-
Notifications
You must be signed in to change notification settings - Fork 1
/
mlp_attack_model.py
253 lines (223 loc) · 7.81 KB
/
mlp_attack_model.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
import torch
import numpy as np
from torch import nn
from torch.autograd import Variable
import torch.nn.functional as F
from torch.utils.data import DataLoader
from sklearn.model_selection import train_test_split
topk = 100
num_latent = 100
class MLP(nn.Module):
def __init__(self):
super(MLP, self).__init__()
self.fc1 = nn.Linear(num_latent, 32)
self.fc2 = nn.Linear(32, 8)
self.fc3 = nn.Linear(8, 2)
def forward(self, x):
x = x.view(-1, num_latent)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = F.softmax(self.fc3(x))
return x
num_target # number of users in target dataset
num_shadow # number of users in shadow dataset
num_vector_shadow # number of items in shadow dataset
num_vector_target # nubmer of items in target dataset
f_shadow = open("file name of shadow interactions", 'r') # interactions for shadow
f_target = open("file name of target interactions", 'r') # interactions for target
fr_shadow = open("file name of shadow recommendations", 'r') # recommendations for shadow
fr_target = open("file name of target recommendations", 'r') # recommendations for target
fr_vector_shadow = open("file name of shadow vectorizations", 'r') # vector for shadow items
fr_vector_target = open("file name of target vectorizations", 'r') # vector for target items
interaction_target = [] # interactions for target
interaction_shadow = [] # interactions for shadow
recommend_target = [] # recommendations for target
recommend_shadow = [] # recommendations for shadow
vector_target = [] # vectors for target
vector_shadow = [] # vectors for shadow
vectors1 = [] # vectors for shadow items
vectors2 = [] # vectors for target items
label_target = []
label_shadow = []
# vectors for shadow items
for i in range(num_vector_shadow):
vectors1.append([])
line = fr_vector_shadow.readline()
arr = line.split('\t')
for j in range(num_latent):
arr[j] = float(arr[j])
vectors1[i].append(arr[j])
vectors1[i] = torch.tensor(vectors1[i])
# vectors for target items
for i in range(num_vector_target):
vectors2.append([])
line = fr_vector_target.readline()
arr = line.split('\t')
for j in range(num_latent):
arr[j] = float(arr[j])
vectors2[i].append(arr[j])
vectors2[i] = torch.tensor(vectors2[i])
# init for target
for i in range(num_target):
recommend_target.append([])
interaction_target.append([])
# read recommendations
line = fr_target.readline()
while line != '' and line is not None and line != ' ':
arr = line.split('\t')
recommend_target[int(arr[0])].append(int(arr[1]))
line = fr_target.readline()
# read interactions
line = f_target.readline()
while line != '' and line is not None and line != ' ':
arr = line.split('\t')
interaction_target[int(arr[0])].append(int(arr[1]))
line = f_target.readline()
# init for shadow
for i in range(num_shadow):
recommend_shadow.append([])
interaction_shadow.append([])
# read recommendations
line = fr_shadow.readline()
while line != '' and line is not None and line != ' ':
arr = line.split('\t')
recommend_shadow[int(arr[0])].append(int(arr[1]))
line = fr_shadow.readline()
# read interactions
line = f_shadow.readline()
while line != '' and line is not None and line != ' ':
arr = line.split('\t')
interaction_shadow[int(arr[0])].append(int(arr[1]))
line = f_shadow.readline()
# vectorization for shadow
member_bound = 0
random_bound = 0
for i in range(num_shadow):
if i < num_shadow//2:
# member
label_shadow.append([1])
else:
# random
label_shadow.append([0])
temp_vector = torch.zeros(num_latent)
# the center of the ineractions
len_shadow = len(interaction_shadow[i])
for j in range(len_shadow):
temp_vector = temp_vector + vectors1[interaction_shadow[i][j]]
temp_vector = temp_vector / len_shadow
temp_vector = temp_vector.numpy().tolist()
# subtracted
vector_shadow.append([])
vector_shadow[i].append(temp_vector) # list
temp_vector = torch.zeros(num_latent)
#the center of the recommendations
sum_temp = 0
for j in range(topk):
sum_temp += np.exp(j+1)
for j in range(topk):
temp_vector = temp_vector + ((topk-j)/5050)*vectors1[recommend_shadow[i][j]]
temp_vector = temp_vector.numpy().tolist()
for j in range(num_latent):
vector_shadow[i][0][j] = (vector_shadow[i][0][j] - temp_vector[j])
# vectorization for target
for i in range(num_target):
if i < num_target // 2:
# member
label_target.append([1])
else:
# random
label_target.append([0])
temp_vector = torch.zeros(num_latent)
# the center of the ineractions
len_target = len(interaction_target[i])
for j in range(len_target):
temp_vector = temp_vector + vectors2[interaction_target[i][j]]
temp_vector = temp_vector / len_target
temp_vector = temp_vector.numpy().tolist()
# subtracted
vector_target.append([])
vector_target[i].append(temp_vector) # lsit
temp_vector = torch.zeros(num_latent)
#the center of the recommendations
sum_temp = 0
for j in range(topk):
sum_temp += np.exp(j+1)
for j in range(topk):
temp_vector = temp_vector + ((topk-j)/5050)*vectors2[recommend_target[i][j]]
temp_vector = temp_vector.numpy().tolist()
for j in range(num_latent):
vector_target[i][0][j] = (vector_target[i][0][j] - temp_vector[j])
for i in range(num_shadow):
vector_shadow[i] = torch.Tensor(np.array(vector_shadow[i])) # train
label_shadow[i] = torch.Tensor(np.array(label_shadow[i])).long() # train
for i in range(num_target):
vector_target[i] = torch.Tensor(np.array(vector_target[i])) # test
label_target[i] = torch.Tensor(np.array(label_target[i])).long() # test
mlp = MLP()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(mlp.parameters(), lr=0.01, momentum=0.9)
# accuarcy
def AccuarcyCompute(pred, label):
pred = pred.cpu().data.numpy()
label = label.cpu().data.numpy()
test_np = (np.argmax(pred, 1) == label)
test_np = np.float32(test_np)
return np.mean(test_np)
losses = []
acces = []
eval_losses = []
eval_acces = []
# SGD: sample-update-sample-update-...
for x in range(20):
train_loss = 0
train_acc = 0
for i in range(num_shadow):
optimizer.zero_grad()
inputs = torch.autograd.Variable(vector_shadow[i])
labels = torch.autograd.Variable(label_shadow[i])
outputs = mlp(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
train_loss += loss.item()
_, pred = outputs.max(1)
if i < num_shadow//2:
# 1
if int(pred)==1:
train_acc += 1
else:
# 0
if int(pred)==0:
train_acc += 1
losses.append(train_loss / num_shadow)
acces.append(train_acc / num_shadow)
print('epoch: {}, Train Loss: {:.6f}, Train Acc: {:.6f}'.format(x, train_loss / (num_shadow), train_acc / (num_shadow)))
acc_ans = 0
TruePositive = 0
FalsePositive = 0
TrueNegative = 0
FalseNegative = 0
for i in range(num_target):
inputs = torch.autograd.Variable(vector_target[i])
labels = torch.autograd.Variable(label_target[i])
outputs = mlp(inputs)
_, pred = outputs.max(1)
if i < num_target//2:
# 1
if int(pred)==1:
acc_ans += 1
TruePositive = TruePositive + 1
else:
FalseNegative = FalseNegative + 1
else:
# 0
if int(pred)==0:
acc_ans += 1
TrueNegative = TrueNegative + 1
else:
FalsePositive = FalsePositive + 1
TPRate = TruePositive / (TruePositive+FalseNegative)
FPRate = FalsePositive / (FalsePositive+TrueNegative)
area = 0.5*TPRate*FPRate+0.5*(TPRate+1)*(1-FPRate)
print("AUC: ")
print(area)