-
Notifications
You must be signed in to change notification settings - Fork 71
/
merit_trainer.py
146 lines (132 loc) · 6.42 KB
/
merit_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
from cProfile import label
import numpy as np
import scipy.sparse as sp
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 sys
sys.path.append(os.getcwd())
import tensorlayerx as tlx
from gammagl.models.merit import MERIT
from process import compute_diff
from gammagl.datasets import Planetoid
from tensorlayerx.model import TrainOneStep, WithLoss
from eval import evaluation
from tqdm import tqdm
from gammagl.utils.corrupt_graph import dfde_norm_g
from gammagl.utils.norm import calc_gcn_norm
class Unsupervised_Loss(WithLoss):
def __init__(self, net):
super(Unsupervised_Loss, self).__init__(backbone=net, loss_fn=None)
def forward(self, data, label):
loss = self._backbone(data['feat1'], data['edge1'], data['weight1'], data['num_node1'],
data['feat2'], data['edge2'], data['weight2'], data['num_node2'])
return loss
def main(args):
# load dataset
if str.lower(args.dataset) not in ['cora', 'pubmed', 'citeseer']:
raise ValueError('Unknown dataset: {}'.format(args.dataset))
dataset = Planetoid(args.dataset_path, args.dataset)
# dataset=Amazon(root='./Amazon/',name='photo')
graph = dataset[0]
num_node = graph.num_nodes
edge_index = graph.edge_index
weight = tlx.ops.convert_to_tensor(calc_gcn_norm(edge_index, graph.num_nodes))
graph.edge_weight = weight
row, col = edge_index[0], edge_index[1]
# origin_graph = Graph(x=graph.x, edge_index=tlx.convert_to_tensor([row, col], dtype=tlx.int64),
# num_nodes=graph.num_nodes, y=graph.y)
# origin_graph.edge_weight = tlx.convert_to_tensor(weight)
features = graph.x
labels = graph.y
train_mask = tlx.convert_to_tensor(graph.train_mask)
val_mask = tlx.convert_to_tensor(graph.val_mask)
test_mask = tlx.convert_to_tensor(graph.test_mask)
weight = np.ones(edge_index.shape[1])
sparse_adj = sp.coo_matrix((weight, (edge_index[0], edge_index[1])), shape=(num_node, num_node))
diff_edge_index, diff_weight = compute_diff(sparse_adj, alpha=args.alpha, eps=0.0001)
# define model
net = MERIT(
feat_size=dataset.num_node_features,
projection_size=args.proj_size,
projection_hidden_size=args.proj_hid,
prediction_size=args.pred_size,
prediction_hidden_size=args.pred_hid,
moving_average_decay=args.momentum, beta=args.beta)
optimizer = tlx.optimizers.Adam(lr=args.lr, weight_decay=args.weight_decay)
train_weights = net.online_encoder.trainable_weights + net.online_predictor.trainable_weights
loss_func = Unsupervised_Loss(net)
train_one_step = TrainOneStep(loss_func, optimizer, train_weights)
best = 0
# cnt_wait = 0
patience_count = 0
results = []
result_over_runs = []
features = tlx.convert_to_tensor(features, dtype='float32')
for epoch in tqdm(range(args.n_epoch)):
net.set_train()
for _ in range(args.batch_size):
graph1 = dfde_norm_g(graph.edge_index, features, args.drop_feat_rate_1,
args.drop_edge_rate_1)
graph2 = dfde_norm_g(graph.edge_index, features, args.drop_feat_rate_2,
args.drop_edge_rate_2)
data = {"feat1": graph1.x, "edge1": graph1.edge_index, "weight1": graph1.edge_weight,
"num_node1": graph1.num_nodes, \
"feat2": graph2.x, "edge2": graph2.edge_index, "weight2": graph2.edge_weight,
"num_node2": graph2.num_nodes,
}
loss = train_one_step(data, label=tlx.convert_to_tensor([0]))
net.update_ma()
if epoch % args.eval_every_epoch == 0:
net.set_eval()
acc = evaluation(graph.edge_index, graph.edge_weight, diff_edge_index, diff_weight, \
features, net.online_encoder.gnn, train_mask, test_mask, labels, tlx.nn.PRelu(args.gnn_dim))
if acc > best:
best = acc
patience_count = 0
else:
patience_count += 1
results.append(acc)
print('\t epoch {:03d} | loss {:.4f} | acc {:.4f}'.format(epoch, loss.item(), acc))
if patience_count >= args.patience:
print('Early Stopping.')
break
result_over_runs.append(max(results))
print('\t best acc {:.5f}'.format(max(results)))
if __name__ == '__main__':
# parameters setting
parser = argparse.ArgumentParser()
parser.add_argument('--num_layers', type=int, default=2)
parser.add_argument('--dataset', type=str, default='cora', help='dataset,cora/pubmed/citeseer')
parser.add_argument("--dataset_path", type=str, default=r'', help="path to save dataset")
parser.add_argument('--runs', type=int, default=1)
parser.add_argument('--eval_every_epoch', type=int, default=1)
parser.add_argument('--n_epoch', type=int, default=500)
parser.add_argument('--lr', type=float, default=3e-4)
parser.add_argument('--weight_decay', type=float, default=0.0)
parser.add_argument('--batch_size', type=int, default=4)
parser.add_argument('--sample_size', type=int, default=2000)
parser.add_argument('--patience', type=int, default=100)
parser.add_argument('--gnn_dim', type=int, default=512)
parser.add_argument('--proj_size', type=int, default=512)
parser.add_argument('--proj_hid', type=int, default=4096)
parser.add_argument('--pred_size', type=int, default=512)
parser.add_argument('--pred_hid', type=int, default=4096)
parser.add_argument('--momentum', type=float, default=0.8)
parser.add_argument('--beta', type=float, default=0.5)
parser.add_argument('--alpha', type=float, default=0.05)
parser.add_argument('--drop_edge_rate_1', type=float, default=0.2)
parser.add_argument('--drop_edge_rate_2', type=float, default=0.2)
parser.add_argument('--drop_feat_rate_1', type=float, default=0.5)
parser.add_argument('--drop_feat_rate_2', type=float, default=0.5)
parser.add_argument("--best_model_path", type=str, default=r'./', help="path to save best model")
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)