-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataset.py
125 lines (96 loc) · 3.99 KB
/
dataset.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
import numpy as np
import torch
import dgl
from dgl.data import (
CoraGraphDataset,
CiteseerGraphDataset,
PubmedGraphDataset,
AmazonCoBuyPhotoDataset, AmazonCoBuyComputerDataset,
CoauthorCSDataset, CoauthorPhysicsDataset,
WikiCSDataset
)
from ogb.nodeproppred import DglNodePropPredDataset
from sklearn.preprocessing import StandardScaler
GRAPH_DICT = {
"cora": CoraGraphDataset,
"citeseer": CiteseerGraphDataset,
"pubmed": PubmedGraphDataset,
"ogbn-arxiv": DglNodePropPredDataset,
"wikics": WikiCSDataset,
"photo": AmazonCoBuyPhotoDataset,
"computer": AmazonCoBuyComputerDataset,
"cs": CoauthorCSDataset,
"physics": CoauthorPhysicsDataset
}
def preprocess(graph):
feat = graph.ndata["feat"]
graph = dgl.to_bidirected(graph)
graph.ndata["feat"] = feat
graph = graph.remove_self_loop().add_self_loop()
graph.create_formats_()
return graph
def scale_feats(x):
scaler = StandardScaler()
feats = x.numpy()
scaler.fit(feats)
feats = torch.from_numpy(scaler.transform(feats)).float()
return feats
def load_dataset(dataset_name):
assert dataset_name in GRAPH_DICT, f"Unknow dataset: {dataset_name}."
if dataset_name.startswith("ogbn"):
dataset = GRAPH_DICT[dataset_name](dataset_name, root='/home/zhengyimei/dataset/')
else:
dataset = GRAPH_DICT[dataset_name]()
citegraph = ['cora', 'citeseer', 'pubmed', 'wikics']
cograph = ['photo', 'computer', 'cs', 'physics']
if dataset_name == "ogbn-arxiv":
graph, labels = dataset[0]
num_nodes = graph.num_nodes()
split_idx = dataset.get_idx_split()
train_idx, val_idx, test_idx = split_idx["train"], split_idx["valid"], split_idx["test"]
graph = preprocess(graph)
if not torch.is_tensor(train_idx):
train_idx = torch.as_tensor(train_idx)
val_idx = torch.as_tensor(val_idx)
test_idx = torch.as_tensor(test_idx)
feat = graph.ndata["feat"]
feat = scale_feats(feat)
graph.ndata["feat"] = feat
train_mask = torch.full((num_nodes,), False).index_fill_(0, train_idx, True)
val_mask = torch.full((num_nodes,), False).index_fill_(0, val_idx, True)
test_mask = torch.full((num_nodes,), False).index_fill_(0, test_idx, True)
graph.ndata["label"] = labels.view(-1)
graph.ndata["train_mask"], graph.ndata["val_mask"], graph.ndata["test_mask"] = train_mask, val_mask, test_mask
elif dataset_name in citegraph:
graph = dataset[0]
graph = graph.remove_self_loop()
graph = graph.add_self_loop()
train_mask = graph.ndata["train_mask"]
val_mask = graph.ndata["val_mask"]
test_mask = graph.ndata["test_mask"]
train_idx = torch.nonzero(train_mask, as_tuple=False).squeeze()
val_idx = torch.nonzero(val_mask, as_tuple=False).squeeze()
test_idx = torch.nonzero(test_mask, as_tuple=False).squeeze()
elif dataset_name in cograph:
graph = dataset[0]
graph = graph.remove_self_loop()
graph = graph.add_self_loop()
# split training/validing/testing
train_ratio = 0.1
val_ratio = 0.1
test_ratio = 0.8
N = graph.number_of_nodes()
train_num = int(N * train_ratio)
val_num = int(N * (train_ratio + val_ratio))
idx = np.arange(N)
np.random.shuffle(idx)
train_idx = torch.tensor(idx[:train_num])
val_idx = torch.tensor(idx[train_num:val_num])
test_idx = torch.tensor(idx[val_num:])
train_mask = torch.full((N,), False).index_fill_(0, train_idx, True)
val_mask = torch.full((N,), False).index_fill_(0, val_idx, True)
test_mask = torch.full((N,), False).index_fill_(0, test_idx, True)
graph.ndata["train_mask"], graph.ndata["val_mask"], graph.ndata["test_mask"] = train_mask, val_mask, test_mask
num_features = graph.ndata["feat"].shape[1]
num_classes = dataset.num_classes
return graph, (num_features, num_classes)