-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_attackDefense.py
158 lines (156 loc) · 6.43 KB
/
model_attackDefense.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
import pickle
import numpy as np
import pandas as pd
import scipy.sparse as sp
import pdb
from deeprobust.graph.global_attack import NodeEmbeddingAttack
from deeprobust.graph.defense import DeepWalk
from deeprobust.graph.data import AmazonPyg
from deeprobust.graph.data import Dataset, Dpr2Pyg, Pyg2Dpr
from torch import optim
from torch.nn import functional as F
from torch.nn.parameter import Parameter
from tqdm import tqdm
from deeprobust.graph import utils
from deeprobust.graph.global_attack import BaseAttack
np.set_printoptions(threshold=np.inf)
pd.set_option('display.max_rows', None)
from deeprobust.graph.defense import GCN
from deeprobust.graph.global_attack import Metattack
from deeprobust.graph.global_attack import Random
from deeprobust.graph.global_attack import PGDAttack
from deeprobust.graph.global_attack import DICE
from deeprobust.graph.utils import preprocess
def Meta():
with open('dprmodel.pkl', 'rb') as f:
data = pickle.load(f)
adj, features, labels = data.adj, data.features, data.labels
idx_train, idx_val, idx_test = data.idx_train, data.idx_val, data.idx_test
idx_unlabeled = np.union1d(idx_val, idx_test)
labels = labels.astype(np.int32)
surrogate = GCN(nfeat=features.shape[1], nclass=labels.max().item() + 1,
nhid=16, dropout=0, with_relu=False, with_bias=False, device='cpu').to('cpu')
surrogate.fit(features, adj, labels, idx_train, idx_val, patience=30)
model = Metattack(surrogate, nnodes=adj.shape[0], feature_shape=features.shape,
attack_structure=True, attack_features=False, device='cpu', lambda_=0).to('cpu')
with open('Metattack.pkl', 'wb') as f:
pickle.dump(model, f)
# with open('Metattack.pkl', 'rb') as f:
# model = pickle.load(f)
model.attack(features, adj, labels, idx_train, idx_unlabeled, n_perturbations=10, ll_constraint=False)
modified_adj = model.modified_adj
sp.save_npz('testmMeta.npz', modified_adj)
def Embed():
with open('dprmodel.pkl', 'rb') as f:
data = pickle.load(f)
print(type(data.adj))
return
adj, features, labels = data.adj, data.features, data.labels
model = NodeEmbeddingAttack()
# with open('NodeEmbeddingAttack.pkl', 'wb') as f:
# pickle.dump(model,f)
with open('NodeEmbeddingAttack.pkl', 'rb') as f:
model = pickle.load(f)
model.attack(adj, attack_type="remove") # 只对adj进行攻击，没有使用features和labels其他的没有变
modified_adj = model.modified_adj
sp.save_npz('testmEmbed.npz', modified_adj)
def RandomAttack():
with open('dprmodel.pkl', 'rb') as f:
data = pickle.load(f)
adj, features, labels = data.adj, data.features, data.labels
model = Random()
# with open('RandomAttack.pkl', 'wb') as f:
# pickle.dump(model,f)
with open('RandomAttack.pkl', 'rb') as f:
model = pickle.load(f)
model.attack(adj, n_perturbations=10)
modified_adj = sp.csr_matrix(model.modified_adj)
print(modified_adj)
print(type(modified_adj))
# sp.save_npz('testmRandom.npz', modified_adj)
def Topo():
with open('dprmodel.pkl', 'rb') as f:
data = pickle.load(f)
adj, features, labels = data.adj, data.features, data.labels
adj, features, labels = preprocess(adj, features, labels, preprocess_adj=False) # conver to tensor
print(1)
idx_train, idx_val, idx_test = data.idx_train, data.idx_val, data.idx_test
# Setup Victim Model
print(2)
victim_model = GCN(nfeat=features.shape[1], nclass=labels.max().item() + 1,
nhid=16, dropout=0.5, weight_decay=5e-4, device='cpu').to('cpu')
print(3)
victim_model.fit(features, adj, labels, idx_train)
print(4)
# Setup Attack Model
model = PGDAttack(model=victim_model, nnodes=adj.shape[0], loss_type='CE', device='cpu').to('cpu')
print(5)
with open('TopoAttack.pkl', 'wb') as f:
pickle.dump(model,f)
with open('TopoAttack.pkl', 'rb') as f:
model = pickle.load(f)
print(6)
model.attack(features, adj, labels, idx_train, n_perturbations=10)
print(7)
modified_adj = model.modified_adj
sp.save_npz('testmTopo.npz', modified_adj)
def Dice():
with open('dprmodel.pkl', 'rb') as f:
data = pickle.load(f)
adj, features, labels = data.adj, data.features, data.labels
model = DICE()
with open('DiceAttack.pkl', 'wb') as f:
pickle.dump(model,f)
with open('DiceAttack.pkl', 'rb') as f:
model = pickle.load(f)
model.attack(adj, labels, n_perturbations=10)
modified_adj = sp.csr_matrix(model.modified_adj)
sp.save_npz('testmDice.npz', modified_adj)
def DeepwalkDefense():
with open('dprmodel.pkl', 'rb') as f:
data = pickle.load(f)
with open('NodeEmbeddingAttack.pkl', 'rb') as f:
attacker = pickle.load(f)
adj, features, labels = data.adj, data.features, data.labels
idx_train, idx_val, idx_test = data.idx_train, data.idx_val, data.idx_test
# set up attack model
# attacker = NodeEmbeddingAttack()
print(0)
attacker.attack(adj, attack_type="remove", n_perturbations=1000)
print(1)
modified_adj = attacker.modified_adj
sp.save_npz('testmEmbed2.npz', modified_adj)
print("Test DeepWalk on clean graph")
model = DeepWalk()
model.fit(adj)
print(2)
model.evaluate_node_classification(labels, idx_train, idx_test)#根据原图进行评估
print("Test DeepWalk on attacked graph")
model.fit(modified_adj)
print(3)
model.evaluate_node_classification(labels, idx_train, idx_test)#根据被攻击的图进行评估，svd是奇异值分解
print("Test DeepWalk SVD")
model = DeepWalk(type="svd")
with open('svd.pkl', 'wb') as f:
pickle.dump(model,f)
model.fit(modified_adj)
print(4)
model.evaluate_node_classification(labels, idx_train, idx_test)
print(5)
def main():
# Meta()
# RandomAttack()
# Topo()
# Embed()
# Dice()
# return
# data = Dataset(root='D:\\ether\\tmp', name='test') # 去除最大连通子图中的节点后的数据集
# with open('dprmodel.pkl', 'wb') as f:
# pickle.dump(data, f)
# with open('dprmodel.pkl', 'rb') as f:
# data = pickle.load(f)
# sp.save_npz('testa.npz',data.adj)
# sp.save_npz('testf.npz',data.features)
DeepwalkDefense()
if __name__ == '__main__':
main()