forked from gondimribeiro/adv-attacks-vae
-
Notifications
You must be signed in to change notification settings - Fork 0
/
normalize_results.py
142 lines (116 loc) · 4.08 KB
/
normalize_results.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
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from mpl_toolkits.mplot3d import Axes3D
import seaborn as sns
sns.set()
def normalize_data(dir, n, title=''):
orig_dist = []
orig_dist_std = []
adv_dist = []
adv_dist_std = []
noise_dist = []
recon_dist = []
orig_target_dist = []
target_recon_dist = []
target_recon_dist_std = []
orig_target_recon_dist = []
orig_target_recon_dist_std = []
C = []
for i in range(n):
df = pd.read_csv(dir + "/results/exp_" + str(i) + ".csv")
orig_dist.append(df['orig_dist'].values)
orig_dist_std.append(df['orig_dist_std'].values)
adv_dist.append(df['adv_dist'].values)
adv_dist_std.append(df['adv_dist_std'].values)
noise_dist.append(df['noise_dist'].values)
recon_dist.append(df['recon_dist'].values)
target_recon_dist.append(df['target_recon_dist'].values)
target_recon_dist_std.append(df['target_recon_dist'].values)
orig_target_dist.append(df['orig_target_dist'].values)
orig_target_recon_dist.append(df['orig_target_recon_dist'].values)
orig_target_recon_dist_std.append(
df['orig_target_recon_dist_std'].values)
C.append(df['C'].values)
normalized_data = []
for i in range(n):
xs = noise_dist[i] / orig_target_dist[i]
zs = (
(adv_dist[i] - target_recon_dist[i]) /
(orig_target_recon_dist[i] - target_recon_dist[i])
)
zsds = (
np.sqrt(adv_dist_std[i]**2.0 + target_recon_dist_std[i]**2.0) /
np.sqrt(orig_target_recon_dist_std[i]
** 2.0 + target_recon_dist_std[i]**2.)
)
idx = np.argsort(xs)
xs = xs[idx]
zs = zs[idx]
zsds = zsds[idx]
(xs, zs, zsds) = zip(*[(x, z, zd) for (x, z, zd)
in zip(xs, zs, zsds) if x >= 0 and x <= 1])
normalized_data.append(dict())
normalized_data[i]['diff_xs'] = xs
normalized_data[i]['diff_zs'] = zs
normalized_data[i]['diff_zsds'] = zsds
ys = noise_dist[i] / orig_target_dist[i]
zs = recon_dist[i] / orig_target_recon_dist[i]
idx = np.argsort(ys)
ys = ys[idx]
zs = zs[idx]
(ys, zs) = zip(*[(y, z)
for (y, z) in zip(ys, zs) if y >= 0 and y <= 1])
normalized_data[i]['close_ys'] = ys
normalized_data[i]['close_zs'] = zs
np.save(dir + "/results/normalized_data", normalized_data)
plot_data(normalized_data, title, dir)
def plot_data(data, title='', dir=None, error_bars=False):
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
for i, d in enumerate(data):
xs = d['diff_xs']
zs = d['diff_zs']
zsds = d['diff_zsds']
npoints = len(xs)
ax.plot(xs, [i] * npoints, zs, alpha=0.5)
if error_bars:
for j in range(len(xs)):
x = np.array([xs[j], xs[j]])
y = np.array([i, i])
z = np.array([zs[j], zs[j]])
zerror = np.array([-zsds[j], zsds[j]])
ax.plot(x, y, z + zerror, marker="_", alpha=0.25)
ax.set_xlabel('Distortion')
ax.set_xlim3d(0, 1)
ax.set_ylabel('Experiment')
ax.set_zlabel('Adversarial rec. - Target')
ax.set_zlim3d(0, 1)
fig.set_figwidth(8)
fig.set_figheight(6)
plt.title(title)
if dir is None:
plt.show()
else:
plt.savefig(dir + '/results/diff_plot.png')
plt.close()
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
for i, d in enumerate(data):
ys = d['close_ys']
zs = d['close_zs']
npoints = len(ys)
ax.plot([i] * npoints, ys, zs, alpha=0.5)
ax.set_ylabel('Distortion')
ax.set_ylim3d(0, 1)
ax.set_xlabel('Experiment')
ax.set_zlabel('Adversarial rec. - Adversarial input')
ax.set_zlim3d(0, 1)
fig.set_figwidth(9)
fig.set_figheight(6)
plt.title(title)
if dir is None:
plt.show()
else:
plt.savefig(dir + '/results/close_plot.png')
plt.close()