sq,
lower_weight_decays=lower_weight_decays
)