/
DGI_inductive.py
116 lines (93 loc) · 4.02 KB
/
DGI_inductive.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
import torch
import os.path as osp
import GCL.losses as L
from torch import nn
from tqdm import tqdm
from torch.optim import Adam
from GCL.eval import get_split, LREvaluator
from GCL.models import SingleBranchContrast
from torch_geometric.nn import SAGEConv
from torch_geometric.nn.inits import uniform
from torch_geometric.data import NeighborSampler
from torch_geometric.datasets import Reddit
class GConv(nn.Module):
def __init__(self, input_dim, hidden_dim, num_layers):
super(GConv, self).__init__()
self.layers = torch.nn.ModuleList()
self.activations = torch.nn.ModuleList()
for i in range(num_layers):
if i == 0:
self.layers.append(SAGEConv(input_dim, hidden_dim))
else:
self.layers.append(SAGEConv(hidden_dim, hidden_dim))
self.activations.append(nn.PReLU(hidden_dim))
def forward(self, x, adjs):
for i, (edge_index, _, size) in enumerate(adjs):
x_target = x[:size[1]]
x = self.layers[i]((x, x_target), edge_index)
x = self.activations[i](x)
return x
class Encoder(torch.nn.Module):
def __init__(self, encoder, hidden_dim):
super(Encoder, self).__init__()
self.encoder = encoder
self.project = torch.nn.Linear(hidden_dim, hidden_dim)
uniform(hidden_dim, self.project.weight)
@staticmethod
def corruption(x, edge_index):
return x[torch.randperm(x.size(0))], edge_index
def forward(self, x, edge_index):
z = self.encoder(x, edge_index)
g = self.project(torch.sigmoid(z.mean(dim=0, keepdim=True)))
zn = self.encoder(*self.corruption(x, edge_index))
return z, g, zn
def train(encoder_model, contrast_model, data, dataloader, optimizer):
encoder_model.train()
total_loss = total_examples = 0
for batch_size, node_id, adjs in dataloader:
adjs = [adj.to('cuda') for adj in adjs]
optimizer.zero_grad()
z, g, zn = encoder_model(data.x[node_id], adjs)
loss = contrast_model(h=z, g=g, hn=zn)
loss.backward()
optimizer.step()
total_loss += loss.item() * z.shape[0]
total_examples += z.shape[0]
return total_loss / total_examples
def test(encoder_model, data, dataloader):
encoder_model.eval()
zs = []
for i, (batch_size, node_id, adjs) in enumerate(dataloader):
adjs = [adj.to('cuda') for adj in adjs]
z, _, _ = encoder_model(data.x[node_id], adjs)
zs.append(z)
x = torch.cat(zs, dim=0)
split = get_split(num_samples=x.size()[0], train_ratio=0.1, test_ratio=0.8)
result = LREvaluator()(x, data.y, split)
return result
def main():
import torch.multiprocessing
torch.multiprocessing.set_sharing_strategy('file_system')
device = torch.device('cuda')
path = osp.join(osp.expanduser('~'), 'datasets', 'Reddit')
dataset = Reddit(path)
data = dataset[0].to(device)
train_loader = NeighborSampler(data.edge_index, node_idx=None,
sizes=[10, 10, 25], batch_size=128,
shuffle=True, num_workers=32)
test_loader = NeighborSampler(data.edge_index, node_idx=None,
sizes=[10, 10, 25], batch_size=128,
shuffle=False, num_workers=32)
gconv = GConv(input_dim=dataset.num_features, hidden_dim=512, num_layers=3).to(device)
encoder_model = Encoder(encoder=gconv, hidden_dim=512).to(device)
contrast_model = SingleBranchContrast(loss=L.JSD(), mode='G2L').to(device)
optimizer = Adam(encoder_model.parameters(), lr=0.0001)
with tqdm(total=30, desc='(T)') as pbar:
for epoch in range(1, 31):
loss = train(encoder_model, contrast_model, data, train_loader, optimizer)
pbar.set_postfix({'loss': loss})
pbar.update()
test_result = test(encoder_model, data, test_loader)
print(f'(E): Best test F1Mi={test_result["micro_f1"]:.4f}, F1Ma={test_result["macro_f1"]:.4f}')
if __name__ == '__main__':
main()