-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathMimicsDataset.py
191 lines (152 loc) · 6.59 KB
/
MimicsDataset.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
import argparse
import pandas as pd
import json
import os
import torch
from IPython import embed
from torch.utils.data import DataLoader, Dataset
from transformers import AlbertTokenizer
from sklearn.model_selection import train_test_split
class MimicsDataset(Dataset):
def __init__(self, tokenizer, args, mode='dev'):
self.tokenizer = tokenizer
self.data_dir = args.data_dir
self.mode = mode
self.max_seq_len = args.max_seq_len
self.text_input = args.text_input
if args.click_explore == 'Click':
df = pd.read_csv(os.path.join(self.data_dir, 'Click_titles_and_snippets_all.tsv'), sep='\t')
elif args.click_explore == 'Explore':
df = pd.read_csv(os.path.join(self.data_dir, 'Explore_titles_and_snippets.tsv'), sep='\t')
df['question'] = df.question.replace('\"{2,}', '', regex=True)
df = df[df['titles'].notna()]
df = df[df['snippets'].notna()]
if args.with_el_only:
df = df[df['engagement_level'] > 0]
qs = list(set(df['query'].unique()))
X_train, X_dev = train_test_split(df, test_size=0.2, random_state=42)
# X_train, X_dev = train_test_split(qs, test_size=0.2, random_state=42)
if mode == 'train':
self.X = X_train
# self.X = df[df['query'].isin(X_train)]
elif mode == 'dev':
self.X = X_dev
# self.X = df[df['query'].isin(X_dev)]
elif mode == 'test':
self.X = df
def __len__(self):
return len(self.X)
def __getitem__(self, idx):
tensors = self.example_to_tensor(idx)
return tensors
def example_to_tensor(self, idx):
x = self.X.iloc[idx]
query = x.query
question = x.question
answs = x[['option_1', 'option_2', 'option_3', 'option_4', 'option_5']].fillna('').str.cat(sep=' ')
if 't' in self.text_input:
second = x.titles
elif 's' in self.text_input:
second = x.snippets
else:
second = ''
label = x.engagement_level / 10
if 'qqa' in self.text_input:
first = ' [SEP] '.join([query, question, answs])
else:
first = query
encoded = self.tokenizer.encode_plus(first, second,
add_special_tokens=True,
max_length=self.max_seq_len,
truncation='only_second',
return_overflowing_tokens=False,
return_special_tokens_mask=False,
return_token_type_ids=True,
padding='max_length'
)
encoded['attention_mask'] = torch.tensor(encoded['attention_mask'])
encoded['input_ids'] = torch.tensor(encoded['input_ids'])
encoded['token_type_ids'] = torch.tensor(encoded['token_type_ids'])
encoded.update({'label': torch.FloatTensor([label]),
'idx': torch.tensor(idx)})
return encoded
@staticmethod
def add_model_specific_args(parser, root_dir):
parser.add_argument('--data_dir', type=str, default='../data/')
parser.add_argument('--mode', type=str, default='dev')
parser.add_argument('--max_seq_len', type=int, default=512)
return parser
class MimicsDatasetNrez(Dataset):
def __init__(self, tokenizer, args, mode='dev'):
self.tokenizer = tokenizer
self.data_dir = args.data_dir
self.mode = mode
self.max_seq_len = args.max_seq_len
self.text_input = args.text_input
self.n_serp_elems = args.n_serp_elems # consider only top N SERP elements
df = pd.read_csv(os.path.join(self.data_dir, 'Click_titles_and_snippets_all_sliced.tsv'), sep='\t')
df['question'] = df.question.replace('\"{2,}', '', regex=True)
df = df[df['titles'].notna()]
df = df[df['snippets'].notna()]
if args.with_el_only:
df = df[df['engagement_level'] > 0]
qs = list(set(df['query'].unique()))
X_train, X_dev = train_test_split(df, test_size=0.2, random_state=42)
# X_train, X_dev = train_test_split(qs, test_size=0.2, random_state=42)
if mode == 'train':
self.X = X_train
# self.X = df[df['query'].isin(X_train)]
elif mode == 'dev':
self.X = X_dev
# self.X = df[df['query'].isin(X_dev)]
elif mode == 'test':
self.X = df
def __len__(self):
return len(self.X)
def __getitem__(self, idx):
tensors = self.example_to_tensor(idx)
return tensors
def example_to_tensor(self, idx):
x = self.X.iloc[idx]
query = x.query
question = x.question
answs = x[['option_1', 'option_2', 'option_3', 'option_4', 'option_5']].fillna('').str.cat(sep=' ')
if 't' in self.text_input:
splt = x.titles.split('|#$')
elif 's' in self.text_input:
splt = x.snippets.split('|#$')
second = ' '.join(splt[:self.n_serp_elems])
label = x.engagement_level / 10
if 'qqa' in self.text_input:
first = ' [SEP] '.join([query, question, answs])
else:
first = query
encoded = self.tokenizer.encode_plus(first, second,
add_special_tokens=True,
max_length=self.max_seq_len,
truncation='only_second',
return_overflowing_tokens=False,
return_special_tokens_mask=False,
return_token_type_ids=True,
padding='max_length'
)
encoded['attention_mask'] = torch.tensor(encoded['attention_mask'])
encoded['input_ids'] = torch.tensor(encoded['input_ids'])
encoded['token_type_ids'] = torch.tensor(encoded['token_type_ids'])
encoded.update({'label': torch.FloatTensor([label]),
'idx': torch.tensor(idx)})
return encoded
@staticmethod
def add_model_specific_args(parser, root_dir):
parser.add_argument('--data_dir', type=str, default='../data/')
parser.add_argument('--mode', type=str, default='dev')
parser.add_argument('--max_seq_len', type=int, default=512)
parser.add_argument('--n_serp_elems', type=int, default=10)
return parser
if __name__ == "__main__":
main_arg_parser = argparse.ArgumentParser(description="MIMICS dataset")
parser = MimicsDataset.add_model_specific_args(main_arg_parser, os.getcwd())
args = parser.parse_args()
tokenizer = AlbertTokenizer.from_pretrained('albert-base-v2')
cd = MimicsDataset(tokenizer, args, args.mode)
embed()