-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathaegis.py
205 lines (171 loc) · 8.07 KB
/
aegis.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
import numpy as np
import scipy.sparse as sp
import torch
import torch.nn as nn
from model_AEGIS import Model
from utils import *
from sklearn.metrics import roc_auc_score
import random
import os
import dgl
from sklearn.metrics import precision_recall_curve, average_precision_score
import argparse
from tqdm import tqdm
os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(map(str, [2]))
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
# Set argument
parser = argparse.ArgumentParser(description='')
parser.add_argument('--dataset', type=str,
default='reddit') # ' tolokers_no_isolated 'BlogCatalog' 'Flickr' 'ACM' 'cora' 'citeseer' 'pubmed'
parser.add_argument('--lr', type=float)
parser.add_argument('--weight_decay', type=float, default=0.0)
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--embedding_dim', type=int, default=300)
parser.add_argument('--num_epoch', type=int)
parser.add_argument('--recon_num_epoch', type=int, default=10)
parser.add_argument('--drop_prob', type=float, default=0.0)
parser.add_argument('--batch_size', type=int, default=300)
parser.add_argument('--subgraph_size', type=int, default=4)
parser.add_argument('--readout', type=str, default='avg') # max min avg weighted_sum
parser.add_argument('--auc_test_rounds', type=int, default=256)
parser.add_argument('--negsamp_ratio', type=int, default=1)
args = parser.parse_args()
if args.lr is None:
if args.dataset in ['Amazon']:
args.lr = 1e-3
elif args.dataset in ['tf_finace']:
args.lr = 5e-4
elif args.dataset in ['reddit']:
args.lr = 1e-3
if args.num_epoch is None:
if args.dataset in ['reddit']:
args.num_epoch = 500
elif args.dataset in ['tf_finace']:
args.num_epoch = 1500
elif args.dataset in ['Amazon']:
args.num_epoch = 800
batch_size = args.batch_size
subgraph_size = args.subgraph_size
print('Dataset: ', args.dataset)
# Set random seed
dgl.random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
# torch.cuda.manual_seed(args.seed)
# torch.cuda.manual_seed_all(args.seed)
random.seed(args.seed)
# Load and preprocess data
adj, features, labels, all_idx, idx_train, idx_val, \
idx_test, ano_label, str_ano_label, attr_ano_label, normal_label_idx, abnormal_label_idx = load_mat(args.dataset)
if args.dataset in ['Amazon', 'tf_finace', 'reddit', 'elliptic']:
features, _ = preprocess_features(features)
else:
features = features.todense()
dgl_graph = adj_to_dgl_graph(adj)
nb_nodes = features.shape[0]
ft_size = features.shape[1]
raw_adj = adj
adj = normalize_adj(adj)
raw_adj = (raw_adj + sp.eye(raw_adj.shape[0])).todense()
adj = (adj + sp.eye(adj.shape[0])).todense()
features = torch.FloatTensor(features[np.newaxis])
adj = torch.FloatTensor(adj[np.newaxis])
raw_adj = torch.FloatTensor(raw_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)
# Initialize model and optimiser
model = Model(ft_size, args.embedding_dim, 'prelu', args.negsamp_ratio, args.readout)
optimiser_ae = torch.optim.Adam(model.parameters(), lr=1e-3, weight_decay=args.weight_decay)
optimiser = torch.optim.Adam(model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
optimiser_gen = torch.optim.Adam(model.generator.parameters(),
lr=args.lr)
if torch.cuda.is_available():
print('Using CUDA')
model.cuda()
features = features.cuda()
adj = adj.cuda()
labels = labels.cuda()
# idx_train = idx_train.cuda()
# idx_val = idx_val.cuda()
# idx_test = idx_test.cuda()
cnt_wait = 0
best = 1e9
best_t = 0
batch_num = nb_nodes // batch_size + 1
import time
# Train model
with tqdm(total=args.num_epoch) as pbar:
pbar.set_description('Training')
for epoch in range(args.recon_num_epoch):
loss_dis, loss_g, loss_ae, score_test, emb_all = model(features, adj, normal_label_idx, idx_test)
# loss_dis, loss_g, loss_ae, score_test, emb_all = model(features, adj, all_idx, idx_test)
loss_ae.backward()
optimiser_ae.step()
print("Epoch:", '%04d' % (epoch), "ae_loss=", "{:.5f}".format(loss_ae.item()))
total_time = 0
for epoch in range(args.num_epoch):
start_time = time.time()
model.train()
optimiser.zero_grad()
optimiser_gen.zero_grad()
# Train model
# loss_dis, loss_g, loss_ae, score_test, emb_all = model(features, adj, normal_label_idx, idx_test)
loss_dis, loss_g, loss_ae, score_test, emb_all = model(features, adj, all_idx, idx_test)
loss_g.backward(retain_graph=True)
loss_dis.backward(retain_graph=True)
# loss = loss_dis + loss_g
optimiser.step()
optimiser_gen.step()
score_test = np.array(score_test.detach().cpu())
emb_inf = torch.norm(emb_all, dim=-1, keepdim=True)
emb_inf = torch.pow(emb_inf, -1)
emb_inf[torch.isinf(emb_inf)] = 0.
emb_norm = emb_all * emb_inf
sim_matrix = torch.mm(emb_norm, emb_norm.T)
raw_adj = torch.squeeze(raw_adj).cuda()
similar_matrix1 = sim_matrix[:int(raw_adj.shape[0]), :int(raw_adj.shape[1])] * raw_adj
similar_matrix2 = sim_matrix[int(raw_adj.shape[0]):, int(raw_adj.shape[1]):] * raw_adj
r_inv = torch.pow(torch.sum(raw_adj, 0), -1)
r_inv[torch.isinf(r_inv)] = 0.
affinity1 = torch.sum(similar_matrix1, 0) * r_inv
affinity2 = torch.sum(similar_matrix2, 0) * r_inv
if epoch % 20 == 0:
# save data for tsne
import scipy.io as io
# tsne_data_path = 'draw/AEGIS2_tfinance/tsne_data_{}.mat'.format(str(epoch))
# # io.savemat(tsne_data_path, {'emb': np.array(emb_all.cpu().detach()), 'ano_label': ano_label,
# # 'abnormal_label_idx': np.array(abnormal_label_idx),
# # 'normal_label_idx': np.array(normal_label_idx)})
real_abnormal_label_idx = np.array(all_idx)[np.argwhere(ano_label == 1).squeeze()].tolist()
real_normal_label_idx = np.array(all_idx)[np.argwhere(ano_label == 0).squeeze()].tolist()
# real_abnormal_label_idx = real_abnormal_label_idx[:50]
real_affinity, index = torch.sort(affinity1[real_abnormal_label_idx])
real_affinity = real_affinity[:50]
draw_pdf_methods('AEGIS', np.array(affinity1[real_normal_label_idx].detach().cpu()),
np.array(affinity2[:500].detach().cpu()),
np.array(real_affinity.detach().cpu()), args.dataset, epoch)
# if epoch % 10 == 0:
# real_abnormal_label_idx = np.array(all_idx)[np.argwhere(ano_label == 1).squeeze()].tolist()
# extend_label = torch.zeros(emb_combine.size(0), 1)
# extend_label[abnormal_label_idx] = 1
# extend_label[real_abnormal_label_idx] = 2
# data_dict = dict([('embedding', emb_combine), ('Label', extend_label)])
#
# scio.savemat('embedding/{}_{}.mat'.format(args.dataset, epoch), data_dict)
#
# draw_pdf(np.array(affinity[normal_label_idx].detach()),
# np.array(affinity[abnormal_label_idx].detach()),
# np.array(affinity[real_abnormal_label_idx].detach()), args.dataset, epoch)
if epoch % 5 == 0:
print("Epoch:", '%04d' % (epoch), "train_loss=", "{:.5f}".format(loss_dis.item()))
model.eval()
auc = roc_auc_score(ano_label[idx_test], score_test)
print('Testing {} AUC:{:.4f}'.format(args.dataset, auc))
AP = average_precision_score(ano_label[idx_test], score_test, average='macro', pos_label=1,
sample_weight=None)
print('Testing AP:', AP)
print('Total time is', total_time)
end_time = time.time()
total_time += end_time - start_time