-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathplot.py
323 lines (262 loc) · 10 KB
/
plot.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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
import os
import numpy as np
import glob
import re
import pandas as pd
from scipy.ndimage.filters import gaussian_filter1d
import seaborn as sns
import matplotlib
import matplotlib.pyplot as plt
from matplotlib.lines import Line2D
from utils import get_interval
def atoi(text):
return int(text) if text.isdigit() else text
def natural_keys(text):
'''
alist.sort(key=natural_keys) sorts in human order
http://nedbatchelder.com/blog/200712/human_sorting.html
(See Toothy's implementation in the comments)
'''
return [ atoi(c) for c in re.split(r'(\d+)', text) ]
def plot_w(indir='w_save/mnist'):
font = {'size' : 22}
matplotlib.rc('font', **font)
files = glob.glob(os.path.join(indir, "w_*.npy"))
files = sorted(files, key=natural_keys)
ws = []
for filename in files:
w = np.load(filename)
ws.append(w)
# print (w)
print (np.mean(w, axis=0))
ws = np.stack(ws, axis=0)
ticks = [str(i) for i in range(0, ws.shape[0], 5)]
def set_box_color(bp, color):
plt.setp(bp['boxes'], color=color)
plt.setp(bp['whiskers'], color=color)
plt.setp(bp['caps'], color=color)
plt.setp(bp['medians'], color=color)
plt.figure(figsize=(16, 8))
medianprops = dict(linestyle='-', linewidth=2.5, color='red')
colors = ['r', 'g', 'b']
arch = list('ABC')
for i in range(w.shape[-1]):
data = ws[:,:,i]
positions = np.array(range(len(data)))*3.0 + 0.7*(i-1)
data = np.transpose(data)
# print (data)
# print (data.shape, len(data))
# print (positions, len(positions))
bp = plt.boxplot(data, positions=positions, sym='', patch_artist=True, medianprops=medianprops)
# set_box_color(bp, color[i]) # colors are from http://colorbrewer2.org/
plt.plot([], c=colors[i], label='Model ' + arch[i])
for patch in bp['boxes']:
patch.set_facecolor(colors[i])
r, g, b, a = patch.get_facecolor()
patch.set_facecolor((r, g, b, .3))
plt.legend(loc='upper left')
plt.xticks(range(0, len(ticks) * 3 * 5, 3 * 5), ticks)
plt.xlim(-3, len(ticks)*3 * 5)
plt.ylim(0, 1)
plt.xlabel('Number of iterations')
plt.ylabel('$w_i$')
plt.tight_layout()
# plt.show()
plt.savefig('w_dist.pdf')
def plot_loss(indir='logs_mnist_avg'):
sns.set(font_scale=1.4)
# sns.set_context("talk")
# sns.set_style("whitegrid")
files = glob.glob(os.path.join(indir, "ens_model*.out"))
results = dict()
for filename in files:
log = open(filename, 'r').readlines()
epochs = int(log[0].split(', ')[6].split("=")[-1])
result = []
for line in log[5:8]:
adv_loss, adv_acc = line.split("\t")
adv_loss = float(adv_loss.split(": ")[-1])
adv_acc = float(adv_acc.split(": ")[-1])
result += [adv_loss, adv_acc]
result += [float(log[-1].split(": ")[-1])]
results[epochs] = result
df = pd.DataFrame.from_dict(results, orient='index')
df.columns = ["loss1", "acc1", "loss2", "acc2", "loss3", "acc3", "foolrate"]
fig, ax1 = plt.subplots(figsize=(9,6))
iters = np.arange(0, 40)
ax1.plot(iters, df['loss1'], 'b^-.', alpha=0.8)
ax1.plot(iters, df['loss2'], 'bo:', alpha=0.8)
ax1.plot(iters, df['loss3'], 'bs--', alpha=0.8)
ax1.set_xlabel('Number of iterations')
# Make the y-axis label, ticks and tick labels match the line color.
ax1.set_ylabel('loss', color='b')
ax1.tick_params('y', colors='b')
ax2 = ax1.twinx()
ax2.plot(iters, df['acc1'], 'm^-.', alpha=0.8)
ax2.plot(iters, df['acc2'], 'mo:', alpha=0.8)
ax2.plot(iters, df['acc3'], 'ms--', alpha=0.8)
legend_elements = [Line2D([0], [0], marker='^', linestyle='-.', lw=2, c='r', label='Model A'),
Line2D([0], [0], marker='o', linestyle=':', lw=2, c='r', label='Model B'),
Line2D([0], [0], marker='s', linestyle='--', lw=2, c='r', label='Model C'),
Line2D([0], [0], marker='x', linestyle='-', lw=2, c='r', label='success rate')]
l7 = ax2.plot(iters, df['foolrate'], 'rx-', linewidth=2.0)
plt.legend(handles=legend_elements, loc='lower right')
ax2.set_ylabel('test accuracy', color='m')
ax2.tick_params('y', colors='m')
ax2.grid(False)
fig.tight_layout()
# plt.show()
plt.savefig('loss_acc_avg.pdf')
def plot_eps(indir="logs_mnist_eps"):
sns.set(font_scale=1.4)
files = glob.glob(os.path.join(indir, "ens_model*.out"))
avg_asr = []
avg_eps = []
mm_asr = []
mm_eps = []
for filename in files:
log = open(filename, 'r').readlines()
avg = (log[0].split(', ')[2].split("=")[-1])
eps = (log[0].split(', ')[7].split("=")[-1])
foolrate = float(log[-1].split(": ")[-1])
if avg == 'True':
avg_asr.append(foolrate)
avg_eps.append(eps)
else:
mm_asr.append(foolrate)
mm_eps.append(eps)
mm_asr = [x for _, x in sorted(zip(mm_eps, mm_asr))]
avg_asr = [x for _, x in sorted(zip(avg_eps, avg_asr))]
mm_eps = sorted(mm_eps)
avg_eps = sorted(avg_eps)
plt.plot(avg_eps, avg_asr, 'o-', label='average case', linewidth=2.0)
plt.plot(mm_eps, mm_asr, '^-', label='minmax optimization', linewidth=2.0)
plt.legend()
plt.xlabel('$\\epsilon$')
plt.ylabel('Attack success rate')
plt.tight_layout()
# plt.show()
plt.savefig('compare_mnist.pdf')
# draw box plot between weight and eps
def plot_w_norm(indir='save_mnist_w'):
font = {'size' : 22}
matplotlib.rc('font', **font)
files = glob.glob(os.path.join(indir, "w_*.npy"))
files = sorted(files, key=natural_keys)
ws = []
for filename in files:
w = np.load(filename)
ws.append(w)
ws = np.stack(ws, axis=0)
ticks = [str(i) for i in range(0, ws.shape[0], 50)]
def set_box_color(bp, color):
plt.setp(bp['boxes'], color=color)
plt.setp(bp['whiskers'], color=color)
plt.setp(bp['caps'], color=color)
plt.setp(bp['medians'], color=color)
plt.figure(figsize=(16, 8))
medianprops = dict(linestyle='-', linewidth=1.0, color='red')
colors = ['r', 'g', 'b']
arch = ["$\\ell_\\infty$","$\\ell_2$", "$\\ell_1$"]
for i in range(ws.shape[-1]):
data = ws[:,:,i]
positions = np.array(range(len(data)))*3.0 + 0.7*(i-1)
data = np.transpose(data)
bp = plt.boxplot(data, positions=positions, widths=0.5, sym='', patch_artist=True, medianprops=medianprops)
plt.plot([], c=colors[i], label=arch[i])
for patch in bp['boxes']:
patch.set_facecolor(colors[i])
r, g, b, a = patch.get_facecolor()
patch.set_facecolor((r, g, b, .3))
plt.legend(loc='upper left')
plt.xticks(range(0, len(ticks) * 3 * 5, 3 * 5), ticks)
plt.xlim(-3, len(ticks)*3 * 5)
plt.ylim(-0.05, 1.05)
plt.xlabel('Number of iterations')
plt.ylabel('$w_i$')
plt.tight_layout()
# plt.show()
plt.savefig('w_dist.pdf')
# draw box plot between weights and eps
def plot_w_eps(indir='save_mnist_w'):
font = {'size': 26}
matplotlib.rc('font', **font)
files = glob.glob(os.path.join(indir, "w_349.npy"))
files = sorted(files, key=natural_keys)[1:8]
ws = []
eps = []
for filename in files:
w = np.load(filename)
e = re.findall(r'\d+\.\d+', filename)[1]
ws.append(w)
eps.append(e)
ws = np.stack(ws, axis=0)
def set_box_color(bp, color):
plt.setp(bp['boxes'], color=color)
plt.setp(bp['whiskers'], color=color)
plt.setp(bp['caps'], color=color)
plt.setp(bp['medians'], color=color)
plt.figure(figsize=(8, 6))
medianprops = dict(linestyle='-', linewidth=0.5, color='red')
colors = ['blue', 'orange', 'b']
arch = ["$\\ell_\\infty$","$\\ell_2$"]
for i in range(ws.shape[-1]):
data = ws[:,:,i]
positions = np.array(range(len(data)))*15.0
data = np.transpose(data)
bp = plt.violinplot(data, positions=positions, widths=3.5, showmeans=True)
plt.plot([], c=colors[i], label=arch[i])
for pc in bp['bodies']:
pc.set_edgecolor('black')
pc.set_linewidth(1)
pc.set_alpha(0.5)
plt.legend(loc='upper right')
plt.xticks(range(0, len(eps) * 3 * 5, 3 * 5), eps)
plt.xlim(-15, len(eps)*3 * 5)
plt.ylim(-0.05, 1.05)
plt.xlabel('$\epsilon (\ell_2)$')
plt.ylabel('$w_i$')
plt.tight_layout()
# plt.show()
plt.savefig('w_dist.pdf')
# draw box plots on data transformation between weight and epoch
def plot_trans_w(indir='save_mnist_w'):
font = {'size' : 22}
matplotlib.rc('font', **font)
ws = []
files = glob.glob(os.path.join(indir, "w_*.npy"))
files = sorted(files, key=natural_keys)
arch = ["fliplr", "adjust_brightness", "rotate_random", "useless"]
for filename in files:
w = np.load(filename)
ws.append(w)
ws = np.stack(ws, axis=0)
ticks = [str(i) for i in range(0, ws.shape[0], 50)]
def set_box_color(bp, color):
plt.setp(bp['boxes'], color=color)
plt.setp(bp['whiskers'], color=color)
plt.setp(bp['caps'], color=color)
plt.setp(bp['medians'], color=color)
plt.figure(figsize=(16, 8))
#medianprops = dict(linestyle='-', linewidth=1, color='red')
colors = ['r', 'g', 'b', 'y']
for i in range(w.shape[-1]):
data = ws[:,:,i]
positions = np.array(range(len(data)))*3.0 + 0.7*(i-1)
data = np.transpose(data)
bp = plt.boxplot(data, positions=positions, sym='', patch_artist=True)
plt.plot([], c=colors[i], label=arch[i])
plt.legend(loc='upper left')
plt.xticks(range(0, len(ticks) * 3 * 5, 3 * 5), ticks)
plt.xlim(-3, len(ticks)*3 * 5)
plt.ylim(-0.1, 1.1)
plt.xlabel('epochs')
plt.ylabel('$w_i$')
plt.tight_layout()
# plt.show()
plt.savefig('w_dist.pdf')
if __name__ == "__main__":
# plot_w_norm()
# plot_trans_w()
# plot_loss()
# plot_eps_new()