-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdataset.py
302 lines (289 loc) · 15.8 KB
/
dataset.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
from torch.utils.data import Dataset
import torch
from torch.utils.data import TensorDataset
from transformers import BertTokenizer,BertModel
from tqdm import tqdm
import pandas as pd
from ace import cal_sim, Passage
import nlpaug.augmenter.word as naw
import nlpaug.augmenter.char as nac
class Data_WithContext:
def __init__(self, config, max_seq_len, model_type):
self.config = config
self.model_type = model_type
self.tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
self.max_seq_len = max_seq_len
self.context_tokenizer = BertTokenizer.from_pretrained("princeton-nlp/sup-simcse-bert-base-uncased")
self.context_model = BertModel.from_pretrained("princeton-nlp/sup-simcse-bert-base-uncased").cuda()
def load_train_and_dev_files(self, train_file, dev_file, hard_sample_con=False, noisy=False):
print('Loading train data...')
train_set = self.load_file(train_file, hard_sample_con, noisy)
print(len(train_set), 'train data loaded.')
print('Loading dev data...')
dev_set = self.load_file(dev_file, False, noisy)
print(len(dev_set), 'dev data loaded.')
return train_set, dev_set
def load_valid_files(self, valid_file, noisy=False):
print('Loading test data...')
valid_set = self.load_file(valid_file, False, noisy)
print(len(valid_set), 'test data loaded.')
return valid_set
def load_file(self, file_path, hard_construction=False, noisy=False) -> TensorDataset:
quo_list, reply_list, label_list = self.loaddata(file_path, hard_construction, noisy=noisy)
dataset = self._convert_sentence_pair_to_bert(quo_list, reply_list, label_list)
return dataset
def loaddata(self, filename, hard_construction, noisy=False):
data_frame = pd.read_csv(filename, sep='#', header=None)
quo_list, reply_list, label_list = [], [], []
for index, row in tqdm(data_frame.iterrows()):
quo_context = row[0]
quo = row[1]
reply_pos = row[2]
reply_pos_context = row[3]
reply_neg = list(row[4::2])
reply_context_candidates = list(row[5::2]) # 取出从第三个数开始的偶数项
quo_partcontext = self._selectblock(quo, quo_context, noisy)
reply_partcontext_list = []
# 1个quo1个reply_pos 4个reply_neg
reply_pos_partcontext = self._selectblock(reply_pos, reply_pos_context, noisy)
# self._selectpartcontext_sim(reply_pos, reply_pos_context,reply_sim_pos) # random.randint(0,len(reply_context_candidates)) # random.randint(0, len(candidates)) # 随机生成label位置
reply_partcontext_list.append(reply_pos_partcontext)
label_list.append(1)
for i in range(len(reply_context_candidates)):
reply_neg_partcontext = self._selectblock(reply_neg[i], reply_context_candidates[i], noisy)
# reply_neg_partcontext = self._selectblock(reply_neg[i], reply_context_candidates[i])# self._selectpartcontext_sim(reply_neg[i], reply_context_candidates[i], reply_sim[i])
reply_partcontext_list.append(reply_neg_partcontext)
label_list.append(0)
if hard_construction:
# pos_reply:neg_reply = 1:2
pos_reply_contrast_1 = self._selectblock_con(reply_pos, reply_pos_context, self.config.pos1_block_size,
self.config.pos1_block_num, noisy)
pos_reply_contrast_2 = self._selectblock_con(reply_pos, reply_pos_context, self.config.pos2_block_size,
self.config.pos2_block_num, noisy)
pos_reply_contrast_3 = self._selectblock_con(reply_pos, reply_pos_context, self.config.pos3_block_size,
self.config.pos3_block_num, noisy)
reply_partcontext_list.append(pos_reply_contrast_1)
reply_partcontext_list.append(pos_reply_contrast_2)
reply_partcontext_list.append(pos_reply_contrast_3)
label_list.append(1)
label_list.append(1)
label_list.append(1)
neg_reply_contrast_1 = self._selectblock_con(reply_neg[0], reply_context_candidates[0],
self.config.pos1_block_size,
self.config.pos1_block_num,
noisy)
neg_reply_contrast_2 = self._selectblock_con(reply_neg[1], reply_context_candidates[1],
self.config.pos2_block_size,
self.config.pos2_block_num,
noisy)
neg_reply_contrast_3 = self._selectblock_con(reply_neg[2], reply_context_candidates[2],
self.config.pos3_block_size,
self.config.pos3_block_num,
noisy)
neg_reply_contrast_4 = self._selectblock_con(reply_neg[3], reply_context_candidates[3],
self.config.block_num,
self.config.block_size,
noisy)
reply_partcontext_list.append(neg_reply_contrast_1)
reply_partcontext_list.append(neg_reply_contrast_2)
reply_partcontext_list.append(neg_reply_contrast_3)
reply_partcontext_list.append(neg_reply_contrast_4)
label_list.append(0)
label_list.append(0)
label_list.append(0)
label_list.append(0)
reply_list.append(reply_partcontext_list)
quo_list.append(quo_partcontext)
return quo_list, reply_list, label_list
def _selectblock(self, query, context, noisy):
if noisy == 'RandomWordAug':
aug_random = naw.RandomWordAug()
query = aug_random.augment(query)[0]
context = aug_random.augment(context)[0]
if noisy == 'BackTranslationAug':
aug_backtranlation = naw.BackTranslationAug()
query = aug_backtranlation.augment(query)[0]
context = aug_backtranlation.augment(context)[0]
if noisy == 'KeyboardAug':
aug_key = nac.KeyboardAug()
query = aug_key.augment(query)[0]
context = aug_key.augment(context)[0]
passage, cnt = Passage.split_document_into_blocks(self.tokenizer.tokenize(context), self.tokenizer,
self.config.block_size, 0, hard=False)
sim_list = []
for con in passage.blocks:
sim_list.append(cal_sim(self.context_tokenizer, self.context_model, query, con))
sorted_id = sorted(range(len(sim_list)), key=lambda k: sim_list[k], reverse=True)
# random.shuffle(sorted_id)#不随机的时候注释掉
newindex = sorted_id[:self.config.block_num]
sorted_new_index = sorted(newindex)
part_list = []
part_list.append(query)
for i in sorted_new_index:
part_list.append(passage.blocks[i].__str__())
con_str = ""
for part_str in part_list:
con_str = con_str + part_str
return con_str
def _selectblock_con(self, query, context, pos_block_size, pos_block_num, noisy):
if noisy == 'RandomWordAug':
aug_random = naw.RandomWordAug()
query = aug_random.augment(query)[0]
context = aug_random.augment(context)[0]
if noisy == 'BackTranslationAug':
aug_backtranlation = naw.BackTranslationAug()
query = aug_backtranlation.augment(query)[0]
context = aug_backtranlation.augment(context)[0]
if noisy == 'KeyboardAug':
aug_key = nac.KeyboardAug()
query = aug_key.augment(query)[0]
context = aug_key.augment(context)[0]
passage, cnt = Passage.split_document_into_blocks(self.tokenizer.tokenize(context), self.tokenizer,
pos_block_size, 0, hard=False)
sim_list = []
for con in passage.blocks:
sim_list.append(cal_sim(self.context_tokenizer, self.context_model, query, con))
sorted_id = sorted(range(len(sim_list)), key=lambda k: sim_list[k], reverse=True)
# random.shuffle(sorted_id) # 不随机的时候注释掉
newindex = sorted_id[:pos_block_num] #
sorted_new_index = sorted(newindex)
part_list = []
part_list.append(query)
for i in sorted_new_index:
part_list.append(passage.blocks[i].__str__())
con_str = ""
for part_str in part_list:
con_str = con_str + part_str
return con_str
def _convert_sentence_pair_to_bert(self, quo_list, reply_list, label_list=None):
all_input_ids, all_input_mask, all_segment_ids = [], [], []
for i, _ in tqdm(enumerate(quo_list), ncols=80):
token_list = []
mask_list = []
segment_list = []
quo_list[i] = self.tokenizer.tokenize(quo_list[i])
for j, _ in enumerate(reply_list[i]):
reply_list[i][j] = self.tokenizer.tokenize(reply_list[i][j])
tokens = ['[CLS]'] + quo_list[i] + ['[SEP]']
segment_ids = [0] * len(tokens)
tokens += reply_list[i][j] + ['[SEP]']
segment_ids += [1] * (len(reply_list[i][j]) + 1)
if len(tokens) > self.max_seq_len:
tokens = tokens[:self.max_seq_len]
segment_ids = segment_ids[:self.max_seq_len]
input_ids = self.tokenizer.convert_tokens_to_ids(tokens)
input_mask = [1] * len(input_ids)
tokens_len = len(input_ids)
input_ids += [0] * (self.max_seq_len - tokens_len)
segment_ids += [0] * (self.max_seq_len - tokens_len)
input_mask += [0] * (self.max_seq_len - tokens_len)
token_list.append(input_ids)
mask_list.append(input_mask)
segment_list.append(segment_ids)
all_input_ids.extend(token_list)
all_input_mask.extend(mask_list)
all_segment_ids.extend(segment_list)
all_input_ids = torch.tensor(all_input_ids, dtype=torch.long)
all_input_mask = torch.tensor(all_input_mask, dtype=torch.long)
all_segment_ids = torch.tensor(all_segment_ids, dtype=torch.long)
# if label_list: # train
all_label_ids = torch.tensor(label_list, dtype=torch.long)
return TensorDataset(
all_input_ids, all_input_mask, all_segment_ids, all_label_ids)
class Data_WithoutContext:
def __init__(self,
config,
max_seq_len: int = 512,
model_type: str = 'bert'):
self.config = config
self.model_type = model_type
self.tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
self.max_seq_len = max_seq_len
def load_train_and_dev_files(self, train_file, dev_file, noisy=False):
print('Loading train records for train...')
train_set = self.load_file(train_file, noisy)
print(len(train_set), 'training records loaded.')
print('Loading dev records...')
dev_set = self.load_file(dev_file, noisy)
print(len(dev_set), 'dev records loaded.')
return train_set, dev_set
def load_valid_files(self, valid_file, noisy=False):
print('Loading valid records...')
valid_set = self.load_file(valid_file, noisy)
print(len(valid_set), 'valid records loaded.')
return valid_set
def load_file(self, file_path='train.txt', noisy=False) -> TensorDataset:
quo_list, reply_list, label_list = self._load_file(file_path, noisy)
dataset = self._convert_sentence_pair_to_bert(
quo_list, reply_list, label_list)
return dataset
def _augnoisy(self, text, noisy):
if noisy != False:
augmented_text = text
if noisy == 'RandomWordAug':
aug_random = naw.RandomWordAug()
augmented_text = aug_random.augment(text)[0]
if noisy == 'BackTranslationAug':
aug_backtranlation = naw.BackTranslationAug()
augmented_text = aug_backtranlation.augment(text)[0]
# print('augmented_text_BackTranslationAug', augmented_text)
if noisy == 'KeyboardAug':
aug_key = nac.KeyboardAug()
augmented_text = aug_key.augment(text)[0]
return augmented_text
def _load_file(self, filename, noisy):
data_frame = pd.read_csv(filename, sep='#', header=None)[:500]
quo_list, reply_list, label_list = [], [], []
for index, row in data_frame.iterrows():
candidates = list(row[4::2])
quo_tokens = row[1]
quo_tokens = self._augnoisy(quo_tokens, noisy)
reply_label = row[2]
reply_label = self._augnoisy(reply_label, noisy)
reply_list_every_row = []
reply_list_every_row.append(reply_label)
label_list.append(1)
for i in range(len(candidates)):
reply_tokens_no_label = candidates[i]
reply_tokens_no_label = self._augnoisy(reply_tokens_no_label, noisy)
reply_list_every_row.append(reply_tokens_no_label)
label_list.append(0)
reply_list.append(reply_list_every_row)
quo_list.append(quo_tokens)
return quo_list, reply_list, label_list
def _convert_sentence_pair_to_bert(self, quo_list, reply_list, label_list=None, ):
all_input_ids, all_input_mask, all_segment_ids = [], [], []
for i, _ in tqdm(enumerate(quo_list), ncols=80):
token_list = []
mask_list = []
segment_list = []
quo_list[i] = self.tokenizer.tokenize(quo_list[i])
for j, _ in enumerate(reply_list[i]):
reply_list[i][j] = self.tokenizer.tokenize(reply_list[i][j])
tokens = ['[CLS]'] + quo_list[i] + ['[SEP]']
segment_ids = [0] * len(tokens)
tokens += reply_list[i][j] + ['[SEP]']
# segment_ids += [j+1] * (len(s2_list[i][j]) + 1)
segment_ids += [1] * (len(reply_list[i][j]) + 1)
if len(tokens) > self.max_seq_len:
tokens = tokens[:self.max_seq_len]
segment_ids = segment_ids[:self.max_seq_len]
input_ids = self.tokenizer.convert_tokens_to_ids(tokens)
input_mask = [1] * len(input_ids)
tokens_len = len(input_ids)
input_ids += [0] * (self.max_seq_len - tokens_len)
segment_ids += [0] * (self.max_seq_len - tokens_len)
input_mask += [0] * (self.max_seq_len - tokens_len)
token_list.append(input_ids)
mask_list.append(input_mask)
segment_list.append(segment_ids)
all_input_ids.extend(token_list)
all_input_mask.extend(mask_list)
all_segment_ids.extend(segment_list)
all_input_ids = torch.tensor(all_input_ids, dtype=torch.long)
all_input_mask = torch.tensor(all_input_mask, dtype=torch.long)
all_segment_ids = torch.tensor(all_segment_ids, dtype=torch.long)
# if label_list: # train
all_label_ids = torch.tensor(label_list, dtype=torch.long)
return TensorDataset(
all_input_ids, all_input_mask, all_segment_ids, all_label_ids)