-
Notifications
You must be signed in to change notification settings - Fork 92
/
train_partial_ner.py
207 lines (161 loc) · 10.2 KB
/
train_partial_ner.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
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
from __future__ import print_function
import datetime
import time
import torch
import torch.autograd as autograd
import torch.nn as nn
import torch.optim as optim
import codecs
import pickle
import math
from model_partial_ner.ner import NER
import model_partial_ner.utils as utils
from model_partial_ner.object import softCE
from model_partial_ner.basic import BasicRNN
from model_partial_ner.dataset import NERDataset, TrainDataset
from torch_scope import wrapper
import argparse
import logging
import json
import os
import sys
import itertools
import functools
logger = logging.getLogger(__name__)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--gpu', type=str, default="auto")
parser.add_argument('--cp_root', default='./checkpoint')
parser.add_argument('--checkpoint_name', default='autoner0')
parser.add_argument('--git_tracking', action='store_true')
parser.add_argument('--eval_dataset', default='./data/hqner/train_0.pk')
parser.add_argument('--train_dataset', default='./data/hqner/test.pk')
parser.add_argument('--batch_token_number', type=int, default=3000)
parser.add_argument('--label_dim', type=int, default=50)
parser.add_argument('--hid_dim', type=int, default=300)
parser.add_argument('--word_dim', type=int, default=100)
parser.add_argument('--char_dim', type=int, default=30)
parser.add_argument('--layer_num', type=int, default=2)
parser.add_argument('--droprate', type=float, default=0.5)
parser.add_argument('--sample_ratio', type=float, default=1.0)
parser.add_argument('--batch_norm', action='store_true')
parser.add_argument('--epoch', type=int, default=500)
parser.add_argument('--clip', type=float, default=5)
parser.add_argument('--update', choices=['Adam', 'Adagrad', 'Adadelta', 'SGD'], default='Adam')
parser.add_argument('--rnn_layer', choices=['Basic'], default='Basic')
parser.add_argument('--rnn_unit', choices=['gru', 'lstm', 'rnn'], default='lstm')
parser.add_argument('--lr', type=float, default=-1)
parser.add_argument('--tolerance', type=int, default=5)
parser.add_argument('--lr_decay', type=float, default=0.05)
parser.add_argument('--interval', type=int, default=30)
parser.add_argument('--check', type=int, default=1000)
parser.add_argument('--seed', type=int, default=None)
args = parser.parse_args()
pw = wrapper(os.path.join(args.cp_root, args.checkpoint_name), args.checkpoint_name, enable_git_track=args.git_tracking, seed = args.seed)
gpu_index = pw.auto_device() if 'auto' == args.gpu else int(args.gpu)
device = torch.device("cuda:" + str(gpu_index) if gpu_index >= 0 else "cpu")
logger.info('loading dataset')
key_list = ['emb_array', 'w_map', 'c_map', 'tl_map', 'cl_map', 'range', 'test_data', 'dev_data']
dataset = pickle.load(open(args.eval_dataset, 'rb'))
emb_array, w_map, c_map, tl_map, cl_map, range_idx, test_data, dev_data = [dataset[tup] for tup in key_list]
id2label = {v: k for k, v in tl_map.items()}
assert len(emb_array) == len(w_map)
train_loader = TrainDataset(args.train_dataset, w_map['<\n>'], c_map['<\n>'], args.batch_token_number, sample_ratio = args.sample_ratio)
test_loader = NERDataset(test_data, w_map['<\n>'], c_map['<\n>'], args.batch_token_number)
dev_loader = NERDataset(dev_data, w_map['<\n>'], c_map['<\n>'], args.batch_token_number)
logger.info('building model')
rnn_map = {'Basic': BasicRNN}
rnn_layer = rnn_map[args.rnn_layer](args.layer_num, args.rnn_unit, args.word_dim + args.char_dim, args.hid_dim, args.droprate, args.batch_norm)
ner_model = NER(rnn_layer, len(w_map), args.word_dim, len(c_map), args.char_dim, args.label_dim, len(tl_map), args.droprate)
ner_model.rand_ini()
ner_model.load_pretrained_word_embedding(torch.FloatTensor(emb_array))
ner_config = ner_model.to_params()
ner_model.to(device)
optim_map = {'Adam' : optim.Adam, 'Adagrad': optim.Adagrad, 'Adadelta': optim.Adadelta, 'SGD': functools.partial(optim.SGD, momentum=0.9)}
if args.lr > 0:
optimizer=optim_map[args.update](ner_model.parameters(), lr=args.lr)
else:
optimizer=optim_map[args.update](ner_model.parameters())
crit_chunk = nn.BCEWithLogitsLoss()
crit_type = softCE()
# writer = SummaryWriter(log_dir='./partial/'+args.log_dir)
# name_list = ['loss_chunk', 'loss_type', 'scores_chunking', 'scores_typing', 'scores_ner']
# c_loss, t_loss, chunk_sco, type_sco, ner_sco = [args.log_dir+'/'+tup for tup in name_list]
logger.info('Saving configues.')
pw.save_configue(args)
pw.save_configue({'w_map': w_map, 'c_map': c_map, 'tl_map': tl_map, 'cl_map': cl_map}, 'dict.json')
logger.info('Setting up training environ.')
batch_index = 0
best_eval = float('-inf')
best_f1, best_pre, best_rec = -1, -1, -1
best_type2f1, best_type2pre, best_type2rec = {}, {}, {}
patience = 0
current_lr = args.lr
tolerance = args.tolerance
try:
for indexs in range(args.epoch):
logger.info('############')
logger.info('Epoch: {}'.format(indexs))
pw.nvidia_memory_map(gpu_index = gpu_index)
ner_model.train()
for word_t, char_t, chunk_mask, chunk_label, type_mask, type_label in train_loader.get_tqdm(device):
ner_model.zero_grad()
output = ner_model(word_t, char_t, chunk_mask)
chunk_score = ner_model.chunking(output)
chunk_loss = crit_chunk(chunk_score, chunk_label)
type_score = ner_model.typing(output, type_mask)
type_loss = crit_type(type_score, type_label)
loss = type_loss + chunk_loss
loss.backward()
torch.nn.utils.clip_grad_norm_(ner_model.parameters(), args.clip)
optimizer.step()
batch_index += 1
if 0 == batch_index % args.interval:
pw.add_loss_vs_batch({'loss_chunk': utils.to_scalar(chunk_loss), 'loss_type': utils.to_scalar(type_loss)}, batch_index, use_logger = False)
if 0 == batch_index % args.check:
# NER evaluation
pre_dev, rec_dev, f1_dev, type2pre_dev, type2rec_dev, type2f1_dev = utils.evaluate_ner(dev_loader.get_tqdm(device), ner_model, tl_map['None'], id2label)
pw.add_loss_vs_batch({'dev_pre': pre_dev, 'dev_rec': rec_dev}, batch_index, use_logger = False)
pw.add_loss_vs_batch({'dev_f1': f1_dev}, batch_index, use_logger = True)
pw.save_checkpoint(model = ner_model, is_best = f1_dev > best_eval, s_dict = {'config': ner_config, 'w_map': w_map, 'c_map': c_map, 'tl_map': tl_map, 'cl_map': cl_map})
if f1_dev > best_eval:
best_eval = f1_dev
# best_f1, best_pre, best_rec, best_type2pre, best_type2rec, best_type2f1 = utils.evaluate_ner(test_loader.get_tqdm(device), ner_model, tl_map['None'])
best_pre, best_rec, best_f1, best_type2pre, best_type2rec, best_type2f1 = utils.evaluate_ner(test_loader.get_tqdm(device), ner_model, tl_map['None'], id2label)
pw.add_loss_vs_batch({'test_pre': best_pre, 'test_rec': best_rec}, batch_index, use_logger = False)
pw.add_loss_vs_batch({'test_f1': best_f1}, batch_index, use_logger = True)
patience = 0
for entity_type in best_type2f1:
pw.add_loss_vs_batch({'per_{}_f1'.format(entity_type): best_type2f1[entity_type],
'per_{}_pre'.format(entity_type): best_type2pre[entity_type],
'per_{}_rec'.format(entity_type): best_type2rec[entity_type]}, batch_index, use_logger = False)
logger.info('\ttype: %s, f1: %.6f, pre: %.6f, rec: %.6f' % (entity_type, best_type2f1[entity_type], best_type2pre[entity_type], best_type2rec[entity_type]))
else:
patience += 1
if patience >= tolerance:
patience = 0
current_lr *= 0.9 ** tolerance
if args.update == 'SGD':
utils.adjust_learning_rate(optimizer, current_lr)
logger.info('current_lr = %.10f' % current_lr)
ner_model.train()
except KeyboardInterrupt:
print('Exiting from training early')
# NER evaluation
pre_dev, rec_dev, f1_dev, type2pre_dev, type2rec_dev, type2f1_dev = utils.evaluate_ner(dev_loader.get_tqdm(device), ner_model, tl_map['None'], id2label)
pw.add_loss_vs_batch({'dev_pre': pre_dev, 'dev_rec': rec_dev}, batch_index, use_logger = False)
pw.add_loss_vs_batch({'dev_f1': f1_dev}, batch_index, use_logger = True)
pw.save_checkpoint(model = ner_model, is_best = f1_dev > best_eval, s_dict = {'config': ner_config, 'w_map': w_map, 'c_map': c_map, 'tl_map': tl_map, 'cl_map': cl_map})
if f1_dev > best_eval:
best_eval = f1_dev
best_pre, best_rec, best_f1, best_type2pre, best_type2rec, best_type2f1 = utils.evaluate_ner(test_loader.get_tqdm(device), ner_model, tl_map['None'], id2label)
pw.add_loss_vs_batch({'test_pre': best_f1, 'test_rec': best_pre}, batch_index, use_logger = False)
pw.add_loss_vs_batch({'test_f1': best_rec}, batch_index, use_logger = True)
patience = 0
for entity_type in best_type2f1:
pw.add_loss_vs_batch({'per_%s_f1'.format(entity_type): best_type2f1[entity_type],
'per_%s_pre'.format(entity_type): best_type2pre[entity_type],
'per_%s_rec'.format(entity_type): best_type2rec[entity_type]}, batch_index, use_logger = False)
logger.info('\ttype: %s, f1: %.6f, pre: %.6f, rec: %.6f' % (entity_type, best_type2f1[entity_type], best_type2pre[entity_type], best_type2rec[entity_type]))
print ('\nbest dev f1: %.6f, corresponding test f1: %.6f' % (best_eval, best_f1))
pw.close()