-
Notifications
You must be signed in to change notification settings - Fork 0
/
KMBERT_ner.py
383 lines (321 loc) · 15 KB
/
KMBERT_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
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
from pretrain.examples.extract_features import *
from pytorch_pretrained_bert.tokenization import BertTokenizer
from pytorch_pretrained_bert.modeling import BertForTokenClassification
from keras.preprocessing.sequence import pad_sequences
from sklearn import metrics
from torch.utils.data import TensorDataset, DataLoader, RandomSampler
from torch.nn import CrossEntropyLoss
from kumc_dataset import InputDataset
from torch.optim import Adam
from tqdm import tqdm
from torch import nn
import numpy as np
import pandas as pd
import scipy.stats
import pickle, re
import torch
import json
import os
import csv
import argparse
def to_np(t):
return t.cpu().detach().numpy()
parser = argparse.ArgumentParser(description='KUMC NER')
parser.add_argument('--pretrained', type=str)
args = parser.parse_args()
pretrained_model = args.pretrained
tag_dict = dict()
tag_dict['B-BD'] = 0
tag_dict['I-BD'] = 0
tag_dict['B-DS'] = 1
tag_dict['I-DS'] = 1
tag_dict['B-ST'] = 2
tag_dict['I-ST'] = 2
tag_dict['O'] = 3
raw_tag_dict = dict()
raw_tag_dict['B-BD'] = 0
raw_tag_dict['I-BD'] = 4
raw_tag_dict['B-DS'] = 1
raw_tag_dict['I-DS'] = 5
raw_tag_dict['B-ST'] = 2
raw_tag_dict['I-ST'] = 6
raw_tag_dict['O'] = 3
# HYPERPARAMETERS
device = 'cpu' # gpu usage will be updated
epoch = 5
learning_rate = 2e-5
batch_size = 32
seq_len = 128
num_label = 4
savepath = "res_ner/"
if pretrained_model=='kmbert':
tokenizer = BertTokenizer(vocab_file="./kmbert/kmbert_vocab.txt", do_lower_case=False)
model = BertForTokenClassification.from_pretrained('./kmbert/', num_labels=num_label)
savepath += "kmbert/"
if pretrained_model=='kmbert_vocab':
tokenizer = BertTokenizer(vocab_file="./kmbert_vocab/kmbert_vocab.txt", do_lower_case=False)
model = BertForTokenClassification.from_pretrained('./kmbert_vocab/', num_labels=num_label)
savepath += "kmbert_vocab/"
if not os.path.isdir(savepath):
os.mkdir(savepath)
model.to(device)
loss_fct = CrossEntropyLoss(ignore_index=-1).to(device)
def read_ner_dataset(filename):
reader = open(filename, 'r', encoding='utf-8')
data = []
for line in reader:
data.append(line)
ner_ori = []
for i, v in enumerate(data):
if v == '\n':
ner_ori.append(['-', '-', '-'])
elif '-DOCSTART-' in v:
continue
else:
ner_ori.append(v.strip().split(' '))
ner_ori = np.array(ner_ori)
sentence, tag, raw_tag = [], [], []
tmp_sent, tmp_tag, tmp_raw_tag = [], [], []
for i, v in enumerate(ner_ori):
if v[0] == '-' and v[1] == '-' and v[2] == '-':
if tmp_sent or tmp_tag:
sentence.append(tmp_sent)
tag.append(tmp_tag)
raw_tag.append(tmp_raw_tag)
tmp_sent, tmp_tag, tmp_raw_tag = [], [], []
continue
tokens = tokenizer.tokenize(v[0])
token_ids = tokenizer.convert_tokens_to_ids(tokens)
tmp_sent.extend(token_ids)
tmp_tag.extend(list(np.repeat(tag_dict[v[2]], len(tokens))))
tmp_raw_tag.extend(list(np.repeat(raw_tag_dict[v[2]], len(tokens))))
return sentence, tag, raw_tag
def get_matched(pred, label, ignore_idx=-1):
idx = label!=ignore_idx
matched = ((pred==label)[idx]).sum()
target = idx.sum()
return matched, target
savepath_ = savepath
for fold in range(5):
filename_train = 'Path for NER train dataset %d (5-fold cross validation)'%(fold+1)
filename_test = 'Path for NER test dataset %d (5-fold cross validation)'%(fold+1)
savepath = savepath_ + "%d/"%(fold+1)
if not os.path.isdir(savepath):
os.mkdir(savepath)
if pretrained_model=='kmbert':
model = BertForTokenClassification.from_pretrained('./kmbert/', num_labels=num_label)
if pretrained_model=='kmbert_vocab':
model = BertForTokenClassification.from_pretrained('./kmbert_vocab/', num_labels=num_label)
model.to(device)
print("======= Fold %d ========" % (fold+1))
sent_train, tag_train, raw_tag_train = read_ner_dataset(filename_train)
sent_test, tag_test, raw_tag_test = read_ner_dataset(filename_test)
# DATA
token_cls = tokenizer.convert_tokens_to_ids(['[CLS]'])[0]
token_sep = tokenizer.convert_tokens_to_ids(['[SEP]'])[0]
input_ids_train = pad_sequences(sent_train, maxlen=seq_len, dtype="long", truncating="post", padding="post")
attention_masks_train = np.array([[float(i > 0) for i in ii] for ii in input_ids_train])
labels_train = pad_sequences(tag_train, maxlen=seq_len, dtype="long", truncating="post", padding="post", value=-1)
labels_raw_train = pad_sequences(raw_tag_train, maxlen=seq_len, dtype="long", truncating="post", padding="post", value=-1)
input_ids_test = pad_sequences(sent_test, maxlen=seq_len, dtype="long", truncating="post", padding="post")
attention_masks_test = np.array([[float(i > 0) for i in ii] for ii in input_ids_test])
labels_test = pad_sequences(tag_test, maxlen=seq_len, dtype="long", truncating="post", padding="post", value=-1)
labels_raw_test = pad_sequences(raw_tag_test, maxlen=seq_len, dtype="long", truncating="post", padding="post", value=-1)
# DATASET & DATALOADER
input_ids_train = torch.tensor(input_ids_train)
attention_masks_train = torch.tensor(attention_masks_train)
labels_train = torch.tensor(labels_train)
labels_raw_train = torch.tensor(labels_raw_train)
train_data = TensorDataset(input_ids_train, attention_masks_train, labels_train, labels_raw_train)
train_sampler = RandomSampler(train_data)
train_dataloader = DataLoader(train_data, sampler=train_sampler, batch_size=batch_size)
input_ids_test = torch.tensor(input_ids_test)
attention_masks_test = torch.tensor(attention_masks_test)
labels_test = torch.tensor(labels_test)
labels_raw_test = torch.tensor(labels_raw_test)
test_data = TensorDataset(input_ids_test, attention_masks_test, labels_test, labels_raw_test)
test_dataloader = DataLoader(test_data, batch_size=batch_size)
# Fine-tuning hyper-parameter
FULL_FINETUNING = True
if FULL_FINETUNING:
param_optimizer = list(model.named_parameters())
no_decay = ['bias', 'gamma', 'beta']
optimizer_grouped_parameters = [
{'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)],
'weight_decay_rate': 0.01},
{'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)],
'weight_decay_rate': 0.0}
]
else:
param_optimizer = list(model.classifier.named_parameters())
optimizer_grouped_parameters = [{"params": [p for n, p in param_optimizer]}]
optim = Adam(optimizer_grouped_parameters, lr=learning_rate)
# optim = Adam(model.parameters(), lr=learning_rate)
out_file_train = open(savepath + 'train_'+str(batch_size)+'_'+str(learning_rate)+'.txt', 'wt', encoding='utf-8', newline='')
tsv_writer_train = csv.writer(out_file_train, delimiter='\t')
tsv_writer_train.writerow(['Epoch', 'Loss', 'Unit Accuracy'])
out_file_train.flush()
out_file_test = open(savepath + 'test_'+str(batch_size)+'_'+str(learning_rate)+'.txt', 'wt', encoding='utf-8', newline='')
tsv_writer_test = csv.writer(out_file_test, delimiter='\t')
# tsv_writer_test.writerow(['Epoch',
# 'Avg. precision', 'Avg. recall', 'Avg. F1',
# 'BD precision', 'BD recall', 'BD F1',
# 'DS precision', 'DS recall', 'DS F1',
# 'ST precision', 'ST recall', 'ST F1',
# 'Loss', 'Unit Accuracy', 'Micro Accuracy'])
tsv_writer_test.writerow(['Epoch', 'Micro F1',
'Avg. precision', 'Avg. recall', 'Avg. F1',
'BBD precision', 'BBD recall', 'BBD F1',
'BDS precision', 'BDS recall', 'BDS F1',
'BST precision', 'BST recall', 'BST F1',
# 'IBD precision', 'IBD recall', 'IBD F1',
# 'IDS precision', 'IDS recall', 'IDS F1',
# 'IST precision', 'IST recall', 'IST F1',
'Loss', 'Unit Accuracy', 'Micro Accuracy'])
out_file_test.flush()
for e in range(epoch):
str_code = "TRAIN"
loss_sum = 0.0
matched_sum = 0.0
N_target = 0.0
data_iter = tqdm(enumerate(train_dataloader),
desc="Epoch_%s:%d" % (str_code, e+1),
total=len(train_dataloader),
bar_format="{l_bar}{r_bar}")
# Training loop
model.train()
for i, data in data_iter:
data = [x.to(device) for x in data]
pred = model(input_ids=data[0],
attention_mask=data[1],
token_type_ids=None)
# CrossEntropy Loss
loss = loss_fct(pred.view(-1, num_label), data[2].view(-1))
_, ner_pred = torch.max(pred, 2)
# matched
ner_loss = loss.item()
loss_sum += ner_loss*data[0].shape[0]
matched, target = get_matched(to_np(ner_pred), to_np(data[2]))
matched_sum += matched
N_target += target
optim.zero_grad()
loss.backward()
optim.step()
# print("\nLoss: %.4f / Accuracy: %.2f" % (ner_loss, matched/target))
del data
loss_t = loss_sum/input_ids_train.shape[0]
accuracy = matched_sum/N_target
tsv_writer_train.writerow([e+1, loss_t, accuracy])
out_file_train.flush()
#########################################################
str_code = "TEST"
loss_sum = 0.0
matched_sum = 0.0
N_target = 0.0
t_matched = 0.0
N_count = 0.0
res_ner = []
data_iter = tqdm(enumerate(test_dataloader),
desc="Epoch_%s:%d" % (str_code, e + 1),
total=len(test_dataloader),
bar_format="{l_bar}{r_bar}")
out_file_res = open(savepath + 'res_' + str(batch_size) + '_' + str(learning_rate) + '_' + str(e+1) + '.txt', 'wt',
encoding='utf-8', newline='')
tsv_writer_res = csv.writer(out_file_res, delimiter='\t')
tsv_writer_res.writerow(['Text', 'Label', 'Pred'])
# Test loop
model.eval()
for _, data in data_iter:
data = [x.to(device) for x in data]
pred = model(input_ids=data[0],
attention_mask=data[1],
token_type_ids=None)
# CrossEntropy Loss
loss = loss_fct(pred.view(-1, num_label), data[2].view(-1))
_, ner_pred = torch.max(pred, 2)
# matched
ner_loss = loss.item()
loss_sum += ner_loss * data[0].shape[0]
matched, target = get_matched(to_np(ner_pred), to_np(data[2]))
matched_sum += matched
N_target += target
t_ids = to_np(data[0])
t_pred = to_np(pred)
t_label = to_np(data[2])
t_raw_label = to_np(data[3])
for i, v in enumerate(t_ids):
txt = tokenizer.convert_ids_to_tokens(v)
pred_sample = np.zeros(len(tag_dict))
tmp_txt = ''
tmp_label = -1
tmp_raw_label = -1
for j, u in enumerate(txt):
if u=='[PAD]':
tsv_writer_res.writerow([])
break
if u.startswith('##'):
pred_sample += t_pred[i][j]
tmp_txt += u[2:]
if not u.startswith("##"):
tmp_txt = u
pred_sample = t_pred[i][j]
tmp_label = t_label[i][j]
tmp_raw_label = t_raw_label[i][j]
N_count += 1.0
if j==(seq_len-1):
pred_label = pred_sample.argmax()
true_label = tmp_label
tsv_writer_res.writerow([tmp_txt, true_label, pred_label, tmp_raw_label])
res_ner.append([true_label, pred_label])
if pred_label==true_label:
t_matched += 1.0
tsv_writer_res.writerow([])
break
if not txt[j+1].startswith('##'):
pred_label = pred_sample.argmax()
true_label = tmp_label
tsv_writer_res.writerow([tmp_txt, true_label, pred_label, tmp_raw_label])
res_ner.append([true_label, pred_label])
if pred_label==true_label:
t_matched += 1.0
out_file_res.flush()
# print("\nLoss: %.4f / Unit Accuracy: %.2f" % (ner_loss, matched/target))
del data
out_file_res.close()
res_ner = np.array(res_ner)
performance = np.zeros((3, 3))
perf_matched = res_ner[:, 0]==res_ner[:, 1]
for kk in range(3):
precision_idx = (res_ner[:, 1] == kk)
recall_idx = (res_ner[:, 0] == kk)
if precision_idx.sum()==0:
performance[kk, 0] = 0
else:
performance[kk, 0] = (perf_matched[precision_idx].sum()) / (precision_idx.sum())
if recall_idx.sum()==0:
performance[kk, 1] = 0
else:
performance[kk, 1] = (perf_matched[recall_idx].sum()) / (recall_idx.sum())
if (performance[kk, 0]==0) and (performance[kk, 1]==0):
performance[kk, 2] = 0
else:
performance[kk, 2] = 2 * (performance[kk, 0] * performance[kk, 1]) / (performance[kk, 0] + performance[kk, 1])
loss_t = loss_sum/input_ids_test.shape[0]
unit_accuracy = matched_sum/N_target
micro_accuracy = t_matched/N_count
confusion = metrics.confusion_matrix(res_ner[:, 0], res_ner[:, 1])
tsv_writer_test.writerow([e + 1,
(confusion[0,0]+confusion[1,1]+confusion[2,2])/confusion[:3,:3].sum(),
performance[:,0].mean(),
performance[:,1].mean(),
performance[:,2].mean(),
performance[0,0], performance[0,1], performance[0,2],
performance[1,0], performance[1,1], performance[1,2],
performance[2,0], performance[2,1], performance[2,2],
# performance[3, 0], performance[3, 1], performance[3, 2],
# performance[4, 0], performance[4, 1], performance[4, 2],
# performance[5, 0], performance[5, 1], performance[5, 2],
loss_t, accuracy, micro_accuracy])
out_file_test.flush()
out_file_train.close()
out_file_test.close()