-
Notifications
You must be signed in to change notification settings - Fork 8
/
upperbound.py
293 lines (234 loc) · 13.3 KB
/
upperbound.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
from __future__ import print_function
from NeuralNetwork import *
from DataSet import *
from CompetitiveMCTS import *
from CooperativeMCTS import *
def upperbound(dataSetName, bound, tau, gameType, image_index, eta):
start_time = time.time()
MCTS_all_maximal_time = 300
MCTS_level_maximal_time = 60
NN = NeuralNetwork(dataSetName)
NN.load_network()
print("Dataset is %s." % NN.data_set)
NN.model.summary()
dataset = DataSet(dataSetName, 'testing')
image = dataset.get_input(image_index)
(label, confident) = NN.predict(image)
origClassStr = NN.get_label(int(label))
print("Working on input with index %s, whose class is '%s' and the confidence is %s."
% (image_index, origClassStr, confident))
print("the second player is %s." % gameType)
# tau = 1
# choose between "cooperative" and "competitive"
if gameType == 'cooperative':
mctsInstance = MCTSCooperative(dataSetName, NN, image_index, image, tau, eta)
mctsInstance.initialiseMoves()
start_time_all = time.time()
runningTime_all = 0
start_time_level = time.time()
runningTime_level = 0
currentBest = eta[1]
while runningTime_all <= MCTS_all_maximal_time:
'''
if runningTime_level > MCTS_level_maximal_time:
bestChild = mctsInstance.bestChild(mctsInstance.rootIndex)
# pick the current best move to take
mctsInstance.makeOneMove(bestChild)
start_time_level = time.time()
'''
# Here are three steps for MCTS
(leafNode, availableActions) = mctsInstance.treeTraversal(mctsInstance.rootIndex)
newNodes = mctsInstance.initialiseExplorationNode(leafNode, availableActions)
for node in newNodes:
(_, value) = mctsInstance.sampling(node, availableActions)
mctsInstance.backPropagation(node, value)
if currentBest > mctsInstance.bestCase[0]:
print("best distance up to now is %s" % (str(mctsInstance.bestCase[0])))
currentBest = mctsInstance.bestCase[0]
bestChild = mctsInstance.bestChild(mctsInstance.rootIndex)
# store the current best
(_, bestManipulation) = mctsInstance.bestCase
image1 = mctsInstance.applyManipulation(bestManipulation)
path0 = "%s_pic/%s_Unsafe_currentBest.png" % (dataSetName, image_index)
NN.save_input(image1, path0)
runningTime_all = time.time() - start_time_all
runningTime_level = time.time() - start_time_level
(_, bestManipulation) = mctsInstance.bestCase
print("the number of sampling: %s" % mctsInstance.numOfSampling)
print("the number of adversarial examples: %s\n" % mctsInstance.numAdv)
image1 = mctsInstance.applyManipulation(bestManipulation)
(newClass, newConfident) = NN.predict(image1)
newClassStr = NN.get_label(int(newClass))
if newClass != label:
path0 = "%s_pic/%s_%s_modified_into_%s_with_confidence_%s.png" % (
dataSetName, image_index, origClassStr, newClassStr, newConfident)
NN.save_input(image1, path0)
path0 = "%s_pic/%s_diff.png" % (dataSetName, image_index)
NN.save_input(np.absolute(image - image1), path0)
print("\nfound an adversary image within pre-specified bounded computational resource. "
"The following is its information: ")
print("difference between images: %s" % (diffImage(image, image1)))
print("number of adversarial examples found: %s" % mctsInstance.numAdv)
l2dist = l2Distance(mctsInstance.image, image1)
l1dist = l1Distance(mctsInstance.image, image1)
l0dist = l0Distance(mctsInstance.image, image1)
percent = diffPercent(mctsInstance.image, image1)
print("L2 distance %s" % l2dist)
print("L1 distance %s" % l1dist)
print("L0 distance %s" % l0dist)
print("manipulated percentage distance %s" % percent)
print("class is changed into '%s' with confidence %s\n" % (newClassStr, newConfident))
return time.time() - start_time_all, newConfident, percent, l2dist, l1dist, l0dist, 0
else:
print("\nfailed to find an adversary image within pre-specified bounded computational resource. ")
return 0, 0, 0, 0, 0, 0, 0
elif gameType == 'competitive':
mctsInstance = MCTSCompetitive(dataSetName, NN, image_index, image, tau, eta)
mctsInstance.initialiseMoves()
start_time_all = time.time()
runningTime_all = 0
currentBest = eta[1]
currentBestIndex = 0
while runningTime_all <= MCTS_all_maximal_time:
(leafNode, availableActions) = mctsInstance.treeTraversal(mctsInstance.rootIndex)
newNodes = mctsInstance.initialiseExplorationNode(leafNode, availableActions)
for node in newNodes:
(_, value) = mctsInstance.sampling(node, availableActions)
mctsInstance.backPropagation(node, value)
if currentBest > mctsInstance.bestCase[0]:
print("best distance up to now is %s" % (str(mctsInstance.bestCase[0])))
currentBest = mctsInstance.bestCase[0]
currentBestIndex += 1
# store the current best
(_, bestManipulation) = mctsInstance.bestCase
image1 = mctsInstance.applyManipulation(bestManipulation)
path0 = "%s_pic/%s_Unsafe_currentBest_%s.png" % (dataSetName, image_index, currentBestIndex)
NN.save_input(image1, path0)
runningTime_all = time.time() - start_time_all
(bestValue, bestManipulation) = mctsInstance.bestCase
print("the number of sampling: %s" % mctsInstance.numOfSampling)
print("the number of adversarial examples: %s\n" % mctsInstance.numAdv)
print("the number of max features is %s" % mctsInstance.bestFeatures()[0])
maxfeatures = mctsInstance.bestFeatures()[0]
if bestValue < eta[1]:
image1 = mctsInstance.applyManipulation(bestManipulation)
(newClass, newConfident) = NN.predict(image1)
newClassStr = NN.get_label(int(newClass))
if newClass != label:
path0 = "%s_pic/%s_%s_modified_into_%s_with_confidence_%s.png" % (
dataSetName, image_index, origClassStr, newClassStr, newConfident)
NN.save_input(image1, path0)
path0 = "%s_pic/%s_diff.png" % (dataSetName, image_index)
NN.save_input(np.absolute(image - image1), path0)
print("\nfound an adversary image within pre-specified bounded computational resource. "
"The following is its information: ")
print("difference between images: %s" % (diffImage(image, image1)))
print("number of adversarial examples found: %s" % mctsInstance.numAdv)
l2dist = l2Distance(mctsInstance.image, image1)
l1dist = l1Distance(mctsInstance.image, image1)
l0dist = l0Distance(mctsInstance.image, image1)
percent = diffPercent(mctsInstance.image, image1)
print("L2 distance %s" % l2dist)
print("L1 distance %s" % l1dist)
print("L0 distance %s" % l0dist)
print("manipulated percentage distance %s" % percent)
print("class is changed into '%s' with confidence %s\n" % (newClassStr, newConfident))
return time.time() - start_time_all, newConfident, percent, l2dist, l1dist, l0dist, maxfeatures
else:
print("\nthe robustness of the (input, model) is under control, "
"with the first player is able to defeat the second player "
"who aims to find adversarial example by "
"playing suitable strategies on selecting features. ")
return 0, 0, 0, 0, 0, 0, 0
else:
print("\nthe robustness of the (input, model) is under control, "
"with the first player is able to defeat the second player "
"who aims to find adversarial example by "
"playing suitable strategies on selecting features. ")
return 0, 0, 0, 0, 0, 0, 0
else:
print("Unrecognised game type. Try 'cooperative' or 'competitive'.")
runningTime = time.time() - start_time
'''
if gameType == 'cooperative':
mctsInstance = MCTSCooperative(dataSetName, NN, image_index, image, tau, eta)
mctsInstance.initialiseMoves()
start_time_all = time.time()
runningTime_all = 0
numberOfMoves = 0
while (not mctsInstance.terminalNode(mctsInstance.rootIndex) and
not mctsInstance.terminatedByEta(mctsInstance.rootIndex) and
runningTime_all <= MCTS_all_maximal_time):
print("the number of moves we have made up to now: %s" % numberOfMoves)
l2dist = mctsInstance.l2Dist(mctsInstance.rootIndex)
l1dist = mctsInstance.l1Dist(mctsInstance.rootIndex)
l0dist = mctsInstance.l0Dist(mctsInstance.rootIndex)
percent = mctsInstance.diffPercent(mctsInstance.rootIndex)
diffs = mctsInstance.diffImage(mctsInstance.rootIndex)
print("L2 distance %s" % l2dist)
print("L1 distance %s" % l1dist)
print("L0 distance %s" % l0dist)
print("manipulated percentage distance %s" % percent)
print("manipulated dimensions %s" % diffs)
start_time_level = time.time()
runningTime_level = 0
childTerminated = False
currentBest = eta[1]
while runningTime_level <= MCTS_level_maximal_time:
# Here are three steps for MCTS
(leafNode, availableActions) = mctsInstance.treeTraversal(mctsInstance.rootIndex)
newNodes = mctsInstance.initialiseExplorationNode(leafNode, availableActions)
for node in newNodes:
(childTerminated, value) = mctsInstance.sampling(node, availableActions)
mctsInstance.backPropagation(node, value)
runningTime_level = time.time() - start_time_level
if currentBest > mctsInstance.bestCase[0]:
print("best possible distance up to now is %s" % (str(mctsInstance.bestCase[0])))
currentBest = mctsInstance.bestCase[0]
bestChild = mctsInstance.bestChild(mctsInstance.rootIndex)
# pick the current best move to take
mctsInstance.makeOneMove(bestChild)
image1 = mctsInstance.applyManipulation(mctsInstance.manipulation[mctsInstance.rootIndex])
diffs = mctsInstance.diffImage(mctsInstance.rootIndex)
path0 = "%s_pic/%s_temp_%s.png" % (dataSetName, image_index, len(diffs))
NN.save_input(image1, path0)
(newClass, newConfident) = NN.predict(image1)
print("confidence: %s" % newConfident)
# break if we found that one of the children is a misclassification
if childTerminated is True:
break
# store the current best
(_, bestManipulation) = mctsInstance.bestCase
image1 = mctsInstance.applyManipulation(bestManipulation)
path0 = "%s_pic/%s_currentBest.png" % (dataSetName, image_index)
NN.save_input(image1, path0)
numberOfMoves += 1
runningTime_all = time.time() - start_time_all
(_, bestManipulation) = mctsInstance.bestCase
image1 = mctsInstance.applyManipulation(bestManipulation)
(newClass, newConfident) = NN.predict(image1)
newClassStr = NN.get_label(int(newClass))
if newClass != label:
path0 = "%s_pic/%s_%s_modified_into_%s_with_confidence_%s.png" % (
dataSetName, image_index, origClassStr, newClassStr, newConfident)
NN.save_input(image1, path0)
path0 = "%s_pic/%s_diff.png" % (dataSetName, image_index)
NN.save_input(np.subtract(image, image1), path0)
print("\nfound an adversary image within pre-specified bounded computational resource. "
"The following is its information: ")
print("difference between images: %s" % (diffImage(image, image1)))
print("number of adversarial examples found: %s" % mctsInstance.numAdv)
l2dist = l2Distance(mctsInstance.image, image1)
l1dist = l1Distance(mctsInstance.image, image1)
l0dist = l0Distance(mctsInstance.image, image1)
percent = diffPercent(mctsInstance.image, image1)
print("L2 distance %s" % l2dist)
print("L1 distance %s" % l1dist)
print("L0 distance %s" % l0dist)
print("manipulated percentage distance %s" % percent)
print("class is changed into '%s' with confidence %s\n" % (newClassStr, newConfident))
return time.time() - start_time_all, newConfident, percent, l2dist, l1dist, l0dist, 0
else:
print("\nfailed to find an adversary image within pre-specified bounded computational resource. ")
return 0, 0, 0, 0, 0, 0, 0
'''