/
train_heterophily.py
188 lines (162 loc) · 7.86 KB
/
train_heterophily.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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
from __future__ import division
from __future__ import print_function
import time
import random
import argparse
import numpy as np
import torch
import torch.nn.functional as F
import torch.optim as optim
from process import *
from utils import *
from model import *
import uuid
from layers import *
from networkx.drawing.tests.test_pylab import plt
from sklearn.manifold import TSNE
from torch.optim.lr_scheduler import CosineAnnealingLR,CosineAnnealingWarmRestarts,StepLR
# Training settings
parser = argparse.ArgumentParser()
parser.add_argument('--seed', type=int, default=40, help='Random seed.')
parser.add_argument('--epochs', type=int, default=1500, help='Number of epochs to train.')
parser.add_argument('--lr', type=float, default=0.01, help='learning rate.')
parser.add_argument('--layer', type=int, default=64, help='Number of layers.')
parser.add_argument('--hidden', type=int, default=64, help='hidden dimensions.')
parser.add_argument('--dropout', type=float, default=0.5, help='Dropout rate (1 - keep probability).')
parser.add_argument('--patience', type=int, default=200, help='Patience')
parser.add_argument('--data', default='cora', help='dateset')
parser.add_argument('--dev', type=int, default=0, help='device id')
parser.add_argument('--al', type=float, default=0.9, help='initial proportion')#0.1
parser.add_argument('--all', type=float, default=0.1, help='initial proportion')#0.1
parser.add_argument('--model', default='LEGNN', help='model')
parser.add_argument('--bias', default=False, help='use or not bias.')
parser.add_argument('--gama1', type=float, default=1e-3, help='refine')
parser.add_argument('--gama2', type=float, default=1e-2, help='refine')
parser.add_argument('--norm', default=False, help='use or not norm.')
parser.add_argument('--lal_rate', type=float, default=0, help='label noise ratio')#0.1
parser.add_argument('--wd1', type=float, default=0, help='weight decay (L2 loss on convs parameters).')#0.01 1e-4
parser.add_argument('--wd2', type=float, default=5e-4, help='weight decay (L2 loss on fcs parameters).')#5e-4
parser.add_argument("--normalization", default="AugNormAdj",
help="The normalization on the adj matrix.")
parser.add_argument('--no_cuda', action='store_true', default=True,
help='Disables CUDA training.')
args = parser.parse_args()
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
cudaid = "cuda:"+str(args.dev)
device = torch.device(cudaid)
checkpt_file = 'pretrained/'+uuid.uuid4().hex+'.pt'
print(cudaid,checkpt_file)
def train_step(model,optimizer,features,labels,adj,idx_train,labels_for_lpa):
model.train()
optimizer.zero_grad()
if args.model == 'LEGNN':
output, Pseudo_label = model(features, adj, labels_for_lpa, idx_train)
loss_gcn = F.nll_loss(output[idx_train], labels[idx_train].to(device))
loss_label = F.nll_loss(Pseudo_label[idx_train], labels[idx_train].to(device))
Pseudo_label = torch.clamp(Pseudo_label, min=1e-7, max=1.0)
label_one_hot = torch.nn.functional.one_hot(labels, int(labels.max()) + 1).float().to(device)
label_one_hot = torch.clamp(label_one_hot, min=1e-4, max=1.0)
rce = (-1*torch.sum(Pseudo_label * torch.log(label_one_hot), dim=1)).mean()
acc_train = accuracy(output[idx_train], labels[idx_train].to(device))
loss_train = 1*loss_gcn + args.gama1*loss_label + args.gama2*rce
loss_train.backward(retain_graph=True)
optimizer.step()
return loss_train.item(), acc_train.item(),\
output[idx_train].cpu().detach().numpy(), \
labels[idx_train].cpu().detach().numpy()
def validate_step(model,features,labels,adj,idx_val,labels_for_lpa, idx_train):
model.eval()
with torch.no_grad():
output, label= model(features, adj,labels_for_lpa, idx_train)
loss_val = F.nll_loss(output[idx_val], labels[idx_val].to(device))
acc_val = accuracy(output[idx_val], labels[idx_val].to(device))
return loss_val.item(),acc_val.item()
def test_step(model,features,labels,adj,idx_test,labels_for_lpa,idx_train):
model.load_state_dict(torch.load(checkpt_file))
model.eval()
with torch.no_grad():
if args.model == 'LEGNN':
output,_= model(features, adj,labels_for_lpa,idx_train)
else:
output = model(features, adj,idx_train,labels_for_lpa)
loss_test = F.nll_loss(output[idx_test], labels[idx_test].to(device))
acc_test = accuracy(output[idx_test], labels[idx_test].to(device))
return loss_test.item(),acc_test.item(),output[idx_test].cpu().numpy(),labels[idx_test].cpu().numpy()
def train(datastr,splitstr):
adj, features, labels, idx_train, idx_val, idx_test, num_features, num_labels = full_load_data(datastr,splitstr, args.lal_rate, args.seed)
print('labels', idx_test)
labels_for_lpa = one_hot_embedding(labels, labels.max().item() + 1).type(torch.FloatTensor).to(device)
features = features.to(device)
adj = adj.to(device)
###########################################################
if args.model == "LEGNN":
model = LEGNN(nfeat=features.shape[1],
nlayers=args.layer,
nhidden=args.hidden,
nclass=int(labels.max()) + 1,
dropout=args.dropout,
al=args.al,
all=args.all,
adj = adj,
norm = args.norm).to(device)
# optimizer = optim.Adam(model.parameters(), lr=args.lr,
# weight_decay=args.weight_decay)
optimizer = optim.Adam([
{'params': model.params1, 'weight_decay': args.wd1},
{'params': model.params2, 'weight_decay': args.wd2},
], lr=args.lr)
else:
raise NotImplementedError("model error.")
bad_counter = 0
best = 999999999
best1 = 0
for epoch in range(args.epochs):
if args.model == 'LEGNN':
loss_tra, acc_tra, logits_tra, label_tra = train_step(model, optimizer, features, labels, adj, idx_train,labels_for_lpa)
loss_val, acc_val = validate_step(model, features, labels, adj, idx_val,labels_for_lpa,idx_train)
if(epoch+1)%1 == 0:
print('Epoch:{:04d}'.format(epoch+1),
'train',
'loss:{:.3f}'.format(loss_tra),
'acc:{:.2f}'.format(acc_tra*100),
'| val',
'loss:{:.3f}'.format(loss_val),
'acc:{:.2f}'.format(acc_val*100))
if loss_val < best:
best = loss_val
torch.save(model.state_dict(), checkpt_file)
acc1 = test_step(model,features,labels,adj,idx_test,labels_for_lpa,idx_train)[1]
bad_counter = 0
print(acc1)
else:
bad_counter += 1
# if acc_val > best1:
# best1 = acc_val
# torch.save(model.state_dict(), checkpt_file)
# # wandb.save('checkpt_file')
# acc1 = test_step(model,features,labels,adj,idx_test,labels_for_lpa,idx_train)[1]
# print('test', acc1*100)
# bad_counter = 0
# else:
# bad_counter += 1
if bad_counter == args.patience:
break
loss_test, acc,test_logits, test_label = test_step(model,features,labels,adj,idx_test,labels_for_lpa,idx_train)
return acc*100, test_logits, test_label
t_total = time.time()
acc_list = []
for i in range(10):
datastr = args.data
print("-----data", args.data)
splitstr = 'splits/'+args.data+'_split_0.6_0.2_'+str(i)+'.npz'
acc_test,logits,label = train(datastr,splitstr)
acc_list.append(acc_test)
print(i,": {:.2f}".format(acc_list[-1]))
print(acc_list)
print("Train cost: {:.4f}s".format(time.time() - t_total))
print("Test acc.:{:.2f}".format(np.mean(acc_list)))
print(acc_list)
print(np.std(acc_list, ddof=1))