-
Notifications
You must be signed in to change notification settings - Fork 76
/
metapath2vec_imdb_dblp_trainer.py
170 lines (144 loc) · 6.99 KB
/
metapath2vec_imdb_dblp_trainer.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
import argparse
import os
# os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# os.environ['TL_BACKEND'] = 'torch'
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
# 0:Output all; 1:Filter out INFO; 2:Filter out INFO and WARNING; 3:Filter out INFO, WARNING, and ERROR
import os.path as osp
import tensorlayerx as tlx
from gammagl.datasets import IMDB, DBLP
from gammagl.models import MetaPath2Vec
from tensorlayerx.model import WithLoss, TrainOneStep
from sklearn.linear_model import LogisticRegression
from gammagl.utils import mask_to_index
if tlx.BACKEND == 'torch': # when the backend is torch and you want to use GPU
try:
tlx.set_device(device='GPU', id=0)
except:
print("GPU is not available")
class Unsupervised_Loss(WithLoss):
def __init__(self, net, loss_fn):
super(Unsupervised_Loss, self).__init__(backbone=net, loss_fn=loss_fn)
def forward(self, data, label):
logits = self.backbone_network(data["pos_rw"], data["neg_rw"])
loss = self._loss_fn(logits, label)
return loss
def calculate_acc(train_z, train_y, test_z, test_y, solver='lbfgs', multi_class='auto', max_iter=150):
train_z = tlx.convert_to_numpy(train_z)
train_y = tlx.convert_to_numpy(train_y)
test_z = tlx.convert_to_numpy(test_z)
test_y = tlx.convert_to_numpy(test_y)
clf = LogisticRegression(solver=solver, multi_class=multi_class, max_iter=max_iter).fit(train_z, train_y)
return clf.score(test_z, test_y)
def main(args, log_steps=10):
# load datasets
if str.lower(args.dataset) not in ['imdb', 'dblp']:
raise ValueError('Unknown dataset: {}'.format(args.dataset))
targetType = {
'imdb': 'movie',
'dblp': 'author'
}
if str.lower(args.dataset) == 'imdb':
path = osp.join(osp.dirname(osp.realpath(__file__)), '../../data/IMDB')
dataset = IMDB(path)
graph = dataset[0]
metapath = [
('actor', 'to', 'movie'),
('movie', 'to', 'director'),
('director', 'to', 'movie'),
('movie', 'to', 'actor')
]
elif str.lower(args.dataset) == 'dblp':
path = osp.join(osp.dirname(osp.realpath(__file__)), '../../data/DBLP')
dataset = DBLP(path)
graph = dataset[0]
metapath = [
('author', 'to', 'paper'),
('paper', 'to', 'term'),
('term', 'to', 'paper'),
('paper', 'to', 'author')
]
graph[targetType[str.lower(args.dataset)]].train_mask = tlx.convert_to_tensor(
graph[targetType[str.lower(args.dataset)]].train_mask)
graph[targetType[str.lower(args.dataset)]].val_mask = tlx.convert_to_tensor(
graph[targetType[str.lower(args.dataset)]].val_mask)
graph[targetType[str.lower(args.dataset)]].test_mask = tlx.convert_to_tensor(
graph[targetType[str.lower(args.dataset)]].test_mask)
train_idx = mask_to_index(graph[targetType[str.lower(args.dataset)]].train_mask)
val_idx = mask_to_index(graph[targetType[str.lower(args.dataset)]].val_mask)
test_idx = mask_to_index(graph[targetType[str.lower(args.dataset)]].test_mask)
model = MetaPath2Vec(graph.edge_index_dict,
embedding_dim=args.embedding_dim,
metapath=metapath,
walk_length=args.walk_length,
context_size=args.window_size,
walks_per_node=args.num_walks,
num_negative_samples=args.num_negative_samples,
name="MetaPath2Vec")
loader = model.loader(batch_size=args.batch_size, shuffle=True)
optimizer = tlx.optimizers.Adam(lr=args.lr)
train_weights = model.trainable_weights
loss_func = Unsupervised_Loss(net=model, loss_fn=tlx.losses.absolute_difference_error)
train_one_step = TrainOneStep(loss_func, optimizer, train_weights)
data = {
"y": tlx.convert_to_tensor(graph[targetType[args.dataset]].y),
'train_idx': train_idx,
'val_idx': val_idx,
'test_idx': test_idx
}
best_val_acc = 0
for epoch in range(args.n_epoch):
total_loss = 0
for i, (pos_rw, neg_rw) in enumerate(loader):
data['pos_rw'] = pos_rw
data['neg_rw'] = neg_rw
model.set_train()
train_loss = train_one_step(data, tlx.convert_to_tensor(0, dtype=tlx.float32))
total_loss += train_loss.item()
if (i + 1) % log_steps == 0:
model.set_eval()
z = model.campute(targetType[args.dataset])
val_acc = calculate_acc(tlx.gather(z, data['train_idx']), tlx.gather(data['y'], data['train_idx']),
tlx.gather(z, data['val_idx']), tlx.gather(data['y'], data['val_idx']),
max_iter=150)
print((f'Epoch: {epoch}, Step: {i + 1:05d}/{len(loader)}, '
f'Loss: {total_loss / log_steps:.4f}, '
f'Acc: {val_acc:.4f}'))
total_loss = 0
# save best model on evaluation set
if val_acc > best_val_acc:
best_val_acc = val_acc
model.save_weights(
args.best_model_path + tlx.BACKEND + "_" + args.dataset + "_" + model.name + ".npz",
format='npz_dict')
model.load_weights(args.best_model_path + tlx.BACKEND + "_" + args.dataset + "_" + model.name + ".npz",
format='npz_dict', skip=True)
model.set_eval()
z = model.campute(targetType[args.dataset])
test_acc = calculate_acc(tlx.gather(z, data['train_idx']), tlx.gather(data['y'], data['train_idx']),
tlx.gather(z, data['test_idx']), tlx.gather(data['y'], data['test_idx']),
max_iter=150)
print("Test acc: {:.4f}".format(test_acc))
return test_acc
if __name__ == '__main__':
# parameters setting
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', type=str, default='imdb', help='dataset')
parser.add_argument("--dataset_path", type=str, default=r'', help="path to save dataset")
parser.add_argument("--best_model_path", type=str, default=r'./', help="path to save best model")
parser.add_argument("--lr", type=float, default=0.01, help="learning rate")
parser.add_argument("--n_epoch", type=int, default=50, help="number of epoch")
parser.add_argument("--embedding_dim", type=int, default=16)
parser.add_argument("--walk_length", type=int, default=50)
parser.add_argument("--num_walks", type=int, default=5)
parser.add_argument("--window_size", type=int, default=7)
parser.add_argument("--train_ratio", type=float, default=0.1)
parser.add_argument("--num_negative_samples", type=int, default=5)
parser.add_argument("--batch_size", type=int, default=128)
parser.add_argument("--gpu", type=int, default=0)
args = parser.parse_args()
if args.gpu >= 0:
tlx.set_device("GPU", args.gpu)
else:
tlx.set_device("CPU")
main(args, log_steps=10)