/
dataset_utils.py
775 lines (678 loc) · 27.9 KB
/
dataset_utils.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
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
import os
import random
import time
import warnings
from multiprocessing.pool import ThreadPool
import numpy as np
from keras.api_export import keras_export
from keras.utils import io_utils
from keras.utils.module_utils import tensorflow as tf
@keras_export("keras.utils.split_dataset")
def split_dataset(
dataset, left_size=None, right_size=None, shuffle=False, seed=None
):
"""Splits a dataset into a left half and a right half (e.g. train / test).
Args:
dataset:
A `tf.data.Dataset`, a `torch.utils.data.Dataset` object,
or a list/tuple of arrays with the same length.
left_size: If float (in the range `[0, 1]`), it signifies
the fraction of the data to pack in the left dataset. If integer, it
signifies the number of samples to pack in the left dataset. If
`None`, defaults to the complement to `right_size`.
Defaults to `None`.
right_size: If float (in the range `[0, 1]`), it signifies
the fraction of the data to pack in the right dataset.
If integer, it signifies the number of samples to pack
in the right dataset.
If `None`, defaults to the complement to `left_size`.
Defaults to `None`.
shuffle: Boolean, whether to shuffle the data before splitting it.
seed: A random seed for shuffling.
Returns:
A tuple of two `tf.data.Dataset` objects:
the left and right splits.
Example:
>>> data = np.random.random(size=(1000, 4))
>>> left_ds, right_ds = keras.utils.split_dataset(data, left_size=0.8)
>>> int(left_ds.cardinality())
800
>>> int(right_ds.cardinality())
200
"""
dataset_type_spec = _get_type_spec(dataset)
if dataset_type_spec is None:
raise TypeError(
"The `dataset` argument must be either"
"a `tf.data.Dataset`, a `torch.utils.data.Dataset`"
"object, or a list/tuple of arrays. "
f"Received: dataset={dataset} of type {type(dataset)}"
)
if right_size is None and left_size is None:
raise ValueError(
"At least one of the `left_size` or `right_size` "
"must be specified. Received: left_size=None and "
"right_size=None"
)
dataset_as_list = _convert_dataset_to_list(dataset, dataset_type_spec)
if shuffle:
if seed is None:
seed = random.randint(0, int(1e6))
random.seed(seed)
random.shuffle(dataset_as_list)
total_length = len(dataset_as_list)
left_size, right_size = _rescale_dataset_split_sizes(
left_size, right_size, total_length
)
left_split = list(dataset_as_list[:left_size])
right_split = list(dataset_as_list[-right_size:])
left_split = _restore_dataset_from_list(
left_split, dataset_type_spec, dataset
)
right_split = _restore_dataset_from_list(
right_split, dataset_type_spec, dataset
)
left_split = tf.data.Dataset.from_tensor_slices(left_split)
right_split = tf.data.Dataset.from_tensor_slices(right_split)
# apply batching to the splits if the dataset is batched
if dataset_type_spec is tf.data.Dataset and is_batched(dataset):
batch_size = get_batch_size(dataset)
if batch_size is not None:
left_split = left_split.batch(batch_size)
right_split = right_split.batch(batch_size)
left_split = left_split.prefetch(tf.data.AUTOTUNE)
right_split = right_split.prefetch(tf.data.AUTOTUNE)
return left_split, right_split
def _convert_dataset_to_list(
dataset,
dataset_type_spec,
data_size_warning_flag=True,
ensure_shape_similarity=True,
):
"""Convert `dataset` object to a list of samples.
Args:
dataset: A `tf.data.Dataset`, a `torch.utils.data.Dataset` object,
or a list/tuple of arrays.
dataset_type_spec: the type of the dataset.
data_size_warning_flag: If set to `True`, a warning will
be issued if the dataset takes longer than 10 seconds to iterate.
Defaults to `True`.
ensure_shape_similarity: If set to `True`, the shape of
the first sample will be used to validate the shape of rest of the
samples. Defaults to `True`.
Returns:
List: A list of samples.
"""
dataset_iterator = _get_data_iterator_from_dataset(
dataset, dataset_type_spec
)
dataset_as_list = []
start_time = time.time()
for sample in _get_next_sample(
dataset_iterator,
ensure_shape_similarity,
data_size_warning_flag,
start_time,
):
if dataset_type_spec in [tuple, list]:
# The try-except here is for NumPy 1.24 compatibility, see:
# https://numpy.org/neps/nep-0034-infer-dtype-is-object.html
try:
arr = np.array(sample)
except ValueError:
arr = np.array(sample, dtype=object)
dataset_as_list.append(arr)
else:
dataset_as_list.append(sample)
return dataset_as_list
def _get_data_iterator_from_dataset(dataset, dataset_type_spec):
"""Get the iterator from a dataset.
Args:
dataset: A `tf.data.Dataset`, a `torch.utils.data.Dataset` object,
or a list/tuple of arrays.
dataset_type_spec: The type of the dataset.
Returns:
iterator: An `iterator` object.
"""
if dataset_type_spec == list:
if len(dataset) == 0:
raise ValueError(
"Received an empty list dataset. "
"Please provide a non-empty list of arrays."
)
if _get_type_spec(dataset[0]) is np.ndarray:
expected_shape = dataset[0].shape
for i, element in enumerate(dataset):
if np.array(element).shape[0] != expected_shape[0]:
raise ValueError(
"Received a list of NumPy arrays with different "
f"lengths. Mismatch found at index {i}, "
f"Expected shape={expected_shape} "
f"Received shape={np.array(element).shape}."
"Please provide a list of NumPy arrays with "
"the same length."
)
else:
raise ValueError(
"Expected a list of `numpy.ndarray` objects,"
f"Received: {type(dataset[0])}"
)
return iter(zip(*dataset))
elif dataset_type_spec == tuple:
if len(dataset) == 0:
raise ValueError(
"Received an empty list dataset."
"Please provide a non-empty tuple of arrays."
)
if _get_type_spec(dataset[0]) is np.ndarray:
expected_shape = dataset[0].shape
for i, element in enumerate(dataset):
if np.array(element).shape[0] != expected_shape[0]:
raise ValueError(
"Received a tuple of NumPy arrays with different "
f"lengths. Mismatch found at index {i}, "
f"Expected shape={expected_shape} "
f"Received shape={np.array(element).shape}."
"Please provide a tuple of NumPy arrays with "
"the same length."
)
else:
raise ValueError(
"Expected a tuple of `numpy.ndarray` objects, "
f"Received: {type(dataset[0])}"
)
return iter(zip(*dataset))
elif dataset_type_spec == tf.data.Dataset:
if is_batched(dataset):
dataset = dataset.unbatch()
return iter(dataset)
elif is_torch_dataset(dataset):
return iter(dataset)
elif dataset_type_spec == np.ndarray:
return iter(dataset)
raise ValueError(f"Invalid dataset_type_spec: {dataset_type_spec}")
def _get_next_sample(
dataset_iterator,
ensure_shape_similarity,
data_size_warning_flag,
start_time,
):
"""Yield data samples from the `dataset_iterator`.
Args:
dataset_iterator: An `iterator` object.
ensure_shape_similarity: If set to `True`, the shape of
the first sample will be used to validate the shape of rest of the
samples. Defaults to `True`.
data_size_warning_flag: If set to `True`, a warning will
be issued if the dataset takes longer than 10 seconds to iterate.
Defaults to `True`.
start_time (float): the start time of the dataset iteration. this is
used only if `data_size_warning_flag` is set to true.
Yields:
data_sample: The next sample.
"""
try:
dataset_iterator = iter(dataset_iterator)
first_sample = next(dataset_iterator)
if isinstance(first_sample, (tf.Tensor, np.ndarray)) or is_torch_tensor(
first_sample
):
first_sample_shape = np.array(first_sample).shape
else:
first_sample_shape = None
ensure_shape_similarity = False
yield first_sample
except StopIteration:
raise ValueError(
"Received an empty dataset. Argument `dataset` must "
"be a non-empty list/tuple of `numpy.ndarray` objects "
"or `tf.data.Dataset` objects."
)
for i, sample in enumerate(dataset_iterator):
if ensure_shape_similarity:
if first_sample_shape != np.array(sample).shape:
raise ValueError(
"All `dataset` samples must have same shape, "
f"Expected shape: {np.array(first_sample).shape} "
f"Received shape: {np.array(sample).shape} at index "
f"{i}."
)
if data_size_warning_flag:
if i % 10 == 0:
cur_time = time.time()
# warns user if the dataset is too large to iterate within 10s
if int(cur_time - start_time) > 10 and data_size_warning_flag:
warnings.warn(
"The dataset is taking longer than 10 seconds to "
"iterate over. This may be due to the size of the "
"dataset. Keep in mind that the `split_dataset` "
"utility is only for small in-memory dataset "
"(e.g. < 10,000 samples).",
category=ResourceWarning,
source="split_dataset",
)
data_size_warning_flag = False
yield sample
def is_torch_tensor(value):
if hasattr(value, "__class__"):
for parent in value.__class__.__mro__:
if parent.__name__ == "Tensor" and str(parent.__module__).endswith(
"torch"
):
return True
return False
def is_torch_dataset(dataset):
if hasattr(dataset, "__class__"):
for parent in dataset.__class__.__mro__:
if parent.__name__ == "Dataset" and str(
parent.__module__
).startswith("torch.utils.data"):
return True
return False
def _rescale_dataset_split_sizes(left_size, right_size, total_length):
"""Rescale the dataset split sizes.
We want to ensure that the sum of
the split sizes is equal to the total length of the dataset.
Args:
left_size: The size of the left dataset split.
right_size: The size of the right dataset split.
total_length: The total length of the dataset.
Returns:
tuple: A tuple of rescaled `left_size` and `right_size` integers.
"""
left_size_type = type(left_size)
right_size_type = type(right_size)
# check both left_size and right_size are integers or floats
if (left_size is not None and left_size_type not in [int, float]) and (
right_size is not None and right_size_type not in [int, float]
):
raise TypeError(
"Invalid `left_size` and `right_size` Types. Expected: "
"integer or float or None, Received: type(left_size)="
f"{left_size_type} and type(right_size)={right_size_type}"
)
# check left_size is a integer or float
if left_size is not None and left_size_type not in [int, float]:
raise TypeError(
"Invalid `left_size` Type. Expected: int or float or None, "
f"Received: type(left_size)={left_size_type}. "
)
# check right_size is a integer or float
if right_size is not None and right_size_type not in [int, float]:
raise TypeError(
"Invalid `right_size` Type. "
"Expected: int or float or None,"
f"Received: type(right_size)={right_size_type}."
)
# check left_size and right_size are non-zero
if left_size == 0 and right_size == 0:
raise ValueError(
"Both `left_size` and `right_size` are zero. "
"At least one of the split sizes must be non-zero."
)
# check left_size is non-negative and less than 1 and less than total_length
if (
left_size_type == int
and (left_size <= 0 or left_size >= total_length)
or left_size_type == float
and (left_size <= 0 or left_size >= 1)
):
raise ValueError(
"`left_size` should be either a positive integer "
f"smaller than {total_length}, or a float "
"within the range `[0, 1]`. Received: left_size="
f"{left_size}"
)
# check right_size is non-negative and less than 1 and less than
# total_length
if (
right_size_type == int
and (right_size <= 0 or right_size >= total_length)
or right_size_type == float
and (right_size <= 0 or right_size >= 1)
):
raise ValueError(
"`right_size` should be either a positive integer "
f"and smaller than {total_length} or a float "
"within the range `[0, 1]`. Received: right_size="
f"{right_size}"
)
# check sum of left_size and right_size is less than or equal to
# total_length
if (
right_size_type == left_size_type == float
and right_size + left_size > 1
):
raise ValueError(
"The sum of `left_size` and `right_size` is greater "
"than 1. It must be less than or equal to 1."
)
if left_size_type == float:
left_size = round(left_size * total_length)
elif left_size_type == int:
left_size = float(left_size)
if right_size_type == float:
right_size = round(right_size * total_length)
elif right_size_type == int:
right_size = float(right_size)
if left_size is None:
left_size = total_length - right_size
elif right_size is None:
right_size = total_length - left_size
if left_size + right_size > total_length:
raise ValueError(
"The sum of `left_size` and `right_size` should "
"be smaller than the {total_length}. "
f"Received: left_size + right_size = {left_size+right_size}"
f"and total_length = {total_length}"
)
for split, side in [(left_size, "left"), (right_size, "right")]:
if split == 0:
raise ValueError(
f"With `dataset` of length={total_length}, `left_size`="
f"{left_size} and `right_size`={right_size}."
f"Resulting {side} side dataset split will be empty. "
"Adjust any of the aforementioned parameters"
)
left_size, right_size = int(left_size), int(right_size)
return left_size, right_size
def _restore_dataset_from_list(
dataset_as_list, dataset_type_spec, original_dataset
):
"""Restore the dataset from the list of arrays."""
if dataset_type_spec in [tuple, list]:
return tuple(np.array(sample) for sample in zip(*dataset_as_list))
elif dataset_type_spec == tf.data.Dataset:
if isinstance(original_dataset.element_spec, dict):
restored_dataset = {}
for d in dataset_as_list:
for k, v in d.items():
if k not in restored_dataset:
restored_dataset[k] = [v]
else:
restored_dataset[k].append(v)
return restored_dataset
else:
return tuple(np.array(sample) for sample in zip(*dataset_as_list))
elif is_torch_dataset(original_dataset):
return tuple(np.array(sample) for sample in zip(*dataset_as_list))
return dataset_as_list
def is_batched(dataset):
"""Check if the `tf.data.Dataset` is batched."""
return hasattr(dataset, "_batch_size")
def get_batch_size(dataset):
"""Get the batch size of the dataset."""
if is_batched(dataset):
return dataset._batch_size
else:
return None
def _get_type_spec(dataset):
"""Get the type spec of the dataset."""
if isinstance(dataset, tuple):
return tuple
elif isinstance(dataset, list):
return list
elif isinstance(dataset, np.ndarray):
return np.ndarray
elif isinstance(dataset, dict):
return dict
elif isinstance(dataset, tf.data.Dataset):
return tf.data.Dataset
elif is_torch_dataset(dataset):
from torch.utils.data import Dataset as torchDataset
return torchDataset
else:
return None
def index_directory(
directory,
labels,
formats,
class_names=None,
shuffle=True,
seed=None,
follow_links=False,
):
"""List all files in `directory`, with their labels.
Args:
directory: Directory where the data is located.
If `labels` is `"inferred"`, it should contain
subdirectories, each containing files for a class.
Otherwise, the directory structure is ignored.
labels: Either `"inferred"`
(labels are generated from the directory structure),
`None` (no labels),
or a list/tuple of integer labels of the same size as the number
of valid files found in the directory.
Labels should be sorted according
to the alphanumeric order of the image file paths
(obtained via `os.walk(directory)` in Python).
formats: Allowlist of file extensions to index
(e.g. `".jpg"`, `".txt"`).
class_names: Only valid if `labels="inferred"`. This is the explicit
list of class names (must match names of subdirectories). Used
to control the order of the classes
(otherwise alphanumerical order is used).
shuffle: Whether to shuffle the data. Default: True.
If set to False, sorts the data in alphanumeric order.
seed: Optional random seed for shuffling.
follow_links: Whether to visits subdirectories pointed to by symlinks.
Returns:
tuple (file_paths, labels, class_names).
- file_paths: list of file paths (strings).
- labels: list of matching integer labels (same length as file_paths)
- class_names: names of the classes corresponding to these labels, in
order.
"""
if labels == "inferred":
subdirs = []
for subdir in sorted(tf.io.gfile.listdir(directory)):
if tf.io.gfile.isdir(tf.io.gfile.join(directory, subdir)):
if not subdir.startswith("."):
if subdir.endswith("/"):
subdir = subdir[:-1]
subdirs.append(subdir)
if class_names is not None:
if not set(class_names).issubset(set(subdirs)):
raise ValueError(
"The `class_names` passed did not match the "
"names of the subdirectories of the target directory. "
f"Expected: {subdirs} (or a subset of it), "
f"but received: class_names={class_names}"
)
subdirs = class_names # Keep provided order.
else:
# In the explicit/no-label cases, index from the parent directory down.
subdirs = [""]
if class_names is not None:
if labels is None:
raise ValueError(
"When `labels=None` (no labels), argument `class_names` "
"cannot be specified."
)
else:
raise ValueError(
"When argument `labels` is specified, argument "
"`class_names` cannot be specified (the `class_names` "
"will be the sorted list of labels)."
)
class_names = subdirs
class_indices = dict(zip(class_names, range(len(class_names))))
# Build an index of the files
# in the different class subfolders.
pool = ThreadPool()
results = []
filenames = []
for dirpath in (tf.io.gfile.join(directory, subdir) for subdir in subdirs):
results.append(
pool.apply_async(
index_subdirectory,
(dirpath, class_indices, follow_links, formats),
)
)
labels_list = []
for res in results:
partial_filenames, partial_labels = res.get()
labels_list.append(partial_labels)
filenames += partial_filenames
if labels == "inferred":
# Inferred labels.
i = 0
labels = np.zeros((len(filenames),), dtype="int32")
for partial_labels in labels_list:
labels[i : i + len(partial_labels)] = partial_labels
i += len(partial_labels)
elif labels is None:
class_names = None
else:
# Manual labels.
if len(labels) != len(filenames):
raise ValueError(
"Expected the lengths of `labels` to match the number "
"of files in the target directory. len(labels) is "
f"{len(labels)} while we found {len(filenames)} files "
f"in directory {directory}."
)
class_names = [str(label) for label in sorted(set(labels))]
if labels is None:
io_utils.print_msg(f"Found {len(filenames)} files.")
else:
io_utils.print_msg(
f"Found {len(filenames)} files belonging "
f"to {len(class_names)} classes."
)
pool.close()
pool.join()
file_paths = [tf.io.gfile.join(directory, fname) for fname in filenames]
if shuffle:
# Shuffle globally to erase macro-structure
if seed is None:
seed = np.random.randint(1e6)
rng = np.random.RandomState(seed)
rng.shuffle(file_paths)
if labels is not None:
rng = np.random.RandomState(seed)
rng.shuffle(labels)
return file_paths, labels, class_names
def iter_valid_files(directory, follow_links, formats):
if not follow_links:
walk = tf.io.gfile.walk(directory)
else:
walk = os.walk(directory, followlinks=follow_links)
for root, _, files in sorted(walk, key=lambda x: x[0]):
for fname in sorted(files):
if fname.lower().endswith(formats):
yield root, fname
def index_subdirectory(directory, class_indices, follow_links, formats):
"""Recursively walks directory and list image paths and their class index.
Args:
directory: string, target directory.
class_indices: dict mapping class names to their index.
follow_links: boolean, whether to recursively follow subdirectories
(if False, we only list top-level images in `directory`).
formats: Allowlist of file extensions to index (e.g. ".jpg", ".txt").
Returns:
tuple `(filenames, labels)`. `filenames` is a list of relative file
paths, and `labels` is a list of integer labels corresponding
to these files.
"""
dirname = os.path.basename(directory)
valid_files = iter_valid_files(directory, follow_links, formats)
labels = []
filenames = []
for root, fname in valid_files:
labels.append(class_indices[dirname])
absolute_path = tf.io.gfile.join(root, fname)
relative_path = tf.io.gfile.join(
dirname, os.path.relpath(absolute_path, directory)
)
filenames.append(relative_path)
return filenames, labels
def get_training_or_validation_split(samples, labels, validation_split, subset):
"""Potentially restict samples & labels to a training or validation split.
Args:
samples: List of elements.
labels: List of corresponding labels.
validation_split: Float, fraction of data to reserve for validation.
subset: Subset of the data to return.
Either `"training"`, `"validation"`, or `None`.
If `None`, we return all of the data.
Returns:
tuple (samples, labels), potentially restricted to the specified subset.
"""
if not validation_split:
return samples, labels
num_val_samples = int(validation_split * len(samples))
if subset == "training":
io_utils.print_msg(
f"Using {len(samples) - num_val_samples} " f"files for training."
)
samples = samples[:-num_val_samples]
if labels is not None:
labels = labels[:-num_val_samples]
elif subset == "validation":
io_utils.print_msg(f"Using {num_val_samples} files for validation.")
samples = samples[-num_val_samples:]
if labels is not None:
labels = labels[-num_val_samples:]
else:
raise ValueError(
'`subset` must be either "training" '
f'or "validation", received: {subset}'
)
return samples, labels
def labels_to_dataset(labels, label_mode, num_classes):
"""Create a `tf.data.Dataset` from the list/tuple of labels.
Args:
labels: list/tuple of labels to be converted into a `tf.data.Dataset`.
label_mode: String describing the encoding of `labels`. Options are:
- `"binary"` indicates that the labels (there can be only 2) are encoded
as `float32` scalars with values 0 or 1
(e.g. for `binary_crossentropy`).
- `"categorical"` means that the labels are mapped into a categorical
vector. (e.g. for `categorical_crossentropy` loss).
num_classes: number of classes of labels.
Returns:
A `tf.data.Dataset` instance.
"""
label_ds = tf.data.Dataset.from_tensor_slices(labels)
if label_mode == "binary":
label_ds = label_ds.map(
lambda x: tf.expand_dims(tf.cast(x, "float32"), axis=-1),
num_parallel_calls=tf.data.AUTOTUNE,
)
elif label_mode == "categorical":
label_ds = label_ds.map(
lambda x: tf.one_hot(x, num_classes),
num_parallel_calls=tf.data.AUTOTUNE,
)
return label_ds
def check_validation_split_arg(validation_split, subset, shuffle, seed):
"""Raise errors in case of invalid argument values.
Args:
validation_split: float between 0 and 1, fraction of data to reserve for
validation.
subset: One of `"training"`, `"validation"`, or `"both"`. Only used if
`validation_split` is set.
shuffle: Whether to shuffle the data. Either `True` or `False`.
seed: random seed for shuffling and transformations.
"""
if validation_split and not 0 < validation_split < 1:
raise ValueError(
"`validation_split` must be between 0 and 1, "
f"received: {validation_split}"
)
if (validation_split or subset) and not (validation_split and subset):
raise ValueError(
"If `subset` is set, `validation_split` must be set, and inversely."
)
if subset not in ("training", "validation", "both", None):
raise ValueError(
'`subset` must be either "training", '
f'"validation" or "both", received: {subset}'
)
if validation_split and shuffle and seed is None:
raise ValueError(
"If using `validation_split` and shuffling the data, you must "
"provide a `seed` argument, to make sure that there is no "
"overlap between the training and validation subset."
)