-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathPrepareData_transformers.py
159 lines (119 loc) · 5.64 KB
/
PrepareData_transformers.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
# This script reads the pre-processed data and prepares encoded inputs for BERT/SciBERT models
# Example command:
# python PrepareData_transformers.py --model_type 'SciBERT' --preprocessed_data_loc '/home/msadat3/HTC/SciHTC_data/SciHTC_preprocessed/' --output_dir '/home/msadat3/HTC/SciHTC_data/SciBERT_without_keywords/' --use_keywords 'no'
import numpy as np
import pandas
from transformers import *
import string
from Utils import *
import os.path as p
import os
import nltk
import argparse
from Data_Subset import Data_Subset
def Tokenize_Input(text, add_special_tokens):
text = str(text)
encoded = tokenizer.encode(text,add_special_tokens=add_special_tokens)
return encoded
def pad_seq(seq,max_len,pad_idx):
if len(seq)>max_len:
sep = seq[-1]
seq = seq[0:max_len-1]
seq.append(sep)
while len(seq) != max_len:
seq.append(pad_idx)
return seq
def get_attention_masks(X):
attention_masks = []
# For each sentence...
for sent in X:
att_mask = [int(token_id != tokenizer.pad_token_id) for token_id in sent]
# Store the attention mask for this sentence.
att_mask = np.asarray(att_mask)
attention_masks.append(att_mask)
attention_masks = np.asarray(attention_masks)
return attention_masks
def tag_keywords_or_not(input_tokens, keywords_tokens):
punctuations = ' '.join(string.punctuation)
punctuations = tokenizer.encode(punctuations)
Keyword_labels = []
for token in input_tokens:
if (token in keywords_tokens) and (token not in punctuations):
Keyword_labels.append(1)
else:
Keyword_labels.append(0)
return Keyword_labels
def prepare_data_for_transformers(output_location, trainingSet, testingSet, devSet, use_keywords=False):
if p.exists(output_location) == False:
os.mkdir(output_location)
column_name = ""
if use_keywords == 'yes':
column_name = "Preprocessed_title_abstract_keywords"
elif use_keywords == 'no':
column_name = "Preprocessed_title_abstract"
# print(use_keywords, column_name)
# quit()
X_train = trainingSet.apply(lambda x: Tokenize_Input(x[column_name], add_special_tokens=True), axis=1)
X_test = testingSet.apply(lambda x: Tokenize_Input(x[column_name], add_special_tokens=True), axis=1)
X_dev = devSet.apply(lambda x: Tokenize_Input(x[column_name], add_special_tokens=True), axis=1)
X_train = pandas.Series(X_train)
X_test = pandas.Series(X_test)
X_dev = pandas.Series(X_dev)
trainingSet['BERT_tokenized'] = X_train
testingSet['BERT_tokenized'] = X_test
devSet['BERT_tokenized'] = X_dev
max_len = 0
for x in X_train:
if len(x) > max_len:
max_len = len(x)
for x in X_test:
if len(x) > max_len:
max_len = len(x)
for x in X_dev:
if len(x) > max_len:
max_len = len(x)
#print(max_len)
X_train = X_train.apply(pad_seq, max_len=max_len, pad_idx=tokenizer.pad_token_id)
X_test = X_test.apply(pad_seq, max_len=max_len, pad_idx=tokenizer.pad_token_id)
X_dev = X_dev.apply(pad_seq, max_len=max_len, pad_idx=tokenizer.pad_token_id)
X_train = np.array(X_train.values.tolist())
X_test = np.array(X_test.values.tolist())
X_dev = np.array(X_dev.values.tolist())
att_mask_train = get_attention_masks(X_train)
att_mask_test = get_attention_masks(X_test)
att_mask_dev = get_attention_masks(X_dev)
save_data(X_train, output_location+'X_train.pkl')
save_data(X_test, output_location + 'X_test.pkl')
save_data(X_dev, output_location + 'X_dev.pkl')
save_data(att_mask_train, output_location + 'att_mask_train.pkl')
save_data(att_mask_test, output_location + 'att_mask_test.pkl')
save_data(att_mask_dev, output_location + 'att_mask_dev.pkl')
print('Prepared data shape:')
print(X_train.shape, att_mask_train.shape)
print(X_test.shape, att_mask_test.shape)
print(X_dev.shape, att_mask_dev.shape)
if use_keywords == 'no':
X_train_keywords = trainingSet.apply(lambda x: Tokenize_Input(x['Preprocessed_keywords'], add_special_tokens=False), axis=1)
X_train_keywords = pandas.Series(X_train_keywords)
trainingSet['BERT_tokenized_keywords'] = X_train_keywords
y_train_keywords = trainingSet.apply(lambda x: tag_keywords_or_not(x['BERT_tokenized'], x['BERT_tokenized_keywords']), axis=1)
y_train_keywords = pandas.Series(y_train_keywords)
y_train_keywords = y_train_keywords.apply(pad_seq, max_len=max_len, pad_idx=tokenizer.pad_token_id)
y_train_keywords = np.array(y_train_keywords.values.tolist())
save_data(y_train_keywords, output_location + 'y_train_keywords.pkl')
print('Keyword labels shape:')
print(y_train_keywords.shape)
parser = argparse.ArgumentParser(description='Prepare data for BERT/SciBERT models.')
parser.add_argument("--model_type", type=str, help="Type of the model: BERT/SciBERT.")
parser.add_argument("--use_keywords", type=str, help="Include keywords with input or not.")
parser.add_argument("--preprocessed_data_loc", type=str, help="Directory containing the pre-processed data.")
parser.add_argument("--output_dir", type=str, help="Output directory for saving the prepared data for experiments.")
args = parser.parse_args()
trainingSet = pandas.read_csv(args.preprocessed_data_loc+'train_preprocessed.csv')
testingSet = pandas.read_csv(args.preprocessed_data_loc+'test_preprocessed.csv')
devSet = pandas.read_csv(args.preprocessed_data_loc+'dev_preprocessed.csv')
if args.model_type == 'SciBERT':
tokenizer = BertTokenizer.from_pretrained('allenai/scibert_scivocab_uncased', do_lower_case=True)
elif args.model_type == 'BERT':
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased', do_lower_case=True)
prepare_data_for_transformers(args.output_dir, trainingSet, testingSet, devSet, use_keywords=args.use_keywords)