-
Notifications
You must be signed in to change notification settings - Fork 9
/
trainer.py
78 lines (57 loc) · 2.92 KB
/
trainer.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
import torch
from test import Tester
from utils import MetricTracker
class Trainer:
def __init__(self, config, model, optimizer, criterion, dataloader):
self.config = config
self.model = model
self.optimizer = optimizer
self.criterion = criterion
self.dataloader = dataloader
self.device = next(self.model.parameters()).device
self.losses = MetricTracker()
self.accs = MetricTracker()
self.tester = Tester(self.config, self.model)
def train(self):
for epoch in range(self.config.num_epochs):
result = self._train_epoch(epoch)
print('Epoch: [{0}]\t Avg Loss {loss:.4f}\t Avg Accuracy {acc:.3f}'.format(epoch, loss=result['loss'], acc=result['acc']))
# NOTE MODIFICATION (TEST)
self.tester.eval()
if self.tester.best_acc == self.tester.accs.avg:
print('Saving Model...')
torch.save({
'epoch': epoch,
'model': self.model,
'optimizer': self.optimizer,
}, 'best_model/model.pth.tar')
def _train_epoch(self, epoch_idx):
self.model.train()
self.losses.reset()
self.accs.reset()
for batch_idx, (docs, labels, doc_lengths, sent_lengths) in enumerate(self.dataloader):
batch_size = labels.size(0)
docs = docs.to(self.device) # (batch_size, padded_doc_length, padded_sent_length)
labels = labels.to(self.device) # (batch_size)
sent_lengths = sent_lengths.to(self.device) # (batch_size, padded_doc_length)
doc_lengths = doc_lengths.to(self.device) # (batch_size)
# (n_docs, n_classes), (n_docs, max_doc_len_in_batch, max_sent_len_in_batch), (n_docs, max_doc_len_in_batch)
scores, word_att_weights, sentence_att_weights = self.model(docs, doc_lengths, sent_lengths)
# NOTE MODIFICATION (BUG)
self.optimizer.zero_grad()
loss = self.criterion(scores, labels)
loss.backward()
if self.config.max_grad_norm is not None:
torch.nn.utils.clip_grad_norm_(self.model.parameters(), self.config.max_grad_norm)
# NOTE MODIFICATION (BUG): clip grad norm should come before optimizer.step()
self.optimizer.step()
# Compute accuracy
predictions = scores.max(dim=1)[1]
correct_predictions = torch.eq(predictions, labels).sum().item()
acc = correct_predictions
self.losses.update(loss.item(), batch_size)
self.accs.update(acc, batch_size)
print('Epoch: [{0}][{1}/{2}]\t Loss {loss.val:.4f}(avg: {loss.avg:.4f})\t Acc {acc.val:.3f} (avg: {acc.avg:.3f})'.format(
epoch_idx, batch_idx, len(self.dataloader), loss=self.losses, acc=self.accs))
log = {'loss': self.losses.avg, 'acc': self.accs.avg}
return log