-
Notifications
You must be signed in to change notification settings - Fork 0
/
Data.py
209 lines (185 loc) · 8.55 KB
/
Data.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
'''
Vineet Kumar, sioom.ai
'''
from pytorch_lightning import LightningDataModule
import torch
from torch.utils.data import Dataset, RandomSampler, DataLoader
from logging import getLogger
from typing import List, Dict, Tuple, Any
import pandas as pd
from sklearn.model_selection import train_test_split
logg = getLogger(__name__)
class Data(LightningDataModule):
def __init__(self, d_params: dict):
super().__init__()
if 'batch_size' not in d_params:
logg.critical('"batch_size" MUST be specified')
exit()
for batch_size_key in ('train', 'val', 'test'):
if batch_size_key not in d_params['batch_size'] or d_params[
'batch_size'][batch_size_key] is None or d_params[
'batch_size'][batch_size_key] == 0:
d_params['batch_size'][batch_size_key] = 1
# Trainer('auto_scale_batch_size': True...) requires self.batch_size
self.batch_size = d_params['batch_size']['train']
self.d_params = d_params
def prepare_data(self,
no_training: bool = False,
no_testing: bool = False) -> None:
self.dataset_metadata, train_data, val_data, test_data =\
_get_trainValTest_data(
data_file_path=self.d_params['default_format_path'],
batch_size=self.d_params['batch_size'],
split=self.d_params['dataset_split'])
if not no_training:
self.train_data = Data_set(train_data)
self.valid_data = Data_set(val_data)
if not no_testing:
self.test_data = Data_set(test_data)
if no_training and no_testing:
logg.debug('No Training and no Testing')
@staticmethod
def app_specific_params() -> Tuple[Dict[Any, Any], Dict[Any, Any]]:
app_specific_init, app_specific = {}, {}
app_specific_init['num_classes'] = 8
app_specific_init['imbalanced_classes'] = [
0.3030, 0.2167, 0.1460, 0.1061, 0.0925, 0.0650, 0.0567, 0.0140
]
app_specific['num_classes'] = 8
return app_specific_init, app_specific
def get_dataset_metadata(self) -> Dict[str, Any]:
return self.dataset_metadata
def put_tokenizer(self, tokenizer):
self.tokenizer = tokenizer
def train_dataloader(self) -> DataLoader:
return DataLoader(self.train_data,
batch_size=self.batch_size,
shuffle=False,
sampler=RandomSampler(self.train_data),
batch_sampler=None,
num_workers=6,
collate_fn=self.gpt2_collater,
pin_memory=True,
drop_last=False,
timeout=0)
def val_dataloader(self) -> DataLoader:
return DataLoader(self.valid_data,
batch_size=self.d_params['batch_size']['val'],
shuffle=False,
sampler=RandomSampler(self.valid_data),
batch_sampler=None,
num_workers=6,
collate_fn=self.gpt2_collater,
pin_memory=True,
drop_last=False,
timeout=0)
def test_dataloader(self) -> DataLoader:
return DataLoader(self.test_data,
batch_size=self.d_params['batch_size']['test'],
shuffle=False,
sampler=RandomSampler(self.test_data),
batch_sampler=None,
num_workers=6,
collate_fn=self.gpt2_collater,
pin_memory=True,
drop_last=False,
timeout=0)
def gpt2_collater(self, examples: List[Dict[str, str]]) -> Dict[str, Any]:
batch_sentence_ids, batch_texts, batch_labels = [], [], []
for example in examples:
batch_sentence_ids.append(example['sentence_id'])
batch_texts.append(example['text'])
batch_labels.append(
self.dataset_metadata['class_info']['names'].index(
example['label']))
# input to Bert model is truncated if it is longer than max allowed
batch_model_inputs = self.tokenizer(text=batch_texts,
padding=True,
truncation=True,
return_tensors='pt',
return_token_type_ids=True,
return_attention_mask=True)
return {
'model_inputs': {
'input_ids':
batch_model_inputs['input_ids'].type(torch.LongTensor),
'attention_mask':
batch_model_inputs['attention_mask'].type(torch.FloatTensor),
'token_type_ids':
batch_model_inputs['token_type_ids'].type(torch.LongTensor)
},
'labels':
(torch.LongTensor(batch_labels)).view(len(batch_labels), 1),
'sentence_ids': tuple(batch_sentence_ids)
}
class Data_set(Dataset):
# example = sentence_id plus text plus label
def __init__(self, examples: List[Dict[str, str]]):
self.examples = examples
def __len__(self) -> int:
return len(self.examples)
def __getitem__(self, idx: int) -> Dict[str, str]:
return (self.examples[idx])
def _get_trainValTest_data(
data_file_path: str, batch_size: Dict[str, int], split: Dict[str, int]
) -> Tuple[Dict[str, Any], List[Dict[str, str]], List[Dict[str, str]],
List[Dict[str, str]]]:
assert split['train'] + split['val'] + split['test']
df = pd.read_csv(data_file_path)
if not split['train'] and split['test']:
# testing a dataset on a checkpoint file; no training
df_train, df_val, df_test, split['val'], split[
'test'] = None, None, df, 0, 100
else:
df_train, df_temp = train_test_split(df,
stratify=df["label"],
train_size=(split['train'] / 100),
random_state=42)
df_val, df_test = train_test_split(
df_temp,
stratify=df_temp["label"],
test_size=(split['test'] / (split['val'] + split['test'])),
random_state=42)
assert len(df) == len(df_train) + len(df_val) + len(df_test)
dataset_metadata = {
'batch_size': batch_size,
'dataset_info': {
'split': (split['train'], split['val'], split['test']),
'lengths': (len(df), len(df_train) if df_train is not None else 0,
len(df_val) if df_val is not None else 0,
len(df_test) if df_test is not None else 0),
},
'class_info': {
'names': [], # this is filled a little later in the code
'dataset_prop':
df.label.value_counts(normalize=True).to_dict(),
'train_prop':
df_train.label.value_counts(
normalize=True).to_dict() if df_train is not None else 0,
'val_prop':
df_val.label.value_counts(
normalize=True).to_dict() if df_val is not None else 0,
'test_prop':
df_test.label.value_counts(
normalize=True).to_dict() if df_test is not None else 0,
'test_lengths':
df_test.label.value_counts(
normalize=False).to_dict() if df_test is not None else 0
}
}
# list of unique labels in original dataset which are ordered by their
# proportion of examples in test dataset; makes it easier to visualize
ordered__unique_labels = [
k for k, v in sorted(dataset_metadata["class_info"]
["test_lengths"].items(),
key=lambda item: item[1],
reverse=True)
]
for dataset_unique_label in df.label.unique().tolist():
if dataset_unique_label not in ordered__unique_labels:
ordered__unique_labels.append(dataset_unique_label)
dataset_metadata["class_info"]["names"] = ordered__unique_labels
return dataset_metadata, df_train.to_dict(
'records') if df_train is not None else 0, df_val.to_dict(
'records') if df_val is not None else 0, df_test.to_dict(
'records') if df_test is not None else 0