-
Notifications
You must be signed in to change notification settings - Fork 8
/
dsner.py
387 lines (282 loc) · 15 KB
/
dsner.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
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
import os
import sys
# necessary to add cwd to path when script run
# by slurm (since it executes a copy)
sys.path.append(os.getcwd())
import random
import argparse
import logging
import torch
import torch.optim as optim
import constants as C
from vocab import Vocab
from utils import load_word_vectors_EC, build_vocab, write_output
from dataset import EC_PA_Datset
from trainer import Trainer_PA_SL_DSNER, Trainer_PA
from model import BiLSTM_CRF_PA_CRF_LightNER, policy_selector
from conlleval import evaluate, tags_to_labels
from weight_init import *
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
def parse_args():
parser = argparse.ArgumentParser(description="PyTorch implementation of Distantly Supervised NER with Partial Annotation Learning and Reinforcement Learning ")
base_dir = os.path.dirname(os.path.dirname(os.path.realpath(__file__)))
data_dir = os.path.join(base_dir, 'data')
parser.add_argument('--data', default=f'{data_dir}/EC/',
help='path to dataset')
parser.add_argument('--glove', default=f'{data_dir}/EC/embeddings/',
help='directory with glove word embeddings')
parser.add_argument('--save', default='checkpoints/',
help='directory to save checkpoints in')
parser.add_argument('--expname', type=str, default='test',
help='Name to identify experiment')
parser.add_argument('--cased', type=bool, default=False,
help='if we consider lower cased true of surface form false')
parser.add_argument('--max_len', default=75, type=int,
help="Max lenght of input")
parser.add_argument('--word_size', default=100, type=int,
help="word embedding size (default: 100)")
parser.add_argument('--freeze_embed', action='store_true', default=False,
help='Freeze word embeddings')
parser.add_argument('--batch_size', default=64, type=int,
help="batchsize for optimizer updates (default: 25)")
parser.add_argument('--epochs', default=800, type=int,
help="number of total epochs to run (default: 10)")
parser.add_argument('--lr', default=1e-3, type=float,
help="initial learning rate (default: 0.001)")
parser.add_argument('--weight_decay', default=1e-4, type=int,
help="weight decay (default: 0.0001)")
parser.add_argument('--dropout', default=0.5, type=float,
help="use dropout (default: 0.5), -1: no dropout")
parser.add_argument('--emblr', default=0.1, type=float,
help="initial embedding learning rate (default: 0.1)")
parser.add_argument('--optim', default="adam",
help="optimizer (default: adam)")
parser.add_argument('--seed', default=123, type=int,
help="random seed (default: 123)")
parser.add_argument('--SL_hidden_size', default=100, type=int,
help="selector hidden size (default: 100)")
parser.add_argument('--biLSTM_hidden_size', default=100, type=int,
help="word BiLSTM hidden size (default: 200)")
parser.add_argument('--iobes', default=False, type=bool,
help="if to use BIOES tagging style")
parser.add_argument('--setup', default='H',
help="which dataset: H, A+H")
parser.add_argument('--mode', default='SL',
help="Mode of training: PA+SL,PA ,SL, CRF")
return parser.parse_args()
def main():
global args
args = parse_args()
# logger
logger = logging.getLogger(__name__)
logger.setLevel(logging.DEBUG)
formatter = logging.Formatter("[%(asctime)s] %(levelname)s:%(name)s:%(message)s")
# console logger
ch = logging.StreamHandler()
ch.setLevel(logging.DEBUG)
ch.setFormatter(formatter)
logger.addHandler(ch)
args.cuda = torch.cuda.is_available()
torch.manual_seed(args.seed)
random.seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
if not os.path.exists(args.save):
os.makedirs(args.save)
##print args:
dataset = 'EC'
print(f'{args.mode}')
train_dir = os.path.join(args.data, 'train/')
dev_dir = os.path.join(args.data, 'dev/')
test_dir = os.path.join(args.data, 'test/')
ds_pa_dir=os.path.join(args.data, 'ds_pa/')
ner_vocab_file = os.path.join(args.data, 'ner.vocab')
if not os.path.isfile(ner_vocab_file):
token_files = [os.path.join(split, 'a.txt') for split in
[train_dir, test_dir, dev_dir, ds_pa_dir] ]#[train_dir, dev_dir, test_dir]
ner_vocab_file = os.path.join(args.data, 'ner.vocab')
build_vocab(token_files, ner_vocab_file)
vocab = Vocab(filename=ner_vocab_file)
vocab.add_specials([C.UNK_WORD, C.PAD_WORD])
ner_vocab_file = os.path.join(args.data, 'ner.tags.vocab' )
if not os.path.isfile(ner_vocab_file):
tags_files = [os.path.join(split, 'tags.txt') for split in
[train_dir, dev_dir, test_dir, ds_pa_dir]] # [train_dir, dev_dir, test_dir]
ner_vocab_file = os.path.join(args.data, 'ner.tags.vocab')
build_vocab(tags_files, ner_vocab_file)
tags_vocab = Vocab(filename=ner_vocab_file)
tags_vocab.add_specials([C.BOS_WORD,C.PAD_WORD])
ner_vocab_file = os.path.join(args.data, 'ner.tags-iobes.vocab')
if not os.path.isfile(ner_vocab_file):
tags_files = [os.path.join(split, 'tags-iobes.txt') for split in
[train_dir, dev_dir, test_dir, ds_pa_dir]] # [train_dir, dev_dir, test_dir]
ner_vocab_file = os.path.join(args.data, 'ner-iobes.tags.vocab')
build_vocab(tags_files, ner_vocab_file)
tags_iobes_vocab = Vocab(filename=ner_vocab_file) #
#tags_iobes_vocab.add_specials([C.BOS_WORD,C.EOS_WORD])
tags_iobes_vocab.add_specials([C.BOS_WORD,C.PAD_WORD])
# load ner dataset splits
#train_file = os.path.join(args.data, 'ner.train.pth')
#if os.path.isfile(train_file):
# train_dataset = torch.load(train_file)
#else:
train_dataset = EC_PA_Datset(train_dir, vocab, tags_vocab, tags_iobes_vocab)
#torch.save(train_dataset, train_file)
logger.debug('==> Size of train data : %d ' % len(train_dataset))
#test_file = os.path.join(args.data, 'ner.test.pth')
# if os.path.isfile(test_file):
# test_dataset = torch.load(test_file)
# else:
test_dataset = EC_PA_Datset(test_dir, vocab, tags_vocab, tags_iobes_vocab)
#torch.save(test_dataset, test_file)
logger.debug('==> Size of test data : %d ' % len(test_dataset))
#dev_file = os.path.join(args.data, 'ner.dev.pth')
# if os.path.isfile(dev_file):
# dev_dataset = torch.load(dev_file)
# else:
dev_dataset = EC_PA_Datset(dev_dir, vocab, tags_vocab, tags_iobes_vocab)
#torch.save(dev_dataset, dev_file)
logger.debug('==> Size of dev data : %d ' % len(dev_dataset))
ds_pa_file = os.path.join(args.data, 'ner.ds_pa.pth')
#if os.path.isfile(ds_pa_file):
# ds_pa_dataset = torch.load(ds_pa_file)
#else:
ds_pa_dataset = EC_PA_Datset(ds_pa_dir, vocab, tags_vocab, tags_iobes_vocab, partial= True if 'PA' in args.mode else False)
#torch.save(ds_pa_dataset, ds_pa_file)
logger.debug('==> Size of ds pa data : %d ' % len(ds_pa_dataset))
#merge_file = os.path.join(args.data, 'ner.merge.pth')
#if os.path.isfile(merge_file):
# merge_dataset = torch.load(merge_file)
#else:
merge_dataset = ds_pa_dataset.merge(train_dataset)
#torch.save(merge_dataset, merge_file)
logger.debug('==> Size of merge data : %d ' % len(merge_dataset))
if args.iobes:
tags_vocab=tags_iobes_vocab
tagger_model = BiLSTM_CRF_PA_CRF_LightNER(
vocab=vocab.label_to_idx,
emb_size=args.word_size,
tags_vocab=tags_vocab.label_to_idx,
freeze_embed=True,
biLSTM_hidden_size=args.biLSTM_hidden_size,
dropout=args.dropout)
sl_model= policy_selector(2*args.biLSTM_hidden_size+args.max_len, args.SL_hidden_size)
tagger_model.apply(weight_init)
sl_model.apply(weight_init)
if args.cuda:
tagger_model=tagger_model.to('cuda')
sl_model=sl_model.to('cuda')
'''
if torch.cuda.device_count() > 1:
print("Let's use", torch.cuda.device_count(), "GPUs!")
model = nn.DataParallel(model)
'''
optimizer_tagger=optim.RMSprop(filter(lambda p: p.requires_grad, tagger_model.parameters()), lr=args.lr,
weight_decay=args.weight_decay)
criterion_sl = torch.nn.BCELoss(reduction='none')
if args.optim == 'adam':
optimizer_sl = optim.Adam(filter(lambda p: p.requires_grad, sl_model.parameters()), lr=args.lr)
elif args.optim == 'adagrad':
optimizer_sl = optim.Adagrad(filter(lambda p: p.requires_grad, sl_model.parameters()), lr=args.lr)
elif args.optim == 'sgd':
optimizer_sl = optim.SGD(filter(lambda p: p.requires_grad, sl_model.parameters()), lr=args.lr,
weight_decay=args.weight_decay)
# word embedding
emb_file = os.path.join(args.data, 'ner.embed.pth')
if os.path.isfile(emb_file):
emb = torch.load(emb_file)
else:
# load glove embeddings and vocab
glove_vocab, glove_emb = load_word_vectors_EC(os.path.join(args.glove, 'pre_trained_100dim.model'))
logger.debug('==> EMBEDDINGS vocabulary size: %d ' % glove_vocab.size())
emb = torch.Tensor(vocab.size(), glove_emb.size(1)).uniform_(-0.05, 0.05)
# zero out the embeddings for padding and other special words
for idx, item in enumerate([C.PAD_WORD, C.UNK_WORD]):
emb[idx].zero_()
for word in vocab.label_to_idx.keys():
if glove_vocab.get_index(word):
emb[vocab.get_index(word)] = glove_emb[glove_vocab.get_index(word)]
torch.save(emb, emb_file)
if args.cuda:
emb = emb.cuda()
tagger_model.embeddings.weight.data.copy_(emb)
if args.mode=='PA+SL':
# if partial = False it will apply only selection with normal crf
trainer = Trainer_PA_SL_DSNER(args, tagger_model, sl_model, optimizer_tagger, optimizer_sl, criterion_sl, partial=True)
else:
# the result of partial = False or partial =True should be same because we do not include the Partial annotation
trainer = Trainer_PA(args, tagger_model, optimizer_tagger, partial=False)
best = -float('inf')
# dataset:
if args.setup=='A+H':
dataset_setup= merge_dataset
else:
dataset_setup = train_dataset
#print(tagger_model)
f1s={}
for epoch in range(args.epochs):
if args.mode == 'PA+SL':
train_loss = trainer.train(dataset_setup, epoch)
train_f1 = 'NAN'
else:
train_loader = torch.utils.data.DataLoader(dataset_setup, batch_size=args.batch_size, shuffle=True,
num_workers =1, pin_memory=False)
_=trainer.train(train_loader,epoch)
train_loader_test = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size, shuffle=False,
num_workers=1, pin_memory=False)
train_loss, train_pred ,actual_tags = trainer.test(train_loader_test)
actual_tags = [[int(t) for t in tags.cpu().numpy()] for tags in actual_tags]
train_pred = [[int(t) for t in tags.cpu().numpy()] for tags in train_pred]
# using conlleval.py
actual_tags_2_label, train_pred_2_label=tags_to_labels(actual_tags, train_pred, tags_vocab.idx_to_label,
args.iobes)
prec, rec, train_f1 = evaluate(actual_tags_2_label,train_pred_2_label, verbose=False)
train_f1 = round(train_f1, 2)
print('==> Epoch {}, Train \tLoss: {:.2f}\tf1: {}'.format(epoch + 1, train_loss, train_f1))
if args.cuda:
torch.cuda.empty_cache()
dev_loader = torch.utils.data.DataLoader(dev_dataset, batch_size=args.batch_size, shuffle=False,
num_workers=1, pin_memory=False)
dev_loss, dev_pred, dev_actual_tags = trainer.test(dev_loader)
dev_actual_tags = [[int(t) for t in tags.cpu().numpy()] for tags in dev_actual_tags]
dev_pred = [[int(t) for t in tags.cpu()] for tags in dev_pred]
dev_actual_tags_2_label, dev_train_pred_2_label = tags_to_labels(dev_actual_tags, dev_pred,
tags_vocab.idx_to_label,
args.iobes)
prec, rec, dev_f1 = evaluate(dev_actual_tags_2_label, dev_train_pred_2_label, verbose=False)
print('==> Epoch {}, dev \tLoss: {:.2f}\tf1: {:.2f}'.format(epoch + 1, dev_loss, dev_f1))
if best < dev_f1:
best = dev_f1
test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False,
num_workers=1, pin_memory=False)
test_loss, test_pred, test_actual_tags = trainer.test(test_loader)
test_actual_tags = [[int(t) for t in tags.cpu().numpy()] for tags in test_actual_tags]
test_pred = [[int(t) for t in tags.cpu()] for tags in test_pred]
test_actual_tags_2_label, test_train_pred_2_label = tags_to_labels(test_actual_tags, test_pred,
tags_vocab.idx_to_label,
args.iobes)
prec, rec, test_f1 = evaluate(test_actual_tags_2_label, test_train_pred_2_label, verbose=False)
print('==> Epoch {}, Test \tLoss: {:.2f}\tf1: {:.2f}'.format(epoch + 1, test_loss, test_f1))
checkpoint = {
'model_tagger': trainer.tagger_model.state_dict(),
'optim_tagger': trainer.optimizer_tagger,
'f1': test_f1,
'f1-tags': test_f1,
'predict': test_pred,
'args': args,
'epoch': epoch + 1
}
dataset = 'ER'
filename = 'ner-{}-{}-{}-{}-{}'.format(dataset, args.mode,args.setup,
args.batch_size,
args.seed)
logger.debug('==> New optimum found, checkpointing everything now...')
torch.save(checkpoint, '%s.pt' % os.path.join(args.save, filename))
train_f1 = train_f1 if train_f1 == 'NAN' else round(train_f1, 2)
f1s[epoch] = {'T': [train_f1], 'D': [round(dev_f1,2)], 'E': [round(test_f1,2)]}
print(f1s)
out_name = 'ner-{}-{}-{}'.format(dataset, args.mode, args.setup)
output_file = os.path.join(args.data, out_name)
write_output(output_file,f1s)
if __name__ == "__main__":
main()