-
Notifications
You must be signed in to change notification settings - Fork 1
/
enhance_main.py
311 lines (281 loc) · 15.5 KB
/
enhance_main.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
294
295
296
297
298
299
300
301
302
303
304
305
"""Unified interface to all dynamic graph model experiments"""
import math
import random
import sys
from tqdm import tqdm
import argparse
import os.path as osp
import h5py
import torch
import pandas as pd
import numpy as np
import os
import warnings
warnings.filterwarnings("ignore")
from sklearn.metrics import average_precision_score
from sklearn.metrics import roc_auc_score
from utils import EarlyStopMonitor, RandEdgeSampler, load_subgraph_margin, get_item, get_item_edge, NeighborFinder
from models import *
from GraphM import GraphMixer
from TGN.tgn import TGN
degree_dict = {"wikipedia":20, "reddit":20 ,"uci":30 ,"mooc":60, "enron": 30, "canparl": 30, "uslegis": 30}
### Argument and global variables
parser = argparse.ArgumentParser('Motif Enhancement Verification')
parser.add_argument('--gpu', type=int, default=0, help='idx for the gpu to use')
parser.add_argument("--base_type", type=str, default="tgn", help="tgn or graphmixer or tgat")
parser.add_argument('--data', type=str, help='data sources to use, try wikipedia or reddit', default='wikipedia')
parser.add_argument('--bs', type=int, default=1000, help='batch_size')
parser.add_argument('--test_bs', type=int, default=1000, help='test batch_size')
parser.add_argument('--n_degree', type=int, default=20, help='number of neighbors to sample')
parser.add_argument('--n_head', type=int, default=3, help='number of heads used in attention layer')
parser.add_argument('--n_epoch', type=int, default=100, help='number of epochs')
parser.add_argument('--out_dim', type=int, default=32, help='number of attention dim')
parser.add_argument('--hid_dim', type=int, default=32, help='number of hidden dim')
parser.add_argument('--temp', type=float, default=0.07, help='temperature')
parser.add_argument('--num_layers', type=int, default=2, help='number of network layers')
parser.add_argument('--lr', type=float, default=1e-3, help='learning rate')
parser.add_argument('--drop_out', type=float, default=0.1, help='dropout probability')
parser.add_argument('--if_bern', type=bool, default=True, help='use bernoulli')
parser.add_argument('--save_model', type=bool, default=False, help='if save model')
parser.add_argument('--verbose', type=int, default=3, help='use dot product attention or mapping based')
parser.add_argument('--weight_decay', type=float, default=0)
parser.add_argument('--lr_decay', type=float, default=0.999)
parser.add_argument('--task_type', type=str, default="motif-enhanced prediction")
parser.add_argument('--wandb_sync', type=str, default="disabled", help='online or disabled')
try:
args = parser.parse_args()
except:
parser.print_help()
sys.exit(0)
def norm_imp(imp):
imp[imp < 0] = 0
imp += 1e-16
return imp / imp.sum()
### Load data and train val test split
def load_data(mode):
g_df = pd.read_csv(osp.join(osp.dirname(osp.realpath(__file__)), 'processed/ml_{}.csv'.format(args.data)))
val_time, test_time = list(np.quantile(g_df.ts, [0.70, 0.85]))
src_l = g_df.u.values
dst_l = g_df.i.values
e_idx_l = g_df.idx.values
label_l = g_df.label.values
ts_l = g_df.ts.values
max_src_index = src_l.max()
max_idx = max(src_l.max(), dst_l.max())
random.seed(2023)
total_node_set = set(np.unique(np.hstack([g_df.u.values, g_df.i.values])))
num_total_unique_nodes = len(total_node_set)
mask_node_set = set(random.sample(set(src_l[ts_l > val_time]).union(set(dst_l[ts_l > val_time])),
int(0.1 * num_total_unique_nodes)))
mask_src_flag = g_df.u.map(lambda x: x in mask_node_set).values
mask_dst_flag = g_df.i.map(lambda x: x in mask_node_set).values
none_node_flag = (1 - mask_src_flag) * (1 - mask_dst_flag)
valid_train_flag = (ts_l <= val_time) * (none_node_flag > 0)
train_src_l = src_l[valid_train_flag]
train_dst_l = dst_l[valid_train_flag]
train_ts_l = ts_l[valid_train_flag]
train_e_idx_l = e_idx_l[valid_train_flag]
train_label_l = label_l[valid_train_flag]
valid_val_flag = (ts_l <= test_time) * (ts_l > val_time)
valid_test_flag = ts_l > test_time
val_src_l = src_l[valid_val_flag]
val_dst_l = dst_l[valid_val_flag]
test_src_l = src_l[valid_test_flag]
test_dst_l = dst_l[valid_test_flag]
test_ts_l = ts_l[valid_test_flag]
test_e_idx_l = e_idx_l[valid_test_flag]
test_label_l = label_l[valid_test_flag]
adj_list = [[] for _ in range(max_idx + 1)]
for src, dst, eidx, ts in zip(train_src_l, train_dst_l, train_e_idx_l, train_ts_l):
adj_list[src].append((dst, eidx, ts))
adj_list[dst].append((src, eidx, ts))
train_ngh_finder = NeighborFinder(adj_list)
full_adj_list = [[] for _ in range(max_idx + 1)]
for src, dst, eidx, ts in zip(src_l, dst_l, e_idx_l, ts_l):
full_adj_list[src].append((dst, eidx, ts))
full_adj_list[dst].append((src, eidx, ts))
full_ngh_finder = NeighborFinder(full_adj_list)
train_rand_sampler = RandEdgeSampler((train_src_l,), (train_dst_l,))
test_rand_sampler = RandEdgeSampler((train_src_l, val_src_l, test_src_l), (train_dst_l, val_dst_l, test_dst_l))
if mode == "test":
return test_rand_sampler, test_src_l, test_dst_l, test_ts_l, test_label_l, test_e_idx_l, full_ngh_finder
else:
return train_rand_sampler, train_src_l, train_dst_l, train_ts_l, train_label_l, train_e_idx_l, train_ngh_finder
def eval_one_epoch(args, base_model, predictor, full_ngh_finder, src, dst, ts, val_e_idx_l, epoch, best_accuracy,
test_pack, test_edge):
test_aps = []
test_auc = []
test_acc = []
test_loss = []
num_test_instance = len(src) - 1
num_test_batch = math.ceil(num_test_instance / args.test_bs) - 1
idx_list = np.arange(num_test_instance)
criterion = torch.nn.BCEWithLogitsLoss()
base_model.set_neighbor_sampler(full_ngh_finder)
for k in tqdm(range(num_test_batch)):
s_idx = k * args.test_bs
e_idx = min(num_test_instance - 1, s_idx + args.test_bs)
if s_idx == e_idx:
continue
batch_idx = idx_list[s_idx:e_idx]
src_l_cut = src[batch_idx]
dst_l_cut = dst[batch_idx]
ts_l_cut = ts[batch_idx]
e_l_cut = val_e_idx_l[batch_idx] if (val_e_idx_l is not None) else None
subgraph_src, subgraph_tgt, subgraph_bgd, walks_src, walks_tgt, walks_bgd, dst_l_fake = get_item(test_pack, batch_idx)
edge_id_feature = get_item_edge(test_edge, batch_idx)
predictor.eval()
with torch.no_grad():
#########################
subgraph_src = base_model.grab_subgraph(src_l_cut, ts_l_cut)
subgraph_tgt = base_model.grab_subgraph(dst_l_cut, ts_l_cut)
subgraph_bgd = base_model.grab_subgraph(dst_l_fake, ts_l_cut)
#########################
src_emb, tgt_emb, bgd_emb = base_model.get_node_emb(src_l_cut, dst_l_cut, dst_l_fake, ts_l_cut, e_l_cut,
subgraph_src, subgraph_tgt, subgraph_bgd)
pos_logit, neg_logit = predictor.enhance_predict_agg(ts_l_cut, walks_src, walks_tgt, walks_bgd,
edge_id_feature, src_emb, tgt_emb, bgd_emb)
size = len(src_l_cut)
pos_label = torch.ones((size, 1), dtype=torch.float, device=args.device, requires_grad=False)
neg_label = torch.zeros((size, 1), dtype=torch.float, device=args.device, requires_grad=False)
loss = criterion(pos_logit, pos_label) + criterion(neg_logit, neg_label)
pos_prob = pos_logit.sigmoid().squeeze(-1)
neg_prob = neg_logit.sigmoid().squeeze(-1)
pred_score = np.concatenate([(pos_prob).cpu().detach().numpy(), (neg_prob).cpu().detach().numpy()])
pred_label = pred_score > 0.5
true_label = np.concatenate([np.ones(size), np.zeros(size)])
test_aps.append(average_precision_score(true_label, pred_score))
test_auc.append(roc_auc_score(true_label, pred_score))
test_acc.append((pred_label == true_label).mean())
test_loss.append(loss.item())
aps_epoch = np.mean(test_aps)
auc_epoch = np.mean(test_auc)
acc_epoch = np.mean(test_acc)
loss_epoch = np.mean(test_loss)
wandb_dict = {'Test Loss': loss_epoch, "Test Aps": aps_epoch, "Test Auc": auc_epoch, 'Test Acc': acc_epoch}
wandb.log(wandb_dict)
print((f'Testing Epoch: {epoch} | '
f'Testing loss: {loss_epoch} | '
f'Testing Aps: {aps_epoch} | '
f'Testing Auc: {auc_epoch} | '
f'Testing Acc: {acc_epoch} | '))
if aps_epoch > best_accuracy:
if args.save_model:
model_path = osp.join(osp.dirname(osp.realpath(__file__)), 'params', f'predictors/{args.base_type}/')
if not osp.exists(model_path):
os.makedirs(model_path)
save_path = f"{args.data}.pt"
torch.save(predictor, osp.join(model_path, save_path))
print(f"Save model to {osp.join(model_path, save_path)}")
return aps_epoch
else:
return best_accuracy
def train(args, base_model, train_pack, test_pack, train_edge, test_edge):
if args.base_type == "tgat":
predictor = TempME_TGAT(base_model, data=args.data, out_dim=args.out_dim, hid_dim=args.hid_dim, temp=args.temp,
dropout_p=args.drop_out, device=args.device)
else:
predictor = TempME(base_model, base_model_type=args.base_type, data=args.data, out_dim=args.out_dim, hid_dim=args.hid_dim,
temp=args.temp, if_cat_feature=True, dropout_p=args.drop_out, device=args.device)
predictor = predictor.to(args.device)
optimizer = torch.optim.Adam(list(predictor.parameters()) + list(base_model.parameters()),
lr=args.lr,
betas=(0.9, 0.999), eps=1e-8,
weight_decay=args.weight_decay)
criterion = torch.nn.BCEWithLogitsLoss()
rand_sampler, src_l, dst_l, ts_l, label_l, e_idx_l, ngh_finder = load_data(mode="training")
test_rand_sampler, test_src_l, test_dst_l, test_ts_l, test_label_l, test_e_idx_l, full_ngh_finder = load_data(
mode="test")
num_instance = len(src_l) - 1
num_batch = math.ceil(num_instance / args.bs)
best_acc = 0
print(f"start training: {args.data}")
print('num of training instances: {}'.format(num_instance))
print('num of batches per epoch: {}'.format(num_batch))
idx_list = np.arange(num_instance)
np.random.shuffle(idx_list)
for epoch in range(args.n_epoch):
base_model.set_neighbor_sampler(ngh_finder)
train_aps = []
train_auc = []
train_acc = []
train_loss = []
np.random.shuffle(idx_list)
predictor.train()
base_model.train()
for k in tqdm(range(num_batch)):
s_idx = k * args.bs
e_idx = min(num_instance - 1, s_idx + args.bs)
if s_idx == e_idx:
continue
batch_idx = idx_list[s_idx:e_idx]
src_l_cut, dst_l_cut = src_l[batch_idx], dst_l[batch_idx]
ts_l_cut = ts_l[batch_idx]
e_l_cut = e_idx_l[batch_idx]
subgraph_src, subgraph_tgt, subgraph_bgd, walks_src, walks_tgt, walks_bgd, dst_l_fake = get_item(train_pack,
batch_idx)
edge_id_feature = get_item_edge(train_edge, batch_idx)
optimizer.zero_grad()
#########################
subgraph_src = base_model.grab_subgraph(src_l_cut, ts_l_cut)
subgraph_tgt = base_model.grab_subgraph(dst_l_cut, ts_l_cut)
subgraph_bgd = base_model.grab_subgraph(dst_l_fake, ts_l_cut)
#########################
src_emb, tgt_emb, bgd_emb = base_model.get_node_emb(src_l_cut, dst_l_cut, dst_l_fake, ts_l_cut, e_l_cut,
subgraph_src, subgraph_tgt, subgraph_bgd)
pos_logit, neg_logit = predictor.enhance_predict_agg(ts_l_cut, walks_src, walks_tgt, walks_bgd,
edge_id_feature, src_emb, tgt_emb, bgd_emb)
size = len(src_l_cut)
pos_label = torch.ones((size, 1), dtype=torch.float, device=args.device, requires_grad=False)
neg_label = torch.zeros((size, 1), dtype=torch.float, device=args.device, requires_grad=False)
loss = criterion(pos_logit, pos_label) + criterion(neg_logit, neg_label)
loss.backward()
optimizer.step()
if args.base_type == "tgn":
base_model.memory.detach_memory()
with torch.no_grad():
pos_prob = pos_logit.sigmoid().squeeze(-1)
neg_prob = neg_logit.sigmoid().squeeze(-1)
pred_score = np.concatenate([(pos_prob).cpu().detach().numpy(), (neg_prob).cpu().detach().numpy()])
pred_label = pred_score > 0.5
true_label = np.concatenate([np.ones(size), np.zeros(size)])
train_aps.append(average_precision_score(true_label, pred_score))
train_auc.append(roc_auc_score(true_label, pred_score))
train_acc.append((pred_label == true_label).mean())
train_loss.append(loss.item())
aps_epoch = np.mean(train_aps)
auc_epoch = np.mean(train_auc)
acc_epoch = np.mean(train_acc)
loss_epoch = np.mean(train_loss)
wandb_dict = {'Train Loss': loss_epoch, "Train Aps": aps_epoch, "Train Auc": auc_epoch, 'Train Acc': acc_epoch}
wandb.log(wandb_dict)
print((f'Training Epoch: {epoch} | '
f'Training loss: {loss_epoch} | '
f'Training Aps: {aps_epoch} | '
f'Training Auc: {auc_epoch} | '
f'Training Acc: {acc_epoch} | '))
### evaluation:
if (epoch + 1) % args.verbose == 0:
if args.base_type == "tgn":
train_memory_backup = base_model.memory.backup_memory()
best_acc = eval_one_epoch(args, base_model, predictor, full_ngh_finder, test_src_l,
test_dst_l, test_ts_l, test_e_idx_l, epoch, best_acc, test_pack, test_edge)
if args.base_type == "tgn":
base_model.memory.restore_memory(train_memory_backup)
if __name__ == '__main__':
args.device = torch.device('cuda:{}'.format(args.gpu))
args.n_degree = degree_dict[args.data]
gnn_model_path = osp.join(osp.dirname(osp.realpath(__file__)), 'params', 'tgnn',
f'{args.base_type}_{args.data}.pt')
base_model = torch.load(gnn_model_path).to(args.device)
pre_load_train = h5py.File(osp.join(osp.dirname(osp.realpath(__file__)), 'processed', f'{args.data}_train_cat.h5'),
'r')
pre_load_test = h5py.File(osp.join(osp.dirname(osp.realpath(__file__)), 'processed', f'{args.data}_test_cat.h5'),
'r')
e_feat = np.load(osp.join(osp.dirname(osp.realpath(__file__)), 'processed', f'ml_{args.data}.npy'))
n_feat = np.load(osp.join(osp.dirname(osp.realpath(__file__)), 'processed', f'ml_{args.data}_node.npy'))
train_pack = load_subgraph_margin(args, pre_load_train)
test_pack = load_subgraph_margin(args, pre_load_test)
train_edge = np.load(osp.join(osp.dirname(osp.realpath(__file__)), 'processed', f'{args.data}_train_edge.npy'))
test_edge = np.load(osp.join(osp.dirname(osp.realpath(__file__)), 'processed', f'{args.data}_test_edge.npy'))
train(args, base_model, train_pack=train_pack, test_pack=test_pack, train_edge=train_edge, test_edge=test_edge)