-
Notifications
You must be signed in to change notification settings - Fork 1
/
execute_bwt.py
211 lines (166 loc) · 5.11 KB
/
execute_bwt.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
import numpy as np
import scipy.sparse as sp
import torch
import torch.nn as nn
import random
from models import DGI, LogReg, DGI_BW
from utils import process
from deeprobust.graph.data import Dataset, PrePtbDataset
from deeprobust.graph.global_attack import DICE
torch.cuda.set_device(5)
seed = 23456
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
dataset = 'Citeseer'
cuda = True
# training params
batch_size = 1
nb_epochs = 500
patience = 20
lr = 0.001
l2_coef = 0.0
drop_prob = 0.0
hid_units = 128
sparse = True
bw = True
gamma = 0.005
nonlinearity = 'prelu' # special name to separate parameters
# 75.2616
# adj, features, labels, idx_train, idx_val, idx_test = process.load_data(dataset)
pdata = PrePtbDataset(root='data/meta', name='cora', attack_method='meta', ptb_rate=0.05)
padj = pdata.adj
tnode = pdata.get_target_nodes()
print(len(tnode))
data = Dataset(root='/tmp/', name='cora', setting='nettack')
adj, features, labels = data.adj, data.features, data.labels
idx_train, idx_val, idx_test = data.idx_train, data.idx_val, data.idx_test
features, _ = process.preprocess_features(features)
model = model = DICE()
model.attack(adj, labels, n_perturbations=100)
# padj = model.modified_adj
# padj = adj
adj = padj
nb_nodes = features.shape[0]
ft_size = features.shape[1]
nb_classes = max(labels) + 1
print(nb_nodes, nb_classes)
adj = process.normalize_adj(adj + sp.eye(adj.shape[0]))
if sparse:
sp_adj = process.sparse_mx_to_torch_sparse_tensor(adj)
else:
adj = (adj + sp.eye(adj.shape[0])).todense()
features = torch.FloatTensor(features[np.newaxis])
if not sparse:
adj = torch.FloatTensor(adj[np.newaxis])
labels = torch.FloatTensor(labels[np.newaxis])
idx_train = torch.LongTensor(idx_train)
idx_val = torch.LongTensor(idx_val)
idx_test = torch.LongTensor(idx_test)
model = DGI_BW(ft_size, hid_units, nonlinearity, cuda)
optimiser = torch.optim.Adam(model.parameters(), lr=lr, weight_decay=l2_coef)
if torch.cuda.is_available() and cuda:
print('Using CUDA')
model.cuda()
features = features.cuda()
if sparse:
sp_adj = sp_adj.cuda()
else:
adj = adj.cuda()
labels = labels.cuda()
idx_train = idx_train.cuda()
idx_val = idx_val.cuda()
idx_test = idx_test.cuda()
b_xent = nn.BCEWithLogitsLoss()
xent = nn.CrossEntropyLoss()
cnt_wait = 0
best = 1e9
best_t = 0
for epoch in range(nb_epochs):
model.train()
optimiser.zero_grad()
idx = np.random.permutation(nb_nodes)
shuf_fts = features[:, idx, :]
lbl_1 = torch.ones(batch_size, nb_nodes)
lbl_2 = torch.zeros(batch_size, nb_nodes)
lbl = torch.cat((lbl_1, lbl_2), 1)
if torch.cuda.is_available() and cuda:
shuf_fts = shuf_fts.cuda()
lbl = lbl.cuda()
logits = model(features, shuf_fts, sp_adj if sparse else adj, sparse, None, None, None)
loss = b_xent(logits, lbl)
if bw:
bw_loss = model.bw_loss(model.gcn.forward1(features, sp_adj if sparse else adj, sparse))
loss += gamma * bw_loss
print('Loss:', loss)
if loss < best:
best = loss
best_t = epoch
cnt_wait = 0
torch.save(model.state_dict(), 'best_dgi.pkl')
else:
cnt_wait += 1
if cnt_wait == patience:
print('Early stopping!')
break
loss.backward()
optimiser.step()
print('Loading {}th epoch'.format(best_t))
model.load_state_dict(torch.load('best_dgi.pkl'))
adj = padj
print(sparse)
if sparse:
sp_adj = process.sparse_mx_to_torch_sparse_tensor(adj)
else:
adj = (adj + sp.eye(adj.shape[0])).todense()
if torch.cuda.is_available() and cuda:
if sparse:
sp_adj = sp_adj.cuda()
else:
adj = adj.cuda()
embeds, _ = model.embed(features, sp_adj if sparse else adj, sparse, None)
print(embeds.shape)
train_embs = embeds[idx_train]
val_embs = embeds[idx_val]
test_embs = embeds[idx_test]
target_embs = embeds[tnode]
labels = labels.squeeze().long()
# train_lbls = torch.argmax(labels[idx_train], dim=1)
# val_lbls = torch.argmax(labels[idx_val], dim=1)
# test_lbls = torch.argmax(labels[idx_test], dim=1)
train_lbls = labels[idx_train]
val_lbls = labels[idx_val]
test_lbls = labels[idx_test]
target_lbls = labels[tnode]
tot = torch.zeros(1)
if torch.cuda.is_available() and cuda:
tot = tot.cuda()
accs = []
for _ in range(50):
log = LogReg(hid_units, nb_classes)
opt = torch.optim.Adam(log.parameters(), lr=0.01, weight_decay=5e-4)
if torch.cuda.is_available() and cuda:
log.cuda()
pat_steps = 0
best_acc = torch.zeros(1)
if torch.cuda.is_available() and cuda:
best_acc = best_acc.cuda()
for _ in range(100):
log.train()
opt.zero_grad()
# print(train_embs.shap)
logits = log(train_embs)
loss = xent(logits, train_lbls)
loss.backward()
opt.step()
logits = log(target_embs)
preds = torch.argmax(logits, dim=1)
acc = torch.sum(preds == target_lbls).float() / target_lbls.shape[0]
accs.append(acc * 100)
print(acc)
tot += acc
print('Average accuracy:', tot / 50)
accs = torch.stack(accs)
print(accs.mean())
print(accs.std())