-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
173 lines (139 loc) · 7.78 KB
/
train.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
from __future__ import absolute_import, division, print_function, unicode_literals
import argparse, random, os, logging, glob
from tqdm import tqdm
import torch
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
import torch.nn as nn
import torch.nn.functional as F
from pytorch_transformers import AdamW, WarmupLinearSchedule
import torch_optimizer as optim
import utils.constant as config
from utils.data_loader import NERDataset, pad_collate
from utils.log import logger, init_logger
from models.cnn_lstm import CNNBiLSTM
device = config.device
def to_cpu(tensors):
for t in tensors:
t.cpu()
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("-epoch", default=config.epoch, type=int)
parser.add_argument("-batch_size", default=config.batch_size, type=int)
parser.add_argument("-lstm_layers", default=config.lstm_layers, type=int)
parser.add_argument("-channel_in", default=config.channel_in, type=int)
parser.add_argument("-channel_out", default=config.channel_out, type=int)
parser.add_argument("-kernel_sizes", default=config.kernel_sizes, type=list)
parser.add_argument("-max_char_len", default=config.max_char_len, type=int)
parser.add_argument('-save_dir', default='/epoch_{}_batch_{}_ch_in_{}_ch_out_{}')
# parser.add_argument("-is_distill", type=str2bool, nargs='?',const=True,default=False)
args = parser.parse_args()
log_path = './result' + args.save_dir.format(args.epoch, args.batch_size, args.channel_in, args.channel_out,)
init_logger(log_path,'/log.txt')
tb_writer = SummaryWriter('{}/runs'.format(log_path))
# Load Entity Dictionary, Train and Test data
vocab_size = len(torch.load('./data/word_vocab.pt'))
char_vocab_size = len(torch.load('./data/char_vocab.pt'))
pos_vocab_size = len(torch.load('./data/pos_vocab.pt'))
entitiy_to_index = torch.load('./data/processed_data/entity_to_index.pt')
num_class = len(entitiy_to_index)
print("Load processed data...")
# Load process train and validation data
train_dataset = NERDataset('tr')
valid_dataset = NERDataset('valid')
# Build train_and validation loaders which generate data with batch_size
train_loader = DataLoader(dataset=train_dataset, batch_size=args.batch_size, shuffle=True,
collate_fn=pad_collate, drop_last=True, num_workers=0)
valid_loader = DataLoader(dataset=valid_dataset,batch_size=args.batch_size, shuffle=True,
collate_fn=pad_collate, drop_last = True, num_workers=0)
print("Build Model...")
model = CNNBiLSTM(config, num_class, vocab_size, char_vocab_size, pos_vocab_size)
criterion = nn.CrossEntropyLoss()
# Prepare optimizer and schedule (linear warmup and decay)
print("Set Optimized and Scheduler...")
param_optimizer = list(model.named_parameters())
no_decay = ['bias', 'LayerNorm.weight', 'LayerNorm.bias']
optimizer_grouped_parameters = [
{'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)],
'weight_decay': 0.01},
{'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)],
'weight_decay': 0.0}]
# t_total = train_examples_len // model_config.gradient_accumulation_steps * model_config.epochs
t_total = len(train_loader) * args.epoch
# optimizer = AdamW(optimizer_grouped_parameters, lr=config.learning_rate)
optimizer = optim.RAdam(model.parameters(), lr=config.learning_rate)
# scheduler = WarmupLinearSchedule(optimizer, int(t_total*0.1), t_total)
# Train
logger.info("***** Running training *****")
logger.info(" Num train examples = %d", len(train_loader))
logger.info(" Num validation examples = %d", len(valid_loader))
logger.info(" Num Epochs = %d", args.epoch)
logger.info(" Batch size = %d", args.batch_size)
logger.info(" Total steps = %d", t_total)
logger.info(" Model save directory = %s", log_path)
global_step = 0
best_eval_acc = 0.0
print('Train Start !')
for e in tqdm(range(args.epoch)):
tr_acc, tr_loss = 0.0, 0.0
model.train()
model.to(device)
# print('Train at Epoch {}'.format(e+1))
for step, batch in enumerate(train_loader):
global_step += 1
token_idx, char_idx, pos_idx, label = batch
# Model Input
logit = model(token_idx, char_idx, pos_idx)
loss = criterion(logit.view(-1, logit.size(-1)), label.view(-1))
tr_loss += loss.item()
loss.backward()
with torch.no_grad():
accuracy = (logit.argmax(-1)==label).float()[label!=0].mean().item()
tr_acc += accuracy
to_cpu([token_idx, char_idx, pos_idx, label])
if global_step >0 and global_step % config.gradient_accumulation_steps == 0:
# global_step += config.gradient_accumulation_steps
optimizer.step()
optimizer.zero_grad()
model.zero_grad()
# scheduler.step()
# if global_step % 100==0: #int(len(self.train_dataloader)/5) ==0:
# tr_avg_acc = tr_acc / global_step
# tr_avg_loss = tr_loss / global_step
# logger.info('epoch : {} /{}, global_step : {} /{}, tr_avg_loss: {:.5f}, tr_loss : {:.5f}, tr_avg_acc: {:.2%}, tr_acc: {:.2%}'.format(
# e+1, args.epoch, global_step, t_total, tr_avg_loss, loss.item(), tr_avg_acc, accuracy))
# tb_writer.add_scalars('tr_loss', {'average': tr_avg_loss, 'current': loss.item()}, global_step)
tr_avg_acc = tr_acc / step
tr_avg_loss = tr_loss / step
logger.info('>>> Train Epoch : {} Tr_avg_loss: {:.5f}, Tr_avg_acc: {:.2%}'.format(e+1, tr_avg_loss, tr_avg_acc))
""" Evaluation"""
model.eval()
eval_acc, eval_loss = 0.0, 0.0
# print('Evaluate at Epoch {}'.format(e+1))
for step, batch in enumerate(valid_loader):
token_idx, char_idx, pos_idx, label = batch
logit = model(token_idx, char_idx, pos_idx)
loss = criterion(logit.view(-1, logit.size(-1)), label.view(-1))
eval_loss += loss.item()
with torch.no_grad():
eval_accuracy = (logit.argmax(-1)==label).float()[label!=0].mean().item()
eval_acc += eval_accuracy
to_cpu([token_idx, char_idx, pos_idx, label])
eval_avg_acc = eval_acc / step
eval_avg_loss = eval_loss / step
logger.info('VALID epoch : {} /{}, Eval_avg_loss: {:.5f}, Eval_avg_acc: {:.2%}'.format(e+1, args.epoch, eval_avg_loss, eval_avg_acc))
tb_writer.add_scalars('eval_loss', {'average': eval_avg_loss, 'current': loss.item()}, global_step)
""""""
# Model Save
if eval_avg_acc > best_eval_acc:
model.to('cpu')
best_eval_acc = eval_avg_acc
state = {'epoch':e+1,'model_state_dict': model.state_dict()}
save_path = '{}/epoch_{}_step_{}_tr_acc_{:.3f}_eval_acc_{:.3f}.pt'.format(
log_path, e+1, global_step, tr_avg_acc, eval_avg_acc)
if len(glob.glob(log_path+'/epoch*.pt'))>0:
os.remove(glob.glob(log_path+'/epoch*.pt')[0])
torch.save(state, save_path)
logger.info('Model saving with best eval acc : {:.2%}'.format(eval_avg_acc))
# if len(glob.glob(log_path+'/epoch*.pt'))==0:
os.mkdir(log_path+'/epoch_{}_step_{}_tr_acc_{:.3f}_eval_acc_{:.3f}'.format(e+1, global_step, tr_avg_acc, eval_avg_acc))