-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain_lstm_crf.py
160 lines (146 loc) · 7.38 KB
/
train_lstm_crf.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
# -*- coding: utf-8 -*-
import os
import sys
import time
import random
import argparse
import logging
import warnings
import collections
import torch
import numpy as np
import torch.nn as nn
from torch.utils.data import DataLoader
#from pytorch_transformers import BertTokenizer
from transformers import BertTokenizer, BertTokenizerFast, get_linear_schedule_with_warmup as linear_warmup_schedule
from transformers.optimization import AdamW
from conf.config import *
from business.data_process.data_utils import parse_json_data, NERDataset, NERDatasetSplit
from business.models.model import AlBertNERModel, BertNERModel, AlBertLSTMModel
from business.model_plant import validate, train
from business.tools import setup_seed, FGM, EMA
logging.basicConfig(level=logging.INFO, format='%(asctime)s: %(message)s')
warnings.simplefilter(action='ignore')
now_time = time.strftime("%Y%m%d%H", time.localtime())
def main(args):
setup_seed(2000)
device = torch.device("cuda:{}".format(args.gpu_index) if torch.cuda.is_available() else "cpu")
tokenizer = BertTokenizerFast.from_pretrained(model_name, do_lower_case=True)
logging.info(20 * "=" + " Preparing for training " + 20 * "=")
if not os.path.exists(args.target_dir):
os.makedirs(args.target_dir)
#if True:
if args.train_file != args.val_file:
logging.info("\t* Loading training data...")
train_data, label_2_ids = parse_json_data(args.train_file, tokenizer)
ids_2_label = {v:k for k,v in label_2_ids.items()}
label_nums = len(label_2_ids)
train_dataset = NERDataset(label_2_ids, *train_data)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=args.batch_size)
print(len(train_dataset))
logging.info("\t* Loading validation data...")
val_data, _ = parse_json_data(args.val_file, tokenizer)
val_dataset = NERDataset(label_2_ids, *val_data)
val_loader = DataLoader(val_dataset, shuffle=False, batch_size=args.batch_size)
print(len(val_dataset))
else:
logging.info("\t* Loading training data...")
train_data, label_2_ids = parse_json_data(args.train_file, tokenizer)
print(label_2_ids)
ids_2_label = {v:k for k,v in label_2_ids.items()}
label_nums = len(label_2_ids)
total_idx = list(range(len(train_data[0])))
random.shuffle(total_idx)
train_dataset = NERDatasetSplit(label_2_ids, total_idx[:int(0.9*len(total_idx))], *train_data)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=args.batch_size)
print(len(train_dataset))
logging.info("\t* Loading validation data...")
val_dataset = NERDatasetSplit(label_2_ids, total_idx[int(0.9*len(total_idx)):], *train_data)
val_loader = DataLoader(val_dataset, shuffle=False, batch_size=args.batch_size)
print(len(val_dataset))
logging.info("\t* Building model...")
config = Config()
config.num_label = label_nums
if model_name.find('albert') >= 0:
model = AlBertLSTMModel(config, use_crf=use_crf).to(device)
elif model_name.find('roberta') >= 0:
model = BertNERModel(config).to(device)
#model = AlBertLSTMModel(config, device, use_crf=False).to(device)
#model = BilstmCRF(config, device, use_crf=False).to(device)
#model = nn.DataParallel(model, device_ids=[1, 2])
#param_optimizer = list(model.module.named_parameters())
bert_optimizer = list(model.bert.named_parameters())
lstm_optimizer = list(model.bilstm.named_parameters())
classifier_optimizer = list(model.classifier.named_parameters())
#params = filter(lambda p: p.requires_grad, model.parameters())
no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']
optimizer_grouped_parameters = [
{'params': [p for n, p in bert_optimizer if not any(nd in n for nd in no_decay)],
'weight_decay': weight_decay},
{'params': [p for n, p in bert_optimizer if any(nd in n for nd in no_decay)],
'weight_decay': 0.0},
{'params': [p for n, p in lstm_optimizer if not any(nd in n for nd in no_decay)],
'lr': learning_rate * lr_mag, 'weight_decay': weight_decay},
{'params': [p for n, p in lstm_optimizer if any(nd in n for nd in no_decay)],
'lr': learning_rate * lr_mag, 'weight_decay': 0.0},
{'params': [p for n, p in classifier_optimizer if not any(nd in n for nd in no_decay)],
'lr': learning_rate * lr_mag, 'weight_decay': weight_decay},
{'params': [p for n, p in classifier_optimizer if any(nd in n for nd in no_decay)],
'lr': learning_rate * lr_mag, 'weight_decay': 0.0},
{'params': model.crf.parameters(), 'lr': learning_rate * lr_mag}
]
optimizer = AdamW(optimizer_grouped_parameters, lr=learning_rate)
#scheduler1 = linear_warmup_schedule(optimizer, num_warmup_steps=100, num_training_steps = len(train_loader)*args.epochs)
scheduler2 = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode="min", factor=sd_factor, patience=0, min_lr=2e-6)
print(optimizer.defaults['lr'])
best_score = 0.0
best_loss = 100
start_epoch = 1
fgm = None
ema = None
if args.use_EMA:
ema = EMA(model, 0.999)
ema.register()
print("use ema success...")
if args.attack_type=='FGM':
fgm = FGM(model, epsilon=1, emb_name='word_embeddings')
print("use fgm success...")
_, valid_loss, _= validate(model, val_loader, device, label_nums, ids_2_label, ema=None)
logging.info("\t* Validation loss before training: {:.4f}".format(valid_loss))
logging.info("\n" + 20 * "=" + "Training Bert model on device: {}".format(device) + 20 * "=")
patience_counter = 0
for epoch in range(start_epoch, args.epochs + 1):
logging.info("* Training epoch {}:".format(epoch))
print(optimizer.param_groups[0]['lr'])
epoch_time, epoch_loss, epoch_accuracy = train(model, train_loader, optimizer,
args.max_grad_norm, device, label_nums, scheduler=None, fgm=fgm, ema=ema)
#for g in optimizer.param_groups:
# g['lr'] *= 0.95
logging.info("-> Training time: {:.4f}s, loss = {:.4f}, accuracy: {:.4f}%"
.format(epoch_time, epoch_loss, (epoch_accuracy * 100)))
logging.info("* Validation for epoch {}:".format(epoch))
epoch_time, epoch_loss, epoch_score = validate(model, val_loader, device, label_nums, ids_2_label, ema=ema)
logging.info("-> Valid. time: {:.4f}s, loss: {:.4f}"
.format(epoch_time, epoch_loss))
scheduler2.step(epoch_loss)
#scheduler.step(epoch_accuracy)
if epoch_score < best_score:
patience_counter += 1
else:
best_score = epoch_score
patience_counter = 0
if ema:
ema.apply_shadow()
torch.save({"epoch": epoch,
"model": model.state_dict(),
"best_score": best_score},
os.path.join(args.target_dir, "best.pth.tar.%s" % now_time))
if ema:
ema.restore()
if patience_counter >= args.patience:
logging.info("-> Early stopping: patience limit reached, stopping...")
break
if __name__ == "__main__":
for k in args.__dict__:
logging.info(k + ": " + str(args.__dict__[k]))
main(args)