-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainer.py
582 lines (493 loc) · 17.3 KB
/
trainer.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
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
import re
import torch
from transformers import BertTokenizer, AutoModelForTokenClassification, TrainingArguments, Trainer, EvalPrediction
from sklearn.metrics import precision_recall_fscore_support
from tqdm import tqdm
from datasets import Dataset, DatasetDict
device = torch.device('cuda' if torch.cuda.is_available else 'cpu')
print("Device: ", device)
# Input Files:
train_corpus = "corpus/train-set.txt"
val_corpus = "corpus/eval-set.txt"
bert_model = "gklmip/bert-tagalog-base-uncased"
tokenizer = BertTokenizer.from_pretrained(bert_model)
# print(tokenizer.additional_special_tokens_ids)
num_added_toks = tokenizer.add_special_tokens({'additional_special_tokens': ['[PMP]', '[PMS]', '[PMC]']})
# special_tokens = ['[PMP]', '[PMS]', '[PMC]']
# tokenizer.add_tokens(special_tokens, special_tokens=True)
# print("[PMP] token ID:", tokenizer.convert_tokens_to_ids('[PMP]'))
# print("[PMS] token ID:", tokenizer.convert_tokens_to_ids('[PMS]'))
# print("[PMC] token ID:", tokenizer.convert_tokens_to_ids('[PMC]'))
pos_tag_mapping = {
'[PAD]': 0,
'NNC': 1,
'NNP': 2,
'NNPA': 3,
'NNCA': 4,
'PR': 5,
'PRS': 6,
'PRP': 7,
'PRSP': 8,
'PRO': 9,
'PRQ': 10,
'PRQP': 11,
'PRL': 12,
'PRC': 13,
'PRF': 14,
'PRI': 15,
'DT': 16,
'DTC': 17,
'DTP': 18,
'DTPP': 19,
'LM': 20,
'CC': 21,
'CCT': 22,
'CCR': 23,
'CCB': 24,
'CCA': 25,
'PM': 26,
'PMP': 27,
'PME': 28,
'PMQ': 29,
'PMC': 30,
'PMSC': 31,
'PMS': 32,
'VB': 33,
'VBW': 34,
'VBS': 35,
'VBN': 36,
'VBTS': 37,
'VBTR': 38,
'VBTF': 39,
'VBTP': 40,
'VBAF': 41,
'VBOF': 42,
'VBOB': 43,
'VBOL': 44,
'VBOI': 45,
'VBRF': 46,
'JJ': 47,
'JJD': 48,
'JJC': 49,
'JJCC': 50,
'JJCS': 51,
'JJCN': 52,
'JJCF': 53,
'JJCB': 54,
'JJT': 55,
'RB': 56,
'RBD': 57,
'RBN': 58,
'RBK': 59,
'RBP': 60,
'RBB': 61,
'RBR': 62,
'RBQ': 63,
'RBT': 64,
'RBF': 65,
'RBW': 66,
'RBM': 67,
'RBL': 68,
'RBI': 69,
'RBS': 70,
'RBJ': 71,
'RBY': 72,
'RBLI': 73,
'TS': 74,
'FW': 75,
'CD': 76,
'CCB_CCP': 77,
'CCR_CCA': 78,
'CCR_CCB': 79,
'CCR_CCP': 80,
'CCR_LM': 81,
'CCT_CCA': 82,
'CCT_CCP': 83,
'CCT_LM': 84,
'CCU_DTP': 85,
'CDB_CCA': 86,
'CDB_CCP': 87,
'CDB_LM': 88,
'CDB_NNC': 89,
'CDB_NNC_CCP': 90,
'JJCC_CCP': 91,
'JJCC_JJD': 92,
'JJCN_CCP': 93,
'JJCN_LM': 94,
'JJCS_CCB': 95,
'JJCS_CCP': 96,
'JJCS_JJC': 97,
'JJCS_JJC_CCP': 98,
'JJCS_JJD': 99,
'[UNK]': 100,
'[CLS]': 101,
'[SEP]': 102,
'JJCS_JJN': 103,
'JJCS_JJN_CCP': 104,
'JJCS_RBF': 105,
'JJCS_VBAF': 106,
'JJCS_VBAF_CCP': 107,
'JJCS_VBN_CCP': 108,
'JJCS_VBOF': 109,
'JJCS_VBOF_CCP': 110,
'JJCS_VBN': 111,
'RBQ_CCP': 112,
'JJC_CCB': 113,
'JJC_CCP': 114,
'JJC_PRL': 115,
'JJD_CCA': 116,
'JJD_CCB': 117,
'JJD_CCP': 118,
'JJD_CCT': 119,
'JJD_NNC': 120,
'JJD_NNP': 121,
'JJN_CCA': 122,
'JJN_CCB': 123,
'JJN_CCP': 124,
'JJN_NNC': 125,
'JJN_NNC_CCP': 126,
'JJD_NNC_CCP': 127,
'NNC_CCA': 128,
'NNC_CCB': 129,
'NNC_CCP': 130,
'NNC_NNC_CCP': 131,
'NN': 132,
'JJN': 133,
'NNP_CCA': 134,
'NNP_CCP': 135,
'NNP_NNP': 136,
'PRC_CCB': 137,
'PRC_CCP': 138,
'PRF_CCP': 139,
'PRQ_CCP': 140,
'PRQ_LM': 141,
'PRS_CCB': 142,
'PRS_CCP': 143,
'PRSP_CCP': 144,
'PRSP_CCP_NNP': 145,
'PRL_CCP': 146,
'PRL_LM': 147,
'PRO_CCB': 148,
'PRO_CCP': 149,
'VBS_CCP': 150,
'VBTR_CCP': 151,
'VBTS_CCA': 152,
'VBTS_CCP': 153,
'VBTS_JJD': 154,
'VBTS_LM': 155,
'VBAF_CCP': 156,
'VBOB_CCP': 157,
'VBOF_CCP': 158,
'VBOF_CCP_NNP': 159,
'VBRF_CCP': 160,
'CCP': 161,
'CDB': 162,
'RBW_CCP': 163,
'RBD_CCP': 164,
'DTCP': 165,
'VBH': 166,
'VBTS_VBOF': 167,
'PRI_CCP': 168,
'VBTR_VBAF_CCP': 169,
'DQL': 170,
'DQR': 171,
'RBT_CCP': 172,
'VBW_CCP': 173,
'RBI_CCP': 174,
'VBN_CCP': 175,
'VBTR_VBAF': 176,
'VBTF_CCP': 177,
'JJCS_JJD_NNC': 178,
'CCU': 179,
'RBL_CCP': 180,
'VBTR_VBRF_CCP': 181,
'PRP_CCP': 182,
'VBTR_VBRF': 183,
'VBH_CCP': 184,
'VBTS_VBAF': 185,
'VBTF_VBOF': 186,
'VBTR_VBOF': 187,
'VBTF_VBAF': 188,
'JJCS_JJD_CCB': 189,
'JJCS_JJD_CCP': 190,
'RBM_CCP': 191,
'NNCS': 192,
'PRI_CCB': 193,
'NNA': 194,
'VBTR_VBOB': 195,
'DC': 196,
'JJD_CP': 197,
'NC': 198,
'NC_CCP': 199,
'VBO': 200,
'JJD_CC': 201,
'VBF': 202,
'CP': 203,
'NP': 204,
'N': 205,
'F': 206,
'CT': 207,
'MS': 208,
'BTF': 209,
'CA': 210,
'VBOF_RBR': 211,
'DP': 212,
}
num_labels = len(pos_tag_mapping)
id2label = {idx: tag for tag, idx in pos_tag_mapping.items()}
label2id = {tag: idx for tag, idx in pos_tag_mapping.items()}
def symbol2token(symbol):
special_symbols = ['-', '&', "\"", "[", "]", "/", "$", "(", ")", "%", ":", "'", '.']
# Check if the symbol is a comma
if symbol == ',':
return '[PMC] '
elif symbol == '.':
return '[PMP] '
# Check if the symbol is in the list of special symbols
elif symbol in special_symbols:
return '[PMS] '
# If the symbol is not a comma or in the special symbols list, keep it as it is
return symbol
def preprocess_sentence(tagged_sentence):
# Remove the line identifier (e.g., SNT.80188.3)
sentence = re.sub(r'SNT\.\d+\.\d+\s+', '', tagged_sentence)
special_symbols = ['-', '&', ",", "\"", "[", "]", "/", "$", "(", ")", "%", ":", "'", '.']
# Construct the regex pattern for extracting words inside <TAGS> including special symbols
special_symbols_regex = '|'.join([re.escape(sym) for sym in special_symbols])
regex_pattern = r'<(?:[^<>]+? )?([a-zA-Z0-9.,&"!?{}]+)>'.format(special_symbols_regex)
words = re.findall(regex_pattern, tagged_sentence)
# Join the words to form a sentence
sentence = ' '.join(words)
sentence = sentence.lower()
# print("---")
# print("Sentence before:", sentence)
# Loop through the sentence and convert hyphen to '[PMP]' if the next character is a space
new_sentence = ""
i = 0
# print("Length: ", len(sentence))
while i < len(sentence):
# print(f"{i+1} == {len(sentence)}: {sentence[i]}")
if any(sentence[i:].startswith(symbol) for symbol in special_symbols):
if i + 2 < len(sentence) and sentence[i:i + 3] == '...':
# Ellipsis found, replace with '[PMS]'
new_sentence += symbol2token(sentence[i])
i += 3
elif i + 1 == len(sentence):
new_sentence += symbol2token(sentence[i])
break
elif sentence[i + 1] == ' ' and i == 0:
new_sentence += symbol2token(sentence[i])
i += 1
elif sentence[i - 1] == ' ' and sentence[i + 1] == ' ':
new_sentence += symbol2token(sentence[i])
i += 1
elif sentence[i - 1] != ' ':
new_sentence += ''
else:
word_after_symbol = ""
while i + 1 < len(sentence) and sentence[i + 1] != ' ' and not any(
sentence[i + 1:].startswith(symbol) for symbol in special_symbols):
word_after_symbol += sentence[i + 1]
i += 1
new_sentence += word_after_symbol
elif any(sentence[i:].startswith(symbol) for symbol in special_symbols):
if i + 1 < len(sentence) and (sentence[i + 1] == ' ' and sentence[i - 1] != ' '):
new_sentence += '[PMS] '
i += 1
elif i + 1 == len(sentence):
new_sentence += '[PMS] '
break
else:
word_after_symbol = ""
while i + 1 < len(sentence) and sentence[i + 1] != ' ' and not any(
sentence[i + 1:].startswith(symbol) for symbol in special_symbols):
word_after_symbol += sentence[i + 1]
i += 1
new_sentence += word_after_symbol
else:
new_sentence += sentence[i]
i += 1
# print("Sentence after:", new_sentence)
# print("---")
return new_sentence
def extract_tags(input_sentence):
tags = re.findall(r'<([A-Z_]+)\s.*?>', input_sentence)
return tags
def align_tokenization(sentence, tags):
print("Sentence \n: ", sentence)
sentence = sentence.split()
print("Sentence Split\n: ", sentence)
tokenized_sentence = tokenizer.tokenize(' '.join(sentence))
tokenized_sentence_string = " ".join(tokenized_sentence)
print("ID2Token_string\n: ", tokenized_sentence_string)
print("Tags\n: ", [id2label[tag_id] for tag_id in tags])
if len(tags) > 12:
print(id2label[tags[11]])
aligned_tagging = []
current_word = ''
index = 0
for token in tokenized_sentence:
if len(tags) > index:
current_word += re.sub(r'^##', '', token)
# print("Current word after replacing ##: ", current_word)
# print("sentence[index]: ", sentence[index])
if sentence[index] == current_word: # if we completed a word
print("completed a word: ", current_word)
current_word = ''
aligned_tagging.append(tags[index])
# print(f"Tag of index {index}: ", id2label[tags[index]])
# print(f"Aligned tag of index {index}: ", (id2label[aligned_tagging[-1]]))
# print("Tags1\n: ", [id2label[tag_id] for tag_id in tags])
# print("Tags2\n: ", [id2label[tag_id] for tag_id in aligned_tagging])
# print(f"{index+1}/{len(tags)} tags consumed")
index += 1
else: # otherwise insert padding
print("incomplete word: ", current_word)
aligned_tagging.append(0)
print("---")
decoded_tags = [list(pos_tag_mapping.keys())[list(pos_tag_mapping.values()).index(tag_id)] for tag_id in
aligned_tagging]
# print("Tokenized Sentence\n: ", tokenized_sentence)
# print("Tokenized Len\n: ", len(tokenized_sentence))
# print("Tags\n: ", decoded_tags)
# print("Tags Count\n: ", len(decoded_tags))
assert len(tokenized_sentence) == len(aligned_tagging)
aligned_tagging = [0] + aligned_tagging
return tokenized_sentence, aligned_tagging
def process_tagged_sentence(tagged_sentence):
# print(tagged_sentence)
sentence = preprocess_sentence(tagged_sentence)
tags = extract_tags(tagged_sentence) # returns the tags (eto ilagay mo sa tags.txt)
encoded_tags = [pos_tag_mapping[tag] for tag in tags]
# Align tokens
tokenized_sentence, encoded_tags = align_tokenization(sentence, encoded_tags)
encoded_sentence = tokenizer(sentence, padding="max_length" ,truncation=True, max_length=128)
# Create attention mask (1 for real tokens, 0 for padding)
attention_mask = [1] * len(encoded_sentence['input_ids'])
print("len(encoded_sentence['input_ids']):", len(encoded_sentence['input_ids']))
while len(encoded_sentence['input_ids']) < 128:
encoded_sentence['input_ids'].append(0) # Pad with zeros
attention_mask.append(0) # Pad attention mask
while len(encoded_tags) < 128:
encoded_tags.append(0) # Pad with the ID of '[PAD]'
encoded_sentence['encoded_tags'] = encoded_tags
decoded_sentence = tokenizer.convert_ids_to_tokens(encoded_sentence['input_ids'], skip_special_tokens=False)
decoded_tags = [list(pos_tag_mapping.keys())[list(pos_tag_mapping.values()).index(tag_id)] for tag_id in
encoded_tags]
#
word_tag_pairs = list(zip(decoded_sentence, decoded_tags))
print(encoded_sentence)
print("Sentence:", decoded_sentence)
print("Tags:", decoded_tags)
print("Decoded Sentence and Tags:", word_tag_pairs)
print("---")
return encoded_sentence
def encode_corpus(input_file):
encoded_sentences = []
with open(input_file, 'r') as f:
lines = f.readlines()
# int = 1
for line in tqdm(lines, desc="Processing corpus"):
# print(int)
# int += 1
input_sentence = line.strip()
# print(input_sentence)
encoded_sentence = process_tagged_sentence(input_sentence)
encoded_sentences.append(encoded_sentence)
return encoded_sentences
def createDataset(train_set, val_set, test_set=None):
train_dataset_dict = {
'input_ids': [],
'attention_mask': [],
'labels': [],
}
for entry in tqdm(train_set, desc="Converting training set"):
train_dataset_dict['input_ids'].append(entry['input_ids'])
train_dataset_dict['attention_mask'].append(entry['attention_mask'])
train_dataset_dict['labels'].append(entry['encoded_tags'])
train_dataset = Dataset.from_dict(train_dataset_dict)
val_dataset_dict = {
'input_ids': [],
'attention_mask': [],
'labels': [],
}
for entry in tqdm(val_set, desc="Converting validation set"):
val_dataset_dict['input_ids'].append(entry['input_ids'])
val_dataset_dict['attention_mask'].append(entry['attention_mask'])
val_dataset_dict['labels'].append(entry['encoded_tags'])
val_dataset = Dataset.from_dict(val_dataset_dict)
dataset_dict = DatasetDict({
'train': train_dataset,
'validation': val_dataset,
})
if test_set is not None:
test_dataset_dict = {
'input_ids': [],
'attention_mask': [],
'labels': [],
}
for entry in tqdm(test_set, desc="Converting test set"):
test_dataset_dict['input_ids'].append(entry['input_ids'])
test_dataset_dict['attention_mask'].append(entry['attention_mask'])
test_dataset_dict['labels'].append(entry['encoded_tags'])
test_dataset = Dataset.from_dict(test_dataset_dict)
dataset_dict['test'] = test_dataset
print("Dataset created.")
return dataset_dict
test_sentence = [
'SNT.108970.2066 <DTC Ang.> <PRI isa> <CCT sa> <DTCP mga> <NNC susog> <CCP na> <PRO ito> <PMC ,> <DTC ang> <NNP Post-9> <PMS /> <CDB 11> <NNP Batas> <NNP Pangtulong> <CCT sa> <NNP Edukasyon> <CCB ng> <DTCP mga> <NNP Beterano> <CDB 2008> <PMC ,> <LM ay> <RBT_CCP pwedeng> <VBAF magpakita> <CCT bilang> <JJD_CCP modernong> <NNC salin> <CCB ng> <NNC panahon> <CCB ng> <NNP_CCP Ikalawang> <NNP_CCP Digmaang> <JJD pangdaigdig> <PMP .>',
'SNT.206230.256 <VBTS -Sinabi-> <CCB n-g> <NNC tag-apag-salita> <CCT para> <CCT sa> <NNP Winner> <NNP International> <CCT sa> <PRI_CCP isang> <NNC pahayag> <PMC ,> <PMS "> <PMS [> <PRO ito> <LM ay> <PMS ]> <JJCS_JJD napakahirap> <CCP na> <NNC panahon> <CCT para> <CCT sa> <PRSP_CCP aming> <PRI lahat> <CCA at> <VBTR hinihiling> <CCB ng> <NNC pamilya> <CCP na> <VBTF igalang> <PRP ninyo> <DTC ang> <PRSP_CCP kanilang> <NNC pribasya> <PMP .> <PMS ">',
'SNT.187937.383 <VBTS Sinabi> <CCB ng> <JJN pangalawang> <PMS -> <NNC tagapangulo> <CCP na> <DTP si> <NNP Lee> <NNP Cheuk-yan> <CCP na> <DTC ang> <VBOF ginawa> <CCB ng> <NNPA C&ED> <LM ay> <RBF hindi> <JJD pangkaraniwan> <PMC ,> <CCA at> <VBOF tinawagan> <RBI na> <PRS niya> <DTC ang> <NNC departamento> <CCB upang> <VBW makita> <CCR kung> <DTC ang> <NNC departamento> <LM ay> <VBTR pinipilit> <CCP na> <VBW makita> <DTC ang> <DTCP mga> <NNC_CCP kagamitang> <VBH may> <NNC kaugnayan> <CCT sa> <NNC_CCP insidenteng> <VBTS naganap> <CCT sa> <NNP Tiananmen> <NNP Square> <PMP .>'
]
train_corpus = encode_corpus(train_corpus)
val_corpus = encode_corpus(val_corpus)
encoded_dataset = createDataset(train_corpus, val_corpus)
print(encoded_dataset)
max_token_length = 128
vocab_size = tokenizer.vocab_size
encoded_dataset.set_format("torch")
model = AutoModelForTokenClassification.from_pretrained(bert_model,
num_labels=num_labels,
id2label=id2label,
label2id=label2id)
model.resize_token_embeddings(len(tokenizer))
batch_size = 16
metric_name = "f1"
args = TrainingArguments(
"checkpoint",
evaluation_strategy="epoch",
save_strategy="epoch",
learning_rate=3e-4,
per_device_train_batch_size=batch_size,
per_device_eval_batch_size=batch_size,
num_train_epochs=5,
weight_decay=0.01,
load_best_model_at_end=True,
metric_for_best_model=metric_name,
label_names=["labels"],
)
def compute_metrics(p):
y_true = p.label_ids #(sentence[num_of_sentences], words[number_of_words]) (800, 128)
y_pred = p.predictions.argmax(-1)
y_true_flat = [tag_id for tags in y_true for tag_id in tags]
y_pred_flat = [tag_id for tags in y_pred for tag_id in tags]
precision, recall, f1, _ = precision_recall_fscore_support(y_true_flat, y_pred_flat, average="micro")
return {
"precision": precision,
"recall": recall,
"f1": f1,
}
trainer = Trainer(
model,
args,
train_dataset=encoded_dataset["train"],
eval_dataset=encoded_dataset["validation"],
tokenizer=tokenizer,
compute_metrics=compute_metrics
)
training = trainer.train()
print(training)
results = trainer.evaluate()
print("Evaluation: ", results)
trainer.save_model("BERTPOS")
print(results)