-
Notifications
You must be signed in to change notification settings - Fork 3
/
data.py
382 lines (347 loc) · 13.1 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
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
from copy import deepcopy as cp
import torchtext
from torchtext.vocab import Vocab
from torchtext import data
from nltk.tokenize import word_tokenize
from dotmap import DotMap
from utils import removeDuplicates, elmo_batch_to_ids
from transformers import BertTokenizer, GPT2Tokenizer, RobertaTokenizer, AutoTokenizer
from collections import defaultdict, Counter
import os
import torch
import re
bos_token = "<sos>"
eos_token = "<eos>"
pad_token = "<pad>"
unk_token = "<unk>"
cls_token = "<cls>"
mask_token = "<mask>"
special_tokens = {
"cls_token": cls_token,
"unk_token": unk_token,
"pad_token": pad_token,
"bos_token": bos_token,
"eos_token": eos_token,
"mask_token": mask_token,
}
special_tokens_list = [cls_token]
clean_text = lambda sent: re.sub(r" +", " ", sent)
class DataMaker(object):
def __init__(self, fields=None, dataset_path: str = None):
"""
Args:
fields (dict || list of dicts): Configurations for each field to read from the JSON. Example below:
-------------
{
"name": "src", # field in question
"field": None, # How it should be referred to by the dataloader
"tokenizer": BertTokenizer.from_pretrained('bert-base-uncased').encode, #Tokenizer function. If not speficied (i.e. `None`) defualts to nltk.tokenize.word_tokenize
"sequential": True, #Sequential Data?
""
"eos": True, #Add eos token?
"sos": False, #Add sos token?
}
-------------
data_path (str): The main path for all the data
"""
if isinstance(fields, dict):
fields = [fields]
self.defualt_fields = {
"tokenizer": "NaN",
"tokenize": None,
"postprocess": None,
"eos": False,
"sos": False,
"sequential": True,
"use_vocab": True,
}
self.dataset_fields = [{**self.defualt_fields, **field} for field in fields]
self.dataset_fields = [i for i in map(lambda field: DotMap(field), fields)]
self.dataset_path = dataset_path
def build_data(
self, dataset_name=None, shared_vocab_fields=None, max_len=None,
):
_fields = {
field.name: (
field.field if field.field != None else field.name,
Field(
sequential=field.sequential,
tokenize=word_tokenize
if field.tokenize is None
else field.tokenize,
lower=field.lowercase,
fix_length=field.fix_length,
max_len=max_len,
init_token=bos_token if (field.sos and field.sequential) else None,
eos_token=eos_token if (field.eos and field.sequential) else None,
include_lengths=field.include_lengths,
batch_first=True,
tokenizer=field.tokenizer,
use_vocab=field.use_vocab,
pad_token=field.pad_token,
postprocessing=field.postprocess,
),
)
for field in self.dataset_fields
}
if shared_vocab_fields:
if isinstance(shared_vocab_fields[0], list) == False:
shared_vocab_fields = [shared_vocab_fields]
self.share_vocab_fields(_fields, shared_vocab_fields)
if shared_vocab_fields:
assert (
_fields[shared_vocab_fields[0][0]][1]
== _fields[shared_vocab_fields[0][1]][1]
)
seperate_fields = removeDuplicates([_fields[i][1] for i in _fields])
print("Loading", dataset_name)
self.train, self.valid, self.test = data.TabularDataset.splits(
path=os.path.join(self.dataset_path, dataset_name),
train="_train.json",
validation="_valid.json",
test="_test.json",
format="json",
fields=_fields,
)
print("Dataset:", dataset_name, "loaded")
for f in seperate_fields:
try:
if f.use_vocab:
f.build_vocab(
self.train,
self.valid,
self.test,
specials=special_tokens_list,
min_freq=1,
max_size=10000 if f.tokenize == word_tokenize else 100000,
)
except:
pass
self.vocab = {}
for key in _fields:
try:
self.vocab[_fields[key][0]] = _fields[key][1].vocab
except:
self.vocab[_fields[key][0]] = _fields[key][1].tokenizer
self.vocab = DotMap(self.vocab)
def share_vocab_fields(self, field_data, shared_fields):
"""
Function to share vocab fields in the case of a shared vocab
Note that the first item in each list of shared fields and its correpsonding properties will be copied to each following field
Args:
field_data(dict): The field data object generated for every field in question. See func. `self.build_data`
shared_fields(list): A 2d list of fields which should have the same vocab object. If it is 1d, it is unsqueezed to make it a 2d list.
"""
if isinstance(shared_fields[0], list) == False:
shared_fields = [shared_fields]
for _set in shared_fields:
relevant_field = field_data[_set[0]][1]
for _field in _set[1:]:
field_data[_field] = list(field_data[_field])
field_data[_field][1] = relevant_field
field_data[_field] = tuple(field_data[_field])
def get_iterator(
self, dataset: str, batch_size: int = None, shuffle=None, device=None
):
"""
Args:
dataset (str): representing the partition of the data to iterate over
Returns:
`torchtext.data.BucketIterator` object to iterate over the partition that was specified
"""
if dataset not in ["test", "train", "valid"]:
raise NotImplementedError
if batch_size is None:
raise NotImplementedError
if dataset == "train":
return data.BucketIterator(
self.train, batch_size, shuffle=shuffle, device=device
)
elif dataset == "valid":
return data.BucketIterator(
self.valid, batch_size, shuffle=shuffle, device=device
)
elif dataset == "test":
return data.BucketIterator(
self.test, batch_size, shuffle=shuffle, device=device
)
def decode(self, input, vocab_partition, batch=False):
"""
Args:
input (list, torch.Tensor): Input of ids
vocab_partition (str): partition of vocab to convert using
batch (bool): 2d if True else 1d list
"""
_vocab = getattr(self.vocab, vocab_partition)
output_sentences = []
if not batch:
if isinstance(input, torch.Tensor):
input = input.tolist()
input = [input]
if isinstance(input, torch.Tensor):
input = [i.tolist() for i in input]
try:
for element in input:
sentence = []
for token_id in element:
token = _vocab.itos[token_id]
if token not in ["<pad>", "<sos>", "<eos>"]:
sentence.append(token)
else:
pass
output_sentences.append(clean_text(" ".join(sentence)))
except:
for element in input:
output_sentences.append(
clean_text(_vocab.decode(element, skip_special_tokens=True))
)
if not batch:
return output_sentences[0]
else:
return output_sentences
def get_dm_conf(_type, field_name, name=None):
if _type is None:
_type = "normal"
if _type in [
"elmo",
"normal",
"char",
]:
conf = cp(dm_conf[_type])
if name is None:
name = field_name
conf["field"] = field_name
conf["name"] = name
return conf
else:
try:
conf = cp(dm_conf.transformer_base)
tokenizer = AutoTokenizer.from_pretrained(_type)
conf["tokenizer"] = tokenizer
conf["tokenize"] = tokenizer.encode
conf["pad_token"] = tokenizer.pad_token_id
conf["field"] = conf["name"] = field_name
return conf
except Exception as e:
raise NotImplementedError(
"We don't have that preprocessing mechanism yet " + str(e)
)
dm_conf = DotMap(
{
"transformer_base": {
"name": None,
"field": "src",
"tokenize": None, # tokenizer.encode
"tokenizer": None, # tokenizer
"sequential": True,
"eos": False,
"sos": False,
"use_vocab": False,
"pad_token": None,
"postprocess": None,
"include_lengths": True,
"lowercase": None,
"fix_length": None,
},
"elmo": {
"name": None,
"field": "src",
"tokenize": elmo_batch_to_ids,
"tokenizer": None,
"sequential": False,
"eos": False,
"sos": False,
"use_vocab": False,
"pad_token": 261,
"postprocess": None,
"include_lengths": True,
"fix_length": None,
"lowercase": True,
},
"normal": {
"name": None,
"field": "src",
"tokenize": None,
"tokenizer": None,
"sequential": True,
"eos": True,
"sos": True,
"use_vocab": True,
"pad_token": "<pad>",
"postprocess": None,
"include_lengths": True,
"fix_length": None,
"lowercase": True,
},
"char": {
"name": None,
"field": "src",
"tokenize": lambda x: list(x),
"tokenizer": None,
"sequential": True,
"eos": False,
"sos": False,
"use_vocab": True,
"pad_token": "<pad>",
"postprocess": None,
"include_lengths": True,
"fix_length": 15,
"lowercase": True,
},
}
)
class Field(data.Field):
def __init__(self, **kwargs):
self.tokenizer = kwargs.pop("tokenizer")
self.max_len = kwargs.pop("max_len")
super().__init__(**kwargs)
def pad(self, minibatch):
"""Pad a batch of examples using this field.
Pads to self.fix_length if provided, otherwise pads to the length of
the longest example in the batch. Prepends self.init_token and appends
self.eos_token if those attributes are not None. Returns a tuple of the
padded list and a list containing lengths of each example if
`self.include_lengths` is `True` and `self.sequential` is `True`, else just
returns the padded list. If `self.sequential` is `False`, no padding is applied.
"""
minibatch = list(minibatch)
if not self.sequential:
return minibatch
if self.max_len is None and self.fix_length is None:
max_len = max(len(x) for x in minibatch)
elif self.fix_length is not None:
max_len = (
self.fix_length + (self.init_token, self.eos_token).count(None) - 2
)
else:
max_len = min(
self.max_len + (self.init_token, self.eos_token).count(None) - 2,
max(len(x) for x in minibatch),
)
padded, lengths = [], []
for x in minibatch:
if self.pad_first:
padded.append(
[self.pad_token] * max(0, max_len - len(x))
+ ([] if self.init_token is None else [self.init_token])
+ list(x[-max_len:] if self.truncate_first else x[:max_len])
+ ([] if self.eos_token is None else [self.eos_token])
)
else:
padded.append(
([] if self.init_token is None else [self.init_token])
+ list(x[-max_len:] if self.truncate_first else x[:max_len])
+ ([] if self.eos_token is None else [self.eos_token])
+ [self.pad_token] * max(0, max_len - len(x))
)
lengths.append(
len(padded[-1])
- max(
0, max_len - len(x)
) # + abs((self.init_token, self.eos_token).count(None) - 4)),
)
if self.include_lengths:
return (padded, lengths)
return padded
def word_idx_getter(input, *args, **kwargs):
return input[1:-1]