-
Notifications
You must be signed in to change notification settings - Fork 3
/
main_bertpho.py
425 lines (324 loc) · 13.7 KB
/
main_bertpho.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
# nohup python -u main_bertpho.py >> main_bertpho.out &
import os
import sys
import sklearn
import numpy as np
import pandas as pd
import re
import xgboost as xgb
import tensorflow as tf
import torch
from sklearn.preprocessing import LabelEncoder
from tensorflow import keras
from tensorflow.keras import layers
import torch
import torch.nn as nn
from torch.nn import CrossEntropyLoss, MSELoss
from torch.utils.data import Dataset
from utils.utils import class_eval, AverageMeter
from utils.utils import check_fields, print_progress, AverageMeter, class_eval, save_checkpoint, \
load_checkpoint
from transformers import RobertaConfig
from transformers import RobertaModel
from transformers import BertTokenizer
from transformers import BertPreTrainedModel
from transformers import BertForSequenceClassification
import seaborn as sns
import matplotlib.pyplot as plt
#%matplotlib inline
import pytz
from datetime import datetime
tz = pytz.timezone('Asia/Saigon')
#date_time = datetime.now(tz).strftime('%Y-%m-%d %H:%M:%S ')
import nltk
#nltk.download('punkt')
#nltk.download('stopwords')
from nltk.tokenize import RegexpTokenizer
tokenizer = RegexpTokenizer(r'\w+')
from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords
stop_words = set(stopwords.words('english'))
from keras.preprocessing.text import Tokenizer
from keras.preprocessing.sequence import pad_sequences
from tqdm import tqdm
os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
import string
string.punctuation.__add__('!!')
string.punctuation.__add__('(')
string.punctuation.__add__(')')
string.punctuation.__add__('?')
string.punctuation.__add__('.')
string.punctuation.__add__(',')
import fairseq
from fairseq.data.encoders.fastbpe import fastBPE
from fairseq.data import Dictionary
from vncorenlp import VnCoreNLP
import argparse
def warn(*args, **kwargs):
pass
import warnings
warnings.warn = warn
def strToBool(str):
return str.lower() in ('true', 'yes', 'on', 't', '1')
parser = argparse.ArgumentParser()
parser.register('type', 'bool', strToBool)
parser.add_argument('--PATH_STAR_RATING_TRAIN', default='dataset/aivivn/train.csv')
parser.add_argument('--PATH_STAR_RATING_TEST', default='dataset/aivivn/test.csv')
parser.add_argument('--TEXT_COLUMN', default='discriptions')
parser.add_argument('--LABEL_COLUMN', default='mapped_rating')
parser.add_argument('--MAX_LEN', type=int, default=128)
parser.add_argument('--VNCORENLP', default='/data/cuong/data/nlp/embedding/vncorenlp/VnCoreNLP-1.1.1.jar')
parser.add_argument('--BERT_VOCAB', default='/data/cuong/data/nlp/embedding/PhoBERT_base_transformers/dict.txt')
parser.add_argument('--BERT_MODEL', default='/data/cuong/data/nlp/embedding/PhoBERT_base_transformers/model.bin')
parser.add_argument('--BERT_CONFIG', default='/data/cuong/data/nlp/embedding/PhoBERT_base_transformers/config.json')
parser.add_argument('--bpe-codes', default='/data/cuong/data/nlp/embedding/PhoBERT_base_transformers/bpe.codes')
parser.add_argument('--OUTPUT_DIR', default='output/aivivn/models/bertpho/')
parser.add_argument('--EPOCHS', type=int, default=10)
param = parser.parse_args()
def printBoth(message):
date_time = datetime.now(tz).strftime('%Y-%m-%d %H:%M:%S ')
print(date_time + message)
def preprocessing(text):
# remove duplicate characters such as đẹppppppp
text = re.sub(r'([A-Z])\1+', lambda m: m.group(1).upper(), text, flags=re.IGNORECASE)
# remove punctuation
translator = str.maketrans(string.punctuation, ' ' * len(string.punctuation))
text = text.translate(translator)
# remove '_'
text = text.replace('_', ' ')
# remove numbers
text = ''.join([i for i in text if not i.isdigit()])
# lower word
text = text.lower()
# replace special words
replace_list = {
'ô kêi': ' ok ', 'o kê': ' ok ',
'kh ':' không ', 'kô ':' không ', 'hok ':' không ',
'kp ': ' không phải ', 'kô ': ' không ', 'ko ': ' không ', 'khong ': ' không ', 'hok ': ' không ',
}
for k, v in replace_list.items():
text = text.replace(k, v)
# split texts
texts = text.split()
if len(texts) < 5:
text = None
return text
def segment(text):
words = []
for w in rdrsegmenter.tokenize(text):
words = words + w
text = ' '.join([i for i in words])
return text
def bert_encode(text):
text = segment(text)
#text = '<s> ' + bpe.encode(text) + ' </s>'
text = '<s> ' + text + ' </s>'
text_ids = vocab.encode_line(text, append_eos=False, add_if_not_exist=False).long().tolist()
return text_ids
def rpad(array, n=param.MAX_LEN):
"""Right padding."""
current_len = len(array)
if current_len > n:
return array[: n]
extra = n - current_len
return array + ([0] * extra)
class DatasetReview(Dataset):
def __init__(self, reviews, ratings):
self.dataset = [
(
rpad(bert_encode(reviews[i]), n=param.MAX_LEN),
ratings[i],
)
for i in range(len(reviews))
]
def __len__(self):
return len(self.dataset)
def __getitem__(self, index):
X, y = self.dataset[index]
X = torch.tensor(X)
y = torch.tensor(y)
return X, y
def train_one_epoch(model, lossfn, optimizer, dataset, batch_size=32):
generator = torch.utils.data.DataLoader(dataset, batch_size=batch_size, shuffle=True)
model.train()
losses = AverageMeter()
accuracies = AverageMeter()
precisions = AverageMeter()
recalls = AverageMeter()
fscores = AverageMeter()
auc_scores = AverageMeter()
for batch, labels in generator:
batch, labels = batch.to(DEVICE), labels.to(DEVICE)
optimizer.zero_grad()
loss, logits = model(input_ids=batch, labels=labels)
err = lossfn(logits, labels)
loss.backward()
optimizer.step()
acc, precision, recall, fscore, auc_score = class_eval(logits.data.cpu(), labels, pred_type=None)
accuracies.update(acc, labels.size(0))
precisions.update(precision, labels.size(0))
recalls.update(recall, labels.size(0))
fscores.update(fscore, labels.size(0))
auc_scores.update(auc_score, labels.size(0))
losses.update(loss.item(), labels.size(0))
#print('batch', batch)
#print('labels', labels)
#print('logits', logits)
return losses.avg, accuracies.avg, precisions.avg, recalls.avg, fscores.avg, auc_scores.avg
def evaluate_one_epoch(model, lossfn, optimizer, dataset, batch_size=32):
generator = torch.utils.data.DataLoader(dataset, batch_size=batch_size, shuffle=True)
model.eval()
losses = AverageMeter()
accuracies = AverageMeter()
precisions = AverageMeter()
recalls = AverageMeter()
fscores = AverageMeter()
auc_scores = AverageMeter()
with torch.no_grad():
for batch, labels in generator:
batch, labels = batch.to(DEVICE), labels.to(DEVICE)
logits = model(input_ids=batch)[0]
error = lossfn(logits, labels)
acc, precision, recall, fscore, auc_score = class_eval(logits.data.cpu(), labels, pred_type=None)
accuracies.update(acc, labels.size(0))
precisions.update(precision, labels.size(0))
recalls.update(recall, labels.size(0))
fscores.update(fscore, labels.size(0))
auc_scores.update(auc_score, labels.size(0))
losses.update(error.item(), labels.size(0))
return losses.avg, accuracies.avg, precisions.avg, recalls.avg, fscores.avg, auc_scores.avg
# load segmenter
printBoth('Loading segmenter ...')
rdrsegmenter = VnCoreNLP(param.VNCORENLP, annotators="wseg", max_heap_size='-Xmx500m')
# load BPE encoder
printBoth('\nLoading BPE encoder ...')
'''
parser_BPE = argparse.ArgumentParser()
parser_BPE.add_argument('--bpe-codes',
default="/data/cuong/data/nlp/embedding/PhoBERT_base_transformers/bpe.codes",
required=False,
type=str,
help='path to fastBPE BPE'
)
args = parser_BPE.parse_args()
'''
bpe = fastBPE(param)
print('bpe', bpe)
# Load the dictionary
vocab = Dictionary()
vocab.add_from_file(param.BERT_VOCAB)
print('vocab', len(vocab))
# load data
printBoth('Reading csv files ...')
df_train = pd.read_csv(param.PATH_STAR_RATING_TRAIN, usecols=[param.TEXT_COLUMN, param.LABEL_COLUMN], sep=',', nrows=None)
df_test = pd.read_csv(param.PATH_STAR_RATING_TEST, usecols=[param.TEXT_COLUMN, param.LABEL_COLUMN], sep=',', nrows=None)
printBoth('df_train={}; counts={}'.format(df_train.shape, df_train[param.LABEL_COLUMN].value_counts()))
printBoth('df_test={}; counts={}'.format(df_test.shape, df_test[param.LABEL_COLUMN].value_counts()))
df_train.dropna(subset=[param.TEXT_COLUMN], inplace=True)
df_test.dropna(subset=[param.TEXT_COLUMN], inplace=True)
df_train[param.TEXT_COLUMN] = df_train[param.TEXT_COLUMN].apply(lambda x:preprocessing(x))
df_test[param.TEXT_COLUMN] = df_test[param.TEXT_COLUMN].apply(lambda x:preprocessing(x))
df_train.drop_duplicates(subset=param.TEXT_COLUMN, keep = 'first', inplace = True)
df_train.dropna(subset=[param.TEXT_COLUMN], inplace=True)
df_train = df_train.reset_index()
df_test.drop_duplicates(subset=param.TEXT_COLUMN, keep = 'first', inplace = True)
df_test.dropna(subset=[param.TEXT_COLUMN], inplace=True)
df_test = df_test.reset_index()
# prepare training data
printBoth('Preparing bert data for training ...')
reviews = list(df_train[param.TEXT_COLUMN])
ratings = df_train[param.LABEL_COLUMN].values
train_reviews, val_reviews, train_ratings, val_ratings = \
sklearn.model_selection.train_test_split(reviews, ratings, test_size=0.2, random_state=42, shuffle=True, stratify=ratings)
test_reviews = list(df_test[param.TEXT_COLUMN])
test_ratings = df_test[param.LABEL_COLUMN].values
printBoth('Preparing DatasetReview ...')
dataset_train = DatasetReview(train_reviews, train_ratings)
dataset_val = DatasetReview(val_reviews, val_ratings)
dataset_test = DatasetReview(test_reviews, test_ratings)
# build model
printBoth('Building BERT model ...')
class BertForSequenceClassification(BertPreTrainedModel):
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
self.bert = RobertaModel.from_pretrained(Config.BERT_MODEL, config=config)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
self.classifier = nn.Linear(config.hidden_size, self.num_labels)
self.init_weights()
def forward(
self,
input_ids=None,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
labels=None,
):
outputs = self.bert(input_ids)
'''
outputs 2
outputs torch.Size([32, 64, 768])
outputs torch.Size([32, 768])
'''
pooled_output = outputs[1]
pooled_output = self.dropout(pooled_output)
logits = self.classifier(pooled_output)
outputs = (logits,) + outputs[2:] # add hidden states and attention if they are here
if labels is not None:
if self.num_labels == 1:
# We are doing regression
loss_fct = MSELoss()
loss = loss_fct(logits.view(-1), labels.view(-1))
else:
loss_fct = CrossEntropyLoss()
loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
outputs = (loss,) + outputs
return outputs # (loss), logits, (hidden_states), (attentions)
printBoth('Loading RobertaConfig ...')
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
config = RobertaConfig.from_pretrained(param.BERT_CONFIG)
config.num_labels = 2
model = BertForSequenceClassification(config=config)
model = model.to(DEVICE)
lossfn = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=1e-5)
if not os.path.exists(param.OUTPUT_DIR):
os.makedirs(param.OUTPUT_DIR)
best_acc = 0
best_precision = 0
best_recall = 0
best_f1 = 0
best_auc = 0
for epoch in range(param.EPOCHS):
print('\n')
# train
losses, accuracies, precisions, recalls, fscores, auc_scores = \
train_one_epoch(model, lossfn, optimizer, dataset_train, batch_size=32)
printBoth('[TRAIN] epoch={}; losses={:0.5}; accuracies={:0.5}; precisions={:0.5}; recalls={:0.5}; fscores={:0.5}; auc_scores={:0.5}'.\
format(epoch, losses, accuracies, precisions, recalls, fscores, auc_scores))
# valid
losses, accuracies, precisions, recalls, fscores, auc_scores = \
evaluate_one_epoch(model, lossfn, optimizer, dataset_val, batch_size=32)
printBoth('[VALID] epoch={}; losses={:0.5}; accuracies={:0.5}; precisions={:0.5}; recalls={:0.5}; fscores={:0.5}; auc_scores={:0.5}'.\
format(epoch, losses, accuracies, precisions, recalls, fscores, auc_scores))
is_best = (auc_scores > best_auc) if config.num_labels == 2 else (accuracies > best_acc)
best_acc1 = max(accuracies, best_acc)
best_precision = max(precisions, best_precision)
best_recall = max(recalls, best_recall)
best_f1 = max(fscores, best_f1)
best_auc = max(auc_scores, best_auc)
# test
losses, accuracies, precisions, recalls, fscores, auc_scores = \
evaluate_one_epoch(model, lossfn, optimizer, dataset_test, batch_size=32)
printBoth('[TEST] epoch={}; losses={:0.5}; accuracies={:0.5}; precisions={:0.5}; recalls={:0.5}; fscores={:0.5}; auc_scores={:0.5}'.\
format(epoch, losses, accuracies, precisions, recalls, fscores, auc_scores))
# save the best model if any
save_checkpoint({
'epoch': epoch,
'state_dict': model.state_dict(),
'best_acc1': best_acc1,
'optimizer': optimizer.state_dict()
}, is_best, param.OUTPUT_DIR)
printBoth('The end.')