-
Notifications
You must be signed in to change notification settings - Fork 1
/
datautils.py
336 lines (288 loc) · 13.8 KB
/
datautils.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
from functools import partial
from io import DEFAULT_BUFFER_SIZE
import pandas as pd
from pytorch_lightning.utilities.cloud_io import load
from sklearn.preprocessing import StandardScaler
import torch
import argparse
import pytorch_lightning as pl
from pl_bolts.datamodules import CIFAR10DataModule
from torchvision.datasets import STL10, CIFAR100
from typing import Any, Callable, Optional, Union, List
from torch.utils.data import DataLoader, random_split
import numpy as np
import os
import random
from torch.utils.data import DataLoader, Dataset, random_split, Subset
import glob
from itertools import combinations
from torchvision.datasets import CIFAR10
from kornia.color.lab import RgbToLab
from torchvision import transforms
from pl_bolts.transforms.dataset_normalizations import cifar10_normalization
import utils
CIFAR_CLASSES = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']
REAL_TASKS = list(combinations(CIFAR_CLASSES, 5))[:126]
class Rgb2L(RgbToLab):
def forward(self, image: torch.Tensor) -> torch.Tensor:
x = super().forward(image)
return x[:1]
class MyCIFAR10(CIFAR10):
FACTORS_DF_PATH='./data/cifar-factors.csv'
def __init__(
self,
root: str,
train: bool = True,
transform: Optional[Callable] = None,
target_transform: Optional[Callable] = None,
download: bool = False,
include_classes: List[int] = None,
return_indicies: bool = False,
factors: List[str] = None,
) -> None:
super().__init__(root, train, transform, target_transform, download)
self.include_classes = include_classes
self.return_indicies = return_indicies
if self.include_classes is not None:
include = np.array([t in self.include_classes for t in self.targets])
self.data = self.data[include]
self.targets = np.array(self.targets)[include]
self.factors = None
if factors is not None:
self.factors = []
factors_df = pd.read_csv(self.FACTORS_DF_PATH, index_col=0)
for f in factors:
if f == 'mean_color':
self.factors.append(self.data.mean((1, 2)))
elif f == 'color_minmax_diff':
self.factors.append((self.data.max((1, 2, 3)) - self.data.min((1, 2, 3)))[..., None])
else:
self.factors.append(torch.load(factors_df.loc[f][f'path_{"train" if train else "test"}']))
self.factors = np.concatenate(self.factors, axis=1).astype(np.float32)
def __getitem__(self, index):
out = list(super().__getitem__(index))
if self.return_indicies:
out.append(index)
if self.factors is not None:
out.append(self.factors[index])
return tuple(out)
@property
def factors_dim(self) -> int:
return self.factors.shape[1] if self.factors is not None else 0
def __repr__(self) -> str:
return super().__repr__() + f'\nClasses: {self.include_classes}'
class MyCIFAR10DataModule(CIFAR10DataModule):
dataset_cls = MyCIFAR10
def __init__(
self,
data_dir: Optional[str] = os.environ.get('DATA_ROOT', os.getcwd()),
val_split: Union[int, float] = 0.1,
num_workers: int = 16,
normalize: bool = True,
batch_size: int = 32,
test_batch_size: Optional[int] = None,
data_seed: int = 42,
shuffle: bool = False,
pin_memory: bool = True,
drop_last: bool = True,
random_labelling: bool = False,
random_labelling_seed: Optional[int] = None,
n_classes: int = 10,
gt2class: Optional[str] = None,
n_train_images: int = -1,
multi_task: bool = "",
path2pool: str = '',
n_tasks: int = 1,
persistent_workers: bool = False,
return_indicies: bool = False,
to_lightness: bool = False,
include_classes: List[int] = None,
augs: bool = False,
factors: Optional[List[str]] = None,
train_val_split: Optional[str] = None,
*args: Any,
**kwargs: Any,
) -> None:
super().__init__( # type: ignore[misc]
data_dir=data_dir,
val_split=val_split,
num_workers=num_workers,
normalize=normalize,
batch_size=batch_size,
seed=data_seed,
shuffle=shuffle,
pin_memory=pin_memory,
drop_last=drop_last,
)
self.EXTRA_ARGS['download'] = True
self.dataset_cls = partial(MyCIFAR10, return_indicies=return_indicies, include_classes=include_classes, factors=factors)
assert n_train_images == -1 or n_train_images >= batch_size or not drop_last
self.test_batch_size = test_batch_size or self.batch_size
self.n_train_images = n_train_images
self.random_labelling_seed = random_labelling_seed if random_labelling_seed is not None else self.seed
print(f'[Datamodule] ===> : Random_labelling={random_labelling}, Shuffle={shuffle}, Data_seed={data_seed}, Persistent_workers={persistent_workers}')
self.random_labelling = random_labelling
self._num_classes = n_classes
# if not random_labelling:
# print(type(gt2class))
# assert self.num_classes == 10 or isinstance(gt2class, str)
self._gt2class = None
if isinstance(gt2class, str) and gt2class != '' and not self.random_labelling:
self._gt2class = {gt: i for i, clss in enumerate(gt2class.split('|')) for gt in clss.split(',') }
print(self._gt2class)
self.persistent_workers = persistent_workers
self.to_lightness = to_lightness
if self.to_lightness:
self.dims = (1, 32, 32)
self.augs = augs
self.train_transforms=self.get_transforms(train=True)
self.val_transforms=self.get_transforms(train=False)
self.test_transforms=self.get_transforms(train=False)
self.train_val_split = train_val_split
@staticmethod
def add_argparse_args(parent_parser):
parser = argparse.ArgumentParser(parents=[parent_parser], add_help=False)
parser.add_argument('--batch_size', type=int, default=256)
parser.add_argument('--data_seed', type=int, default=42)
parser.add_argument('--random_labelling_seed', type=int, default=42)
parser.add_argument('--n_train_images', type=int, default=-1)
parser.add_argument('--no-shuffle', dest='shuffle', action='store_false')
parser.add_argument('--shuffle', dest='shuffle', action='store_true')
parser.set_defaults(shuffle=True)
parser.add_argument('--multi_task', type=str, default='')
# parser.add_argument('--n_tasks', type=int, default=1)
parser.add_argument('--val_split', type=float, default=0.1)
parser.add_argument('--gt2class', type=str, default="")
parser.add_argument('--path2pool', type=str, default="")
parser.add_argument('--random_labelling', action='store_true', default=False)
parser.add_argument('--no_drop_last', dest='drop_last', action='store_false', default=True)
parser.add_argument('--persistent_workers', action='store_true', default=False)
parser.add_argument('--return_indicies', action='store_true', default=False)
parser.add_argument('--to_lightness', action='store_true', default=False)
parser.add_argument('--normalize', action='store_true', default=True)
parser.add_argument('--no_normalize', dest='normalize', action='store_false', default=True)
parser.add_argument('--include_classes', type=int, default=None, nargs='+')
parser.add_argument('--augs', action='store_true', default=False)
parser.add_argument('--no_augs', dest='augs', action='store_false', default=False)
parser.add_argument('--factors', type=str, default=None, nargs='*')
parser.add_argument('--train_val_split', type=str, default=None)
parser.add_argument('--dataset_path', type=str, default='')
return parser
@property
def num_classes(self) -> int:
return self._num_classes
def get_transforms(self, train=False) -> Callable:
t = [transforms.ToTensor()]
if self.augs and train:
t += [
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
]
if self.to_lightness:
t.append(Rgb2L())
if self.normalize:
if self.to_lightness: raise ValueError(f'{self.normalize=} and {self.to_lightness=}')
t.append(cifar10_normalization())
return transforms.Compose(t)
@property
def factors_dim(self) -> int:
return self.dataset_train.dataset.factors_dim
def setup(self, stage: Optional[str] = None) -> None:
"""
Creates train, val, and test dataset
"""
# prepare all datasets
super().setup()
if self.dataset_train.dataset.factors is not None:
scaler = StandardScaler()
self.dataset_train.dataset.factors = scaler.fit_transform(self.dataset_train.dataset.factors)
self.dataset_val.dataset.factors = scaler.transform(self.dataset_val.dataset.factors)
self.dataset_test.factors = scaler.transform(self.dataset_test.factors)
print(f'[Datamodule] ===> {self.dataset_train.indices[:20]=}')
if self.random_labelling:
g = torch.Generator().manual_seed(self.random_labelling_seed)
self.dataset_train.dataset.targets = torch.randint(0, self.num_classes, (len(self.dataset_train.dataset),), generator=g).tolist()
self.dataset_val.dataset.targets = torch.randint(0, self.num_classes, (len(self.dataset_val.dataset),), generator=g).tolist()
self.dataset_test.targets = torch.randint(0, self.num_classes, (len(self.dataset_test),), generator=g).tolist()
elif self._gt2class is not None:
classes = self.dataset_train.dataset.classes
self.dataset_train.dataset.targets = [self._gt2class[classes[t]] for t in self.dataset_train.dataset.targets]
self.dataset_val.dataset.targets = [self._gt2class[classes[t]] for t in self.dataset_val.dataset.targets]
self.dataset_test.targets = [self._gt2class[classes[t]] for t in self.dataset_test.targets]
def _split_dataset(self, dataset: Dataset, train: bool = True) -> Dataset:
"""
Splits the dataset into train and validation set
"""
if self.train_val_split is None:
len_dataset = len(dataset) # type: ignore[arg-type]
splits = self._get_splits(len_dataset)
dataset_train, _, dataset_val = random_split(dataset, splits, generator=torch.Generator().manual_seed(self.seed))
else:
splits = torch.load(self.train_val_split)
dataset_train, dataset_val = [Subset(dataset, indices) for indices in splits]
if train:
return dataset_train
return dataset_val
def _get_splits(self, len_dataset: int) -> List[int]:
"""
Computes split lengths for train and validation set
"""
if isinstance(self.val_split, int):
val_len = self.val_split
elif isinstance(self.val_split, float):
val_len = int(self.val_split * len_dataset)
else:
raise ValueError(f'Unsupported type {type(self.val_split)}')
if self.n_train_images == -1:
train_len = len_dataset - val_len
else:
train_len = self.n_train_images
splits = [train_len, len_dataset - train_len - val_len, val_len]
print('train/_/val splits :', splits)
return splits
def _data_loader(
self,
dataset: torch.utils.data.Dataset,
generator: Any = None,
shuffle: bool = False,
persistent_workers: bool = False,
batch_size: int = None,
drop_last: bool = None,
) -> torch.utils.data.DataLoader:
return torch.utils.data.DataLoader(
dataset,
batch_size=batch_size or self.batch_size,
shuffle=shuffle,
generator=generator,
num_workers=self.num_workers,
drop_last=self.drop_last if drop_last is None else drop_last,
pin_memory=self.pin_memory,
worker_init_fn=MyCIFAR10DataModule._worker_init_fn,
persistent_workers=persistent_workers,
)
def train_dataloader(
self,
generator: Optional[torch.Generator] = None,
persistent_workers: bool = False,
batch_size: int = None,
drop_last: bool = None,
) -> torch.utils.data.DataLoader:
""" The train dataloader """
persistent_workers = persistent_workers or self.persistent_workers
return self._data_loader(self.dataset_train, shuffle=self.shuffle, generator=generator, persistent_workers=persistent_workers, batch_size=batch_size, drop_last=drop_last)
def val_dataloader(self, persistent_workers: bool = False, batch_size: int = None) -> torch.utils.data.DataLoader:
""" The val dataloader """
persistent_workers = persistent_workers or self.persistent_workers
batch_size = batch_size or self.test_batch_size
return self._data_loader(self.dataset_val, persistent_workers=persistent_workers, batch_size=batch_size, drop_last=False)
def test_dataloader(self, persistent_workers: bool = False, batch_size: int = None) -> torch.utils.data.DataLoader:
""" The val dataloader """
batch_size = batch_size or self.test_batch_size
return self._data_loader(self.dataset_test, persistent_workers=persistent_workers, batch_size=batch_size, drop_last=False)
@staticmethod
def _worker_init_fn(_id):
seed = torch.utils.data.get_worker_info().seed % 2**32
random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
np.random.seed(seed)