/
wilds_dataset.py
522 lines (457 loc) · 20.3 KB
/
wilds_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
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
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
import os
import time
import torch
import numpy as np
class WILDSDataset:
"""
Shared dataset class for all WILDS datasets.
Each data point in the dataset is an (x, y, metadata) tuple, where:
- x is the input features
- y is the target
- metadata is a vector of relevant information, e.g., domain.
For convenience, metadata also contains y.
"""
DEFAULT_SPLITS = {'train': 0, 'val': 1, 'test': 2}
DEFAULT_SPLIT_NAMES = {'train': 'Train', 'val': 'Validation', 'test': 'Test'}
DEFAULT_SOURCE_DOMAIN_SPLITS = [0]
def __init__(self, root_dir, download, split_scheme):
if len(self._metadata_array.shape) == 1:
self._metadata_array = self._metadata_array.unsqueeze(1)
self._add_coarse_domain_metadata()
self.check_init()
def __len__(self):
return len(self.y_array)
def __getitem__(self, idx):
# Any transformations are handled by the WILDSSubset
# since different subsets (e.g., train vs test) might have different transforms
x = self.get_input(idx)
y = self.y_array[idx]
metadata = self.metadata_array[idx]
return x, y, metadata
def get_input(self, idx):
"""
Args:
- idx (int): Index of a data point
Output:
- x (Tensor): Input features of the idx-th data point
"""
raise NotImplementedError
def eval(self, y_pred, y_true, metadata):
"""
Args:
- y_pred (Tensor): Predicted targets
- y_true (Tensor): True targets
- metadata (Tensor): Metadata
Output:
- results (dict): Dictionary of results
- results_str (str): Pretty print version of the results
"""
raise NotImplementedError
def get_subset(self, split, frac=1.0, transform=None):
"""
Args:
- split (str): Split identifier, e.g., 'train', 'val', 'test'.
Must be in self.split_dict.
- frac (float): What fraction of the split to randomly sample.
Used for fast development on a small dataset.
- transform (function): Any data transformations to be applied to the input x.
Output:
- subset (WILDSSubset): A (potentially subsampled) subset of the WILDSDataset.
"""
if split not in self.split_dict:
raise ValueError(f"Split {split} not found in dataset's split_dict.")
split_mask = self.split_array == self.split_dict[split]
split_idx = np.where(split_mask)[0]
if frac < 1.0:
# Randomly sample a fraction of the split
num_to_retain = int(np.round(float(len(split_idx)) * frac))
split_idx = np.sort(np.random.permutation(split_idx)[:num_to_retain])
return WILDSSubset(self, split_idx, transform)
def _add_coarse_domain_metadata(self):
"""
Update metadata fields, map and values with coarse-grained domain information.
"""
if hasattr(self, '_metadata_map'):
self._metadata_map['from_source_domain'] = [False, True]
self._metadata_fields.append('from_source_domain')
from_source_domain = torch.as_tensor(
[1 if split in self.source_domain_splits else 0 for split in self.split_array],
dtype=torch.int64
).unsqueeze(dim=1)
self._metadata_array = torch.cat(
[self._metadata_array, from_source_domain],
dim=1
)
def check_init(self):
"""
Convenience function to check that the WILDSDataset is properly configured.
"""
required_attrs = ['_dataset_name', '_data_dir',
'_split_scheme', '_split_array',
'_y_array', '_y_size',
'_metadata_fields', '_metadata_array']
for attr_name in required_attrs:
assert hasattr(self, attr_name), f'WILDSDataset is missing {attr_name}.'
# Check that data directory exists
if not os.path.exists(self.data_dir):
raise ValueError(
f'{self.data_dir} does not exist yet. Please generate the dataset first.')
# Check splits
assert self.split_dict.keys()==self.split_names.keys()
assert 'train' in self.split_dict
assert 'val' in self.split_dict
# Check the form of the required arrays
assert (isinstance(self.y_array, torch.Tensor) or isinstance(self.y_array, list))
assert isinstance(self.metadata_array, torch.Tensor), 'metadata_array must be a torch.Tensor'
# Check that dimensions match
assert len(self.y_array) == len(self.metadata_array)
assert len(self.split_array) == len(self.metadata_array)
# Check metadata
assert len(self.metadata_array.shape) == 2
assert len(self.metadata_fields) == self.metadata_array.shape[1]
# Check that it is not both classification and detection
assert not (self.is_classification and self.is_detection)
# For convenience, include y in metadata_fields if y_size == 1
if self.y_size == 1:
assert 'y' in self.metadata_fields
@property
def latest_version(cls):
def is_later(u, v):
"""Returns true if u is a later version than v."""
u_major, u_minor = tuple(map(int, u.split('.')))
v_major, v_minor = tuple(map(int, v.split('.')))
if (u_major > v_major) or (
(u_major == v_major) and (u_minor > v_minor)):
return True
else:
return False
latest_version = '0.0'
for key in cls.versions_dict.keys():
if is_later(key, latest_version):
latest_version = key
return latest_version
@property
def dataset_name(self):
"""
A string that identifies the dataset, e.g., 'amazon', 'camelyon17'.
"""
return self._dataset_name
@property
def version(self):
"""
A string that identifies the dataset version, e.g., '1.0'.
"""
if self._version is None:
return self.latest_version
else:
return self._version
@property
def versions_dict(self):
"""
A dictionary where each key is a version string (e.g., '1.0')
and each value is a dictionary containing the 'download_url' and
'compressed_size' keys.
'download_url' is the URL for downloading the dataset archive.
If None, the dataset cannot be downloaded automatically
(e.g., because it first requires accepting a usage agreement).
'compressed_size' is the approximate size of the compressed dataset in bytes.
"""
return self._versions_dict
@property
def data_dir(self):
"""
The full path to the folder in which the dataset is stored.
"""
return self._data_dir
@property
def collate(self):
"""
Torch function to collate items in a batch.
By default returns None -> uses default torch collate.
"""
return getattr(self, '_collate', None)
@property
def split_scheme(self):
"""
A string identifier of how the split is constructed,
e.g., 'standard', 'mixed-to-test', 'user', etc.
"""
return self._split_scheme
@property
def split_dict(self):
"""
A dictionary mapping splits to integer identifiers (used in split_array),
e.g., {'train': 0, 'val': 1, 'test': 2}.
Keys should match up with split_names.
"""
return getattr(self, '_split_dict', WILDSDataset.DEFAULT_SPLITS)
@property
def split_names(self):
"""
A dictionary mapping splits to their pretty names,
e.g., {'train': 'Train', 'val': 'Validation', 'test': 'Test'}.
Keys should match up with split_dict.
"""
return getattr(self, '_split_names', WILDSDataset.DEFAULT_SPLIT_NAMES)
@property
def source_domain_splits(self):
"""
List of split IDs that are from the source domain.
"""
return getattr(self, '_source_domain_splits', WILDSDataset.DEFAULT_SOURCE_DOMAIN_SPLITS)
@property
def split_array(self):
"""
An array of integers, with split_array[i] representing what split the i-th data point
belongs to.
"""
return self._split_array
@property
def y_array(self):
"""
A Tensor of targets (e.g., labels for classification tasks),
with y_array[i] representing the target of the i-th data point.
y_array[i] can contain multiple elements.
"""
return self._y_array
@property
def y_size(self):
"""
The number of dimensions/elements in the target, i.e., len(y_array[i]).
For standard classification/regression tasks, y_size = 1.
For multi-task or structured prediction settings, y_size > 1.
Used for logging and to configure models to produce appropriately-sized output.
"""
return self._y_size
@property
def n_classes(self):
"""
Number of classes for single-task classification datasets.
Used for logging and to configure models to produce appropriately-sized output.
None by default.
Leave as None if not applicable (e.g., regression or multi-task classification).
"""
return getattr(self, '_n_classes', None)
@property
def is_classification(self):
"""
Boolean. True if the task is classification, and false otherwise.
"""
return getattr(self, '_is_classification', (self.n_classes is not None))
@property
def is_detection(self):
"""
Boolean. True if the task is detection, and false otherwise.
"""
return getattr(self, '_is_detection', False)
@property
def metadata_fields(self):
"""
A list of strings naming each column of the metadata table, e.g., ['hospital', 'y'].
Must include 'y'.
"""
return self._metadata_fields
@property
def metadata_array(self):
"""
A Tensor of metadata, with the i-th row representing the metadata associated with
the i-th data point. The columns correspond to the metadata_fields defined above.
"""
return self._metadata_array
@property
def metadata_map(self):
"""
An optional dictionary that, for each metadata field, contains a list that maps from
integers (in metadata_array) to a string representing what that integer means.
This is only used for logging, so that we print out more intelligible metadata values.
Each key must be in metadata_fields.
For example, if we have
metadata_fields = ['hospital', 'y']
metadata_map = {'hospital': ['East', 'West']}
then if metadata_array[i, 0] == 0, the i-th data point belongs to the 'East' hospital
while if metadata_array[i, 0] == 1, it belongs to the 'West' hospital.
"""
return getattr(self, '_metadata_map', None)
@property
def original_resolution(self):
"""
Original image resolution for image datasets.
"""
return getattr(self, '_original_resolution', None)
def initialize_data_dir(self, root_dir, download):
"""
Helper function for downloading/updating the dataset if required.
Note that we only do a version check for datasets where the download_url is set.
Currently, this includes all datasets except Yelp.
Datasets for which we don't control the download, like Yelp,
might not handle versions similarly.
"""
self.check_version()
os.makedirs(root_dir, exist_ok=True)
data_dir = os.path.join(root_dir, f'{self.dataset_name}_v{self.version}')
version_file = os.path.join(data_dir, f'RELEASE_v{self.version}.txt')
# If the dataset exists at root_dir, then don't download.
if not self.dataset_exists_locally(data_dir, version_file):
self.download_dataset(data_dir, download)
return data_dir
def dataset_exists_locally(self, data_dir, version_file):
download_url = self.versions_dict[self.version]['download_url']
# There are two ways to download a dataset:
# 1. Automatically through the WILDS package
# 2. From a third party (e.g. OGB-MolPCBA is downloaded through the OGB package)
# Datasets downloaded from a third party need not have a download_url and RELEASE text file.
return (
os.path.exists(data_dir) and (
os.path.exists(version_file) or
(len(os.listdir(data_dir)) > 0 and download_url is None)
)
)
def download_dataset(self, data_dir, download_flag):
version_dict = self.versions_dict[self.version]
download_url = version_dict['download_url']
compressed_size = version_dict['compressed_size']
# Check that download_url exists.
if download_url is None:
raise ValueError(f'{self.dataset_name} cannot be automatically downloaded. Please download it manually.')
# Check that the download_flag is set to true.
if not download_flag:
raise FileNotFoundError(
f'The {self.dataset_name} dataset could not be found in {data_dir}. Initialize the dataset with '
f'download=True to download the dataset. If you are using the example script, run with --download. '
f'This might take some time for large datasets.'
)
from wilds.datasets.download_utils import download_and_extract_archive
print(f'Downloading dataset to {data_dir}...')
print(f'You can also download the dataset manually at https://wilds.stanford.edu/downloads.')
try:
start_time = time.time()
download_and_extract_archive(
url=download_url,
download_root=data_dir,
filename='archive.tar.gz',
remove_finished=True,
size=compressed_size)
download_time_in_minutes = (time.time() - start_time) / 60
print(f"\nIt took {round(download_time_in_minutes, 2)} minutes to download and uncompress the dataset.\n")
except Exception as e:
print(f"\n{os.path.join(data_dir, 'archive.tar.gz')} may be corrupted. Please try deleting it and rerunning this command.\n")
print(f"Exception: ", e)
def check_version(self):
# Check that the version is valid.
if self.version not in self.versions_dict:
raise ValueError(f'Version {self.version} not supported. Must be in {self.versions_dict.keys()}.')
# Check that the specified version is the latest version. Otherwise, warn.
current_major_version, current_minor_version = tuple(map(int, self.version.split('.')))
latest_major_version, latest_minor_version = tuple(map(int, self.latest_version.split('.')))
if latest_major_version > current_major_version:
print(
f'*****************************\n'
f'{self.dataset_name} has been updated to version {self.latest_version}.\n'
f'You are currently using version {self.version}.\n'
f'We highly recommend updating the dataset by not specifying the older version in the '
f'command-line argument or dataset constructor.\n'
f'See https://wilds.stanford.edu/changelog for changes.\n'
f'*****************************\n')
elif latest_minor_version > current_minor_version:
print(
f'*****************************\n'
f'{self.dataset_name} has been updated to version {self.latest_version}.\n'
f'You are currently using version {self.version}.\n'
f'Please consider updating the dataset.\n'
f'See https://wilds.stanford.edu/changelog for changes.\n'
f'*****************************\n')
@staticmethod
def standard_eval(metric, y_pred, y_true):
"""
Args:
- metric (Metric): Metric to use for eval
- y_pred (Tensor): Predicted targets
- y_true (Tensor): True targets
Output:
- results (dict): Dictionary of results
- results_str (str): Pretty print version of the results
"""
results = {
**metric.compute(y_pred, y_true),
}
results_str = (
f"Average {metric.name}: {results[metric.agg_metric_field]:.3f}\n"
)
return results, results_str
@staticmethod
def standard_group_eval(metric, grouper, y_pred, y_true, metadata, aggregate=True):
"""
Args:
- metric (Metric): Metric to use for eval
- grouper (CombinatorialGrouper): Grouper object that converts metadata into groups
- y_pred (Tensor): Predicted targets
- y_true (Tensor): True targets
- metadata (Tensor): Metadata
Output:
- results (dict): Dictionary of results
- results_str (str): Pretty print version of the results
"""
results, results_str = {}, ''
if aggregate:
results.update(metric.compute(y_pred, y_true))
results_str += f"Average {metric.name}: {results[metric.agg_metric_field]:.3f}\n"
g = grouper.metadata_to_group(metadata)
group_results = metric.compute_group_wise(y_pred, y_true, g, grouper.n_groups)
for group_idx in range(grouper.n_groups):
group_str = grouper.group_field_str(group_idx)
group_metric = group_results[metric.group_metric_field(group_idx)]
group_counts = group_results[metric.group_count_field(group_idx)]
results[f'{metric.name}_{group_str}'] = group_metric
results[f'count_{group_str}'] = group_counts
if group_results[metric.group_count_field(group_idx)] == 0:
continue
results_str += (
f' {grouper.group_str(group_idx)} '
f"[n = {group_results[metric.group_count_field(group_idx)]:6.0f}]:\t"
f"{metric.name} = {group_results[metric.group_metric_field(group_idx)]:5.3f}\n")
results[f'{metric.worst_group_metric_field}'] = group_results[f'{metric.worst_group_metric_field}']
results_str += f"Worst-group {metric.name}: {group_results[metric.worst_group_metric_field]:.3f}\n"
return results, results_str
class WILDSSubset(WILDSDataset):
def __init__(self, dataset, indices, transform, do_transform_y=False):
"""
This acts like `torch.utils.data.Subset`, but on `WILDSDatasets`.
We pass in `transform` (which is used for data augmentation) explicitly
because it can potentially vary on the training vs. test subsets.
`do_transform_y` (bool): When this is false (the default),
`self.transform ` acts only on `x`.
Set this to true if `self.transform` should
operate on `(x,y)` instead of just `x`.
"""
self.dataset = dataset
self.indices = indices
inherited_attrs = ['_dataset_name', '_data_dir', '_collate',
'_split_scheme', '_split_dict', '_split_names',
'_y_size', '_n_classes',
'_metadata_fields', '_metadata_map']
for attr_name in inherited_attrs:
if hasattr(dataset, attr_name):
setattr(self, attr_name, getattr(dataset, attr_name))
self.transform = transform
self.do_transform_y = do_transform_y
def __getitem__(self, idx):
x, y, metadata = self.dataset[self.indices[idx]]
if self.transform is not None:
if self.do_transform_y:
x, y = self.transform(x, y)
else:
x = self.transform(x)
return x, y, metadata
def __len__(self):
return len(self.indices)
@property
def split_array(self):
return self.dataset._split_array[self.indices]
@property
def y_array(self):
return self.dataset._y_array[self.indices]
@property
def metadata_array(self):
return self.dataset.metadata_array[self.indices]
def eval(self, y_pred, y_true, metadata):
return self.dataset.eval(y_pred, y_true, metadata)