-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy patheval.py
142 lines (103 loc) · 4.53 KB
/
eval.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
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
def test(embeds, data, num_classes, FLAGS, device="cpu"):
return node_cls_downstream_task_eval(
input_emb=embeds, data=data, num_classes=num_classes,
lr=FLAGS.lr_cls, wd=FLAGS.wd_cls,
cls_epochs=FLAGS.epochs_cls, cls_runs=5, device=device)
def batch_test(embeds, data, num_classes, FLAGS, device="cpu"):
return batch_node_cls_downstream_task_eval(
input_emb=embeds, data=data, num_classes=num_classes,
lr=FLAGS.lr_cls, wd=FLAGS.wd_cls,
cls_epochs=FLAGS.epochs_cls, cls_runs=5, device=device)
def eval_acc(model, x, y):
model.eval()
with torch.no_grad():
output = model(x)
y_pred = torch.argmax(output, dim=1).squeeze(-1)
return (y_pred == y).float().mean().item()
class Classifier(nn.Module):
def __init__(self, in_dim, out_dim):
super(Classifier, self).__init__()
self.linear = nn.Linear(in_dim, out_dim)
def forward(self, x):
x = self.linear(x)
return x.log_softmax(dim=-1)
def reset_parameters(self):
self.linear.reset_parameters()
def batch_train_cls(cls, x, y, train_mask, val_mask, test_mask,
lr=1e-2, weight_decay=1e-5, epochs=100):
cls.reset_parameters()
optimizer = torch.optim.AdamW(
cls.parameters(), lr=lr, weight_decay=weight_decay)
train_x, train_y = x[train_mask], y[train_mask]
val_x, val_y = x[val_mask], y[val_mask]
test_x, test_y = x[test_mask], y[test_mask]
train_loader = DataLoader(torch.arange(train_x.size(0)), pin_memory=False, batch_size=8192, shuffle=True)
best_val_acc, best_test_acc = 0.0, 0.0
for _ in range(epochs):
for train_idx in train_loader:
cls.train()
optimizer.zero_grad()
output = cls(train_x[train_idx])
loss = F.nll_loss(output, train_y[train_idx])
loss.backward()
optimizer.step()
val_acc, test_acc = eval_acc(cls, val_x, val_y), eval_acc(cls, test_x, test_y)
if val_acc > best_val_acc:
best_val_acc = val_acc
best_test_acc = test_acc
return best_val_acc, best_test_acc
def train_cls(cls, x, y, train_mask, val_mask, test_mask,
lr=1e-2, weight_decay=1e-5, epochs=100):
cls.reset_parameters()
optimizer = torch.optim.AdamW(
cls.parameters(), lr=lr, weight_decay=weight_decay)
train_x, train_y = x[train_mask], y[train_mask]
val_x, val_y = x[val_mask], y[val_mask]
test_x, test_y = x[test_mask], y[test_mask]
best_val_acc, best_test_acc = 0.0, 0.0
for _ in range(epochs):
cls.train()
optimizer.zero_grad()
output = cls(train_x)
loss = F.nll_loss(output, train_y)
loss.backward()
optimizer.step()
val_acc, test_acc = eval_acc(cls, val_x, val_y), eval_acc(cls, test_x, test_y)
if val_acc > best_val_acc:
best_val_acc = val_acc
best_test_acc = test_acc
return best_val_acc, best_test_acc
def batch_node_cls_downstream_task_eval(input_emb, data, num_classes,
lr, wd, cls_epochs=100,
cls_runs=10, device="cpu"):
all_val_acc, all_test_acc = [], []
# input_emb = F.normalize(input_emb, dim=1) # l2 normalize
gnn_emb_dim = input_emb.size(1)
classifier = Classifier(gnn_emb_dim, num_classes).to(device)
for _ in range(cls_runs):
best_val_acc, best_test_acc = batch_train_cls(
classifier, input_emb, data.y,
data.train_mask, data.val_mask, data.test_mask,
lr=lr, weight_decay=wd, epochs=cls_epochs)
all_val_acc.append(best_val_acc)
all_test_acc.append(best_test_acc)
return all_val_acc, all_test_acc
def node_cls_downstream_task_eval(input_emb, data, num_classes,
lr, wd, cls_epochs=100,
cls_runs=10, device="cpu"):
all_val_acc, all_test_acc = [], []
# input_emb = F.normalize(input_emb, dim=1) # l2 normalize
gnn_emb_dim = input_emb.size(1)
classifier = Classifier(gnn_emb_dim, num_classes).to(device)
for _ in range(cls_runs):
best_val_acc, best_test_acc = train_cls(
classifier, input_emb, data.y,
data.train_mask, data.val_mask, data.test_mask,
lr=lr, weight_decay=wd, epochs=cls_epochs)
all_val_acc.append(best_val_acc)
all_test_acc.append(best_test_acc)
return all_val_acc, all_test_acc