-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrainRotatE.py
169 lines (143 loc) · 6.53 KB
/
trainRotatE.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
import os
import sys
sys.path.append(os.path.dirname(os.path.dirname(os.path.realpath(__file__))))
import Config as Config
import models.RotatE as RotatE
from examples.base import TrainBase, adjust_learning_rate
import torch.nn.functional as F
import utils.util as util
import torch
from dataloader.dataloader import *
class RotatEModel(TrainBase):
def __init__(self, config):
super(RotatEModel, self).__init__(config)
self.args = config
self.model = RotatE.RotatE(self.args.modelparam)
print(self.model)
def get_iterator(self):
train_dataloader_head = DataLoader(
AdversarialDataset(self.args.trainpath, self.args.modelparam.entTotal, 'head-batch'),
batch_size=self.args.batch_size,
shuffle=self.args.shuffle,
num_workers=self.args.numworkers,
drop_last=self.args.drop_last,
collate_fn=AdversarialDataset.collate_fn)
train_dataloader_tail = DataLoader(
AdversarialDataset(self.args.trainpath, self.args.modelparam.entTotal, 'tail-batch'),
batch_size=self.args.batch_size,
shuffle=self.args.shuffle,
num_workers=self.args.numworkers,
drop_last=self.args.drop_last,
collate_fn=AdversarialDataset.collate_fn)
test_dataloader = DataLoader(
TestDataset(self.args.testpath),
batch_size=self.args.eval_batch_size,
num_workers=self.args.evalnumberworkers,
shuffle=False,
drop_last=False
)
return [train_dataloader_head, train_dataloader_tail], test_dataloader
def train(self):
model = self.model
epochs = self.args.epochs
lr = self.args.learningrate
optimizer = self.load_opt(model)
if self.args.usegpu:
model.cuda()
globalstep = 0
globalepoch = 0
minLoss = float("inf")
train_iterator, test_dataloader = self.get_iterator()
for epoch in range(epochs):
globalepoch += 1
print("=" * 20 + "EPOCHS(%d/%d)" % (globalepoch, epochs) + "=" * 20)
step = 0
model.train()
for dataloader in train_iterator:
for pos_sample, neg_sample, subsampling_weight, mode in dataloader:
if self.args.usegpu:
pos_sample = pos_sample.cuda()
neg_sample = neg_sample.cuda()
subsampling_weight = subsampling_weight.cuda()
negative_score = model(neg_sample, mode=mode[0])
positive_score = model(pos_sample, mode=mode[0])
loss, positive_sample_loss, negative_sample_loss = model.loss(positive_score, negative_score,
subsampling_weight)
if self.args.usegpu:
lossVal = torch.sum(loss).cpu().item()
pl = torch.sum(positive_sample_loss).cpu().item()
nl = torch.sum(negative_sample_loss).cpu().item()
else:
lossVal = torch.sum(loss).item()
pl = torch.sum(positive_sample_loss).item()
nl = torch.sum(negative_sample_loss).item()
# Calculate the gradient and step down
optimizer.zero_grad()
loss.backward()
optimizer.step()
# Print infomation and add to summary
if minLoss > lossVal:
minLoss = lossVal
if step % 50 == 0:
print("[TRAIN-EPOCH(%d/%d)-STEP(%d)]Loss:%f, minLoss:%f" % (
epoch + 1, epochs, step, lossVal, minLoss))
step += 1
globalstep += 1
self.sumwriter.add_scalar('RotatE/train/loss', lossVal, global_step=globalstep)
self.sumwriter.add_scalar('RotatE/train/positive_sample_loss', pl, global_step=globalstep)
self.sumwriter.add_scalar('RotatE/train/negative_sample_loss', nl, global_step=globalstep)
self.sumwriter.add_scalar('RotatE/train/lr', lr, global_step=globalstep)
if globalepoch % self.args.lrdecayepoch == 0:
adjust_learning_rate(optimizer, decay=self.args.lrdecay)
lr = lr * self.args.lrdecay
if globalepoch % self.args.evalepoch == 0:
# eval the model
print('begin eval the model')
model.eval()
for mode in ['head-batch', 'tail-batch']:
evalstep = 0
hit10_ = 0
hit1_ = 0
mr = 0
for data in test_dataloader:
evalstep += 1
if self.args.usegpu:
data = data.cuda()
ranks, hit1, hit10 = model.eval_model(data, mode=mode)
if evalstep % 1000 == 0:
print("[TEST-EPOCH(%d/%d)-STEP(%d)]mr:%f, hit@10:%f" % (
globalepoch, epochs, evalstep, ranks, hit10))
mr += ranks
hit10_ += hit10
hit1_ += hit1
mr /= evalstep
hit10_ /= evalstep
hit1_ /= evalstep
title = 'RotatE/test/' + mode
self.sumwriter.add_scalar(title+'/hit@10', hit10_, global_step=epoch + 1)
self.sumwriter.add_scalar(title+'/hit@1', hit1_, global_step=epoch + 1)
self.sumwriter.add_scalar(title+'/MR', mr, global_step=epoch + 1)
variable_list = {
'step': globalstep,
'lr': lr,
'MR': mr,
'hit@10': hit10_
}
self.save_model(model, optimizer, variable_list)
if __name__ == '__main__':
config = Config.Config()
config.model = 'RotatE'
config.batch_size = 256
config.eval_batch_size = 16
config.learningrate = 0.0001
config.lrdecay = 1.0
config.init()
util.printArgs(config)
util.printArgs(config.modelparam)
model = RotatEModel(config)
if config.init_checkpoint:
optimizer = model.load_opt(model.model)
model.load_model(model.model, optimizer)
model.fit(model.model, optimizer)
else:
model.train()