def setUp(self):
super(ReadInstructionTest, self).setUp()
self.splits = {
'train':
splits.SplitInfo(
name='train',
shard_lengths=[200],
num_bytes=0,
filename_template=_filename_template('train')),
'test':
splits.SplitInfo(
name='test',
shard_lengths=[101],
num_bytes=0,
filename_template=_filename_template('test')),
'validation':
splits.SplitInfo(
name='validation',
shard_lengths=[30],
num_bytes=0,
filename_template=_filename_template('validation')),
'dev-train':
splits.SplitInfo(
name='dev-train',
shard_lengths=[5, 5],
num_bytes=0,
filename_template=_filename_template('dev-train')),
}
def test_even_splits_subsplit():
split_infos = splits_lib.SplitDict(split_infos=[
splits_lib.SplitInfo(
name='train',
shard_lengths=[2, 3, 2, 3], # 10
num_bytes=0,
filename_template=_filename_template('train'),
),
splits_lib.SplitInfo(
name='test',
shard_lengths=[8],
num_bytes=0,
filename_template=_filename_template('test'),
),
])
# Test to split multiple splits
subsplits = subsplits_utils.even_splits('train+test[50%:]', 3)
expected = [
'train[:4]+test[4:6]',
'train[4:7]+test[6:7]',
'train[7:]+test[7:8]',
]
file_instructions = [split_infos[s].file_instructions for s in subsplits]
expected_file_instructions = [
split_infos[s].file_instructions for s in expected
]
assert file_instructions == expected_file_instructions
def test_set_splits_normal(self):
info = dataset_info.DatasetInfo(builder=self._builder)
split_info1 = splits_lib.SplitInfo(name="train",
shard_lengths=[1, 2],
num_bytes=0)
split_info2 = splits_lib.SplitInfo(name="test",
shard_lengths=[1],
num_bytes=0)
split_dict = splits_lib.SplitDict(
split_infos=[split_info1, split_info2])
info.set_splits(split_dict)
self.assertEqual(str(info.splits), str(split_dict))
self.assertEqual(str(info.as_proto.splits),
str([split_info1.to_proto(),
split_info2.to_proto()]))
def __init__(
self,
root_dir: str,
*,
shape: Optional[type_utils.Shape] = None,
dtype: Optional[tf.DType] = None,
):
"""Construct the `DatasetBuilder`.
Args:
root_dir: Path to the directory containing the images.
shape: Image shape forwarded to `tfds.features.Image`.
dtype: Image dtype forwarded to `tfds.features.Image`.
"""
self._image_shape = shape
self._image_dtype = dtype
super(ImageFolder, self).__init__()
self._data_dir = root_dir # Set data_dir to the existing dir.
# Extract the splits, examples, labels
root_dir = os.path.expanduser(root_dir)
self._split_examples, labels = _get_split_label_images(root_dir)
# Update DatasetInfo labels
self.info.features['label'].names = sorted(labels)
# Update DatasetInfo splits
split_dict = split_lib.SplitDict(self.name)
for split_name, examples in self._split_examples.items():
split_dict.add(split_lib.SplitInfo(
name=split_name,
shard_lengths=[len(examples)],
))
self.info.update_splits_if_different(split_dict)
def _initialize_split(split_name: str,
data_directory: Any,
ds_name: str,
filetype_suffix: str,
shard_lengths: Optional[List[int]] = None,
num_bytes: int = 0) -> Split:
"""Initializes a split.
Args:
split_name: name of the split.
data_directory: directory where the split data will be located.
ds_name: name of the dataset.
filetype_suffix: file format.
shard_lengths: if the split already has shards, it contains the list of the
shard lenghts. If None, it assumes that the split is empty.
num_bytes: number of bytes that have been written already.
Returns:
A Split.
"""
if not shard_lengths:
shard_lengths = []
filename_template = naming.ShardedFileTemplate(
dataset_name=ds_name,
data_dir=data_directory,
split=split_name,
filetype_suffix=filetype_suffix,
template='{DATASET}-{SPLIT}.{FILEFORMAT}-{SHARD_INDEX}',
)
return Split(info=splits_lib.SplitInfo(
name=split_name,
shard_lengths=shard_lengths,
num_bytes=num_bytes,
filename_template=filename_template),
complete_shards=len(shard_lengths),
ds_name=ds_name)
def test_missing_shard_lengths(self):
with self.assertRaisesWithPredicateMatch(ValueError, 'Shard empty.'):
split_info = [
splits.SplitInfo(name='train', shard_lengths=[]),
]
tfrecords_reader.make_file_instructions('mnist', split_info,
'train')
def test_nodata_instruction(self):
# Given instruction corresponds to no data.
with self.assertRaisesWithPredicateMatch(AssertionError,
'corresponds to no data!'):
train_info = splits.SplitInfo(name='train',
shard_lengths=[2, 3, 2, 3, 2])
self.reader.read('mnist', 'train[0:0]', [train_info])
def _build_from_generator(
self,
split_name: str,
generator: Iterable[KeyExample],
path: type_utils.PathLike,
) -> _SplitInfoFuture:
"""Split generator for example generators.
Args:
split_name: str,
generator: Iterable[KeyExample],
path: type_utils.PathLike,
Returns:
future: The future containing the `tfds.core.SplitInfo`.
"""
if self._max_examples_per_split is not None:
logging.warning('Splits capped at %s examples max.',
self._max_examples_per_split)
generator = itertools.islice(generator,
self._max_examples_per_split)
total_num_examples = self._max_examples_per_split
else:
# If dataset info has been pre-downloaded from the internet,
# we can use the pre-computed number of example for the progression bar.
split_info = self._split_dict.get(split_name)
if split_info and split_info.num_examples:
total_num_examples = split_info.num_examples
else:
total_num_examples = None
writer = tfrecords_writer.Writer(
example_specs=self._features.get_serialized_info(),
path=path,
hash_salt=split_name,
file_format=self._file_format,
)
for key, example in utils.tqdm(
generator,
desc=f'Generating {split_name} examples...',
unit=' examples',
total=total_num_examples,
leave=False,
):
try:
example = self._features.encode_example(example)
except Exception as e: # pylint: disable=broad-except
utils.reraise(e,
prefix=f'Failed to encode example:\n{example}\n')
writer.write(key, example)
shard_lengths, total_size = writer.finalize()
split_info = splits_lib.SplitInfo(
name=split_name,
shard_lengths=shard_lengths,
num_bytes=total_size,
)
return _SplitInfoFuture(lambda: split_info)
def test_even_splits_add():
# Compatibility of even_splits with other splits
split_infos = splits_lib.SplitDict(split_infos=[
splits_lib.SplitInfo(
name='train',
shard_lengths=[2, 3, 2, 3], # 10
num_bytes=0,
filename_template=_filename_template('train'),
),
splits_lib.SplitInfo(
name='test',
shard_lengths=[8],
num_bytes=0,
filename_template=_filename_template('test'),
),
splits_lib.SplitInfo(
name='validation',
shard_lengths=[8],
filename_template=_filename_template('validation'),
num_bytes=0,
),
])
# Test to split multiple splits
split = subsplits_utils.even_splits('train', 3, drop_remainder=True)[0]
split = split + 'test'
expected = 'train[:3]+test'
file_instructions = split_infos[split].file_instructions
expected_file_instructions = (split_infos[expected].file_instructions)
assert file_instructions == expected_file_instructions
# Test nested `even_splits`
splits = subsplits_utils.even_splits('validation', n=2)
splits = subsplits_utils.even_splits(splits[1], n=2)
assert (split_infos[splits[0]].file_instructions ==
split_infos['validation[4:6]'].file_instructions)
assert (split_infos[splits[1]].file_instructions ==
split_infos['validation[6:8]'].file_instructions)
def _get_files(self, instruction):
split_infos = [
splits.SplitInfo(
name='train',
shard_lengths=[3, 2, 3, 2, 3], # 13 examples.
num_bytes=0,
filename_template=_filename_template(
dataset_name='mnist', split='train'),
)
]
splits_dict = splits.SplitDict(split_infos=split_infos)
return splits_dict[instruction].file_instructions
def _resolve_future():
if self._in_contextmanager:
raise AssertionError('`future.result()` should be called after the '
'`maybe_beam_pipeline` contextmanager.')
logging.info('Retrieving split info for %s...', split_name)
shard_lengths, total_size = beam_writer.finalize()
return splits_lib.SplitInfo(
name=split_name,
shard_lengths=shard_lengths,
num_bytes=total_size,
filename_template=filename_template,
)
def test_missing_shard_lengths(self):
with self.assertRaisesWithPredicateMatch(ValueError, 'Shard empty.'):
filename_template = _filename_template(
split='train', dataset_name='mnist')
split_infos = [
splits.SplitInfo(
name='train',
shard_lengths=[],
num_bytes=0,
filename_template=filename_template),
]
splits_dict = splits.SplitDict(split_infos=split_infos)
_ = splits_dict['train'].file_instructions
def close_shard(self) -> None:
"""Finalizes a shard and updates the split metadata accordingly."""
if not self.current_shard:
return
self.current_shard.close_writer()
self.info = splits_lib.SplitInfo(
name=self.info.name,
shard_lengths=self.info.shard_lengths +
[self.current_shard.num_examples],
num_bytes=self.info.num_bytes + self.current_shard.num_bytes,
filename_template=self.info.filename_template)
self.complete_shards += 1
self.current_shard = None
def test_nodata_instruction(self):
# Given instruction corresponds to no data.
with self.assertRaisesWithPredicateMatch(ValueError,
'corresponds to no data!'):
train_info = splits.SplitInfo(
name='train',
shard_lengths=[2, 3, 2, 3, 2],
num_bytes=0,
filename_template=self._filename_template(split='train'),
)
self.reader.read(
instructions='train[0:0]',
split_infos=[train_info],
)
def test_set_splits_incorrect_dataset_name(self):
info = dataset_info.DatasetInfo(builder=self._builder)
split_info1 = splits_lib.SplitInfo(
name="train",
shard_lengths=[1, 2],
num_bytes=0,
filename_template=naming.ShardedFileTemplate(
dataset_name="some_other_dataset",
split="train",
data_dir=info.data_dir,
filetype_suffix="tfrecord"))
split_dict = splits_lib.SplitDict(split_infos=[split_info1])
with pytest.raises(AssertionError,
match="SplitDict contains SplitInfo for split"):
info.set_splits(split_dict)
def _write_tfrecord(self, split_name, shards_number, records):
path = os.path.join(self.tmp_dir, 'mnist-%s.tfrecord' % split_name)
num_examples = len(records)
with absltest.mock.patch.object(tfrecords_writer, '_get_number_shards',
return_value=shards_number):
shard_specs = tfrecords_writer._get_shard_specs(
num_examples, 0, [num_examples], path)
serialized_records = [six.b(rec) for rec in records]
for shard_spec in shard_specs:
_write_tfrecord_from_shard_spec(
shard_spec, lambda unused_i: iter(serialized_records))
return splits.SplitInfo(
name=split_name,
shard_lengths=[int(s.examples_number) for s in shard_specs],
)
def _merge_shard_info(shard_infos: List[_ShardInfo]) -> split_lib.SplitInfo:
"""Merge all shard info from one splits and returns the json.
Args:
shard_infos: The shard infos from all shards.
Returns:
The json SplitInfo proto
"""
split_name, = {s.file_info.split for s in shard_infos}
shard_infos = sorted(shard_infos, key=lambda s: s.file_info.shard_index)
return split_lib.SplitInfo(
name=split_name,
shard_lengths=[s.num_examples for s in shard_infos],
num_bytes=sum(s.bytes_size for s in shard_infos),
)
def __init__(self, root_dir: str):
# Extract the splits, examples
root_dir = os.path.expanduser(root_dir)
self._split_examples, self._languages = _get_split_language_examples(
root_dir)
super(TranslateFolder, self).__init__()
# Reset `_data_dir` as it should not change to DATA_DIR/Version
self._data_dir = root_dir
# Update DatasetInfo splits
split_dict = split_lib.SplitDict(self.name)
for split_name, examples in self._split_examples.items():
split_dict.add(
split_lib.SplitInfo(
name=split_name,
shard_lengths=[len(next(iter(examples.values())))],
))
self.info.update_splits_if_different(split_dict)
def _write_tfrecord(self, split_name, shards_number, records):
filename_template = self._filename_template(split=split_name)
num_examples = len(records)
with mock.patch.object(
tfrecords_writer, '_get_number_shards', return_value=shards_number):
shard_specs = tfrecords_writer._get_shard_specs(
num_examples=num_examples,
total_size=0,
bucket_lengths=[num_examples],
filename_template=filename_template)
serialized_records = [(key, six.b(rec)) for key, rec in enumerate(records)]
for shard_spec in shard_specs:
_write_tfrecord_from_shard_spec(shard_spec,
lambda unused_i: iter(serialized_records))
return splits.SplitInfo(
name=split_name,
shard_lengths=[int(s.examples_number) for s in shard_specs],
num_bytes=0,
filename_template=filename_template)
def test_even_splits(num_examples, n, drop_remainder, expected):
split_infos = splits_lib.SplitDict(split_infos=[
splits_lib.SplitInfo(
name='train',
shard_lengths=[num_examples],
num_bytes=0,
filename_template=_filename_template('train'),
),
])
subsplits = subsplits_utils.even_splits('train',
n,
drop_remainder=drop_remainder)
file_instructions = [split_infos[s].file_instructions for s in subsplits]
expected_file_instructions = [
split_infos[f'train{s}'].file_instructions for s in expected
]
assert file_instructions == expected_file_instructions
def __init__(self, root_dir: str):
# Extract the splits, examples
root_dir = os.path.expanduser(root_dir)
self._split_examples, self._languages = _get_split_language_examples(
root_dir)
super(TranslateFolder, self).__init__()
# Reset `_data_dir` as it should not change to DATA_DIR/Version
self._data_dir = root_dir
# Update DatasetInfo splits
split_infos = [
split_lib.SplitInfo( # pylint: disable=g-complex-comprehension
name=split_name,
shard_lengths=[len(next(iter(examples.values())))],
num_bytes=0,
) for split_name, examples in self._split_examples.items()
]
split_dict = split_lib.SplitDict(split_infos, dataset_name=self.name)
self.info.set_splits(split_dict)
def __init__(self, root_dir: str):
super(ImageFolder, self).__init__()
self._data_dir = root_dir # Set data_dir to the existing dir.
# Extract the splits, examples, labels
root_dir = os.path.expanduser(root_dir)
self._split_examples, labels = _get_split_label_images(root_dir)
# Update DatasetInfo labels
self.info.features['label'].names = sorted(labels)
# Update DatasetInfo splits
split_dict = split_lib.SplitDict(self.name)
for split_name, examples in self._split_examples.items():
split_dict.add(
split_lib.SplitInfo(
name=split_name,
shard_lengths=[len(examples)],
))
self.info.update_splits_if_different(split_dict)
def _merge_shard_info(
shard_infos: List[_ShardInfo],
filename_template: naming.ShardedFileTemplate,
) -> split_lib.SplitInfo:
"""Merge all shard info from one splits and returns the SplitInfo.
Args:
shard_infos: The shard infos from all shards.
filename_template: filename template of the splits.
Returns:
The json SplitInfo proto
"""
split_name, = {s.file_info.split for s in shard_infos}
shard_infos = sorted(shard_infos, key=lambda s: s.file_info.shard_index)
filename_template = filename_template.replace(split=split_name)
return split_lib.SplitInfo(
name=split_name,
shard_lengths=[s.num_examples for s in shard_infos],
num_bytes=sum(s.bytes_size for s in shard_infos),
filename_template=filename_template,
)
class GetDatasetFilesTest(testing.TestCase):
SPLIT_INFOS = {
'train':
splits.SplitInfo(
name='train',
shard_lengths=[3, 2, 3, 2, 3], # 13 examples.
num_bytes=0,
),
}
PATH_PATTERN = 'mnist-train.tfrecord-0000%d-of-00005'
def _get_files(self, instruction):
file_instructions = tfrecords_reader._make_file_instructions_from_absolutes(
name='mnist',
split_infos=self.SPLIT_INFOS,
absolute_instructions=[instruction],
)
return file_instructions
def test_no_skip_no_take(self):
instruction = tfrecords_reader._AbsoluteInstruction('train', None, None)
files = self._get_files(instruction)
self.assertEqual(files, [
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % i, skip=0, take=-1, num_examples=n)
for i, n in enumerate([3, 2, 3, 2, 3])
])
def test_skip(self):
# One file is not taken, one file is partially taken.
instruction = tfrecords_reader._AbsoluteInstruction('train', 4, None)
files = self._get_files(instruction)
self.assertEqual(files, [
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 1, skip=1, take=-1, num_examples=1),
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 2, skip=0, take=-1, num_examples=3),
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 3, skip=0, take=-1, num_examples=2),
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 4, skip=0, take=-1, num_examples=3),
])
def test_take(self):
# Two files are not taken, one file is partially taken.
instruction = tfrecords_reader._AbsoluteInstruction('train', None, 6)
files = self._get_files(instruction)
self.assertEqual(files, [
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 0, skip=0, take=-1, num_examples=3),
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 1, skip=0, take=-1, num_examples=2),
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 2, skip=0, take=1, num_examples=1),
])
def test_skip_take1(self):
# A single shard with both skip and take.
instruction = tfrecords_reader._AbsoluteInstruction('train', 1, 2)
files = self._get_files(instruction)
self.assertEqual(files, [
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 0, skip=1, take=1, num_examples=1),
])
def test_skip_take2(self):
# 2 elements in across two shards are taken in middle.
instruction = tfrecords_reader._AbsoluteInstruction('train', 7, 9)
files = self._get_files(instruction)
self.assertEqual(files, [
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 2, skip=2, take=-1, num_examples=1),
shard_utils.FileInstruction(
filename=self.PATH_PATTERN % 3, skip=0, take=1, num_examples=1),
])
def test_touching_boundaries(self):
# Nothing to read.
instruction = tfrecords_reader._AbsoluteInstruction('train', 0, 0)
files = self._get_files(instruction)
self.assertEqual(files, [])
instruction = tfrecords_reader._AbsoluteInstruction('train', None, 0)
files = self._get_files(instruction)
self.assertEqual(files, [])
instruction = tfrecords_reader._AbsoluteInstruction('train', 3, 3)
files = self._get_files(instruction)
self.assertEqual(files, [])
instruction = tfrecords_reader._AbsoluteInstruction('train', 13, None)
files = self._get_files(instruction)
self.assertEqual(files, [])
def test_missing_shard_lengths(self):
with self.assertRaisesWithPredicateMatch(ValueError, 'Shard empty.'):
split_info = [
splits.SplitInfo(name='train', shard_lengths=[], num_bytes=0),
]
tfrecords_reader.make_file_instructions('mnist', split_info, 'train')
def _build_from_generator(
self,
split_name: str,
generator: Iterable[KeyExample],
filename_template: naming.ShardedFileTemplate,
disable_shuffling: bool,
) -> _SplitInfoFuture:
"""Split generator for example generators.
Args:
split_name: str,
generator: Iterable[KeyExample],
filename_template: Template to format the filename for a shard.
disable_shuffling: Specifies whether to shuffle the examples,
Returns:
future: The future containing the `tfds.core.SplitInfo`.
"""
if self._max_examples_per_split is not None:
logging.warning('Splits capped at %s examples max.',
self._max_examples_per_split)
generator = itertools.islice(generator, self._max_examples_per_split)
total_num_examples = self._max_examples_per_split
else:
# If dataset info has been pre-downloaded from the internet,
# we can use the pre-computed number of example for the progression bar.
split_info = self._split_dict.get(split_name)
if split_info and split_info.num_examples:
total_num_examples = split_info.num_examples
else:
total_num_examples = None
writer = writer_lib.Writer(
serializer=example_serializer.ExampleSerializer(
self._features.get_serialized_info()),
filename_template=filename_template,
hash_salt=split_name,
disable_shuffling=disable_shuffling,
# TODO(weide) remove this because it's already in filename_template?
file_format=self._file_format,
)
for key, example in utils.tqdm(
generator,
desc=f'Generating {split_name} examples...',
unit=' examples',
total=total_num_examples,
leave=False,
):
try:
example = self._features.encode_example(example)
except Exception as e: # pylint: disable=broad-except
utils.reraise(e, prefix=f'Failed to encode example:\n{example}\n')
writer.write(key, example)
shard_lengths, total_size = writer.finalize()
split_info = splits_lib.SplitInfo(
name=split_name,
shard_lengths=shard_lengths,
num_bytes=total_size,
filename_template=filename_template,
)
return _SplitInfoFuture(lambda: split_info)
class ReaderTest(testing.TestCase):
SPLIT_INFOS = [
splits.SplitInfo(name='train', shard_lengths=[2, 3, 2, 3,
2]), # 12 ex.
splits.SplitInfo(name='test', shard_lengths=[2, 3, 2]), # 7 ex.
]
def setUp(self):
super(ReaderTest, self).setUp()
with absltest.mock.patch.object(example_parser, 'ExampleParser',
testing.DummyParser):
self.reader = tfrecords_reader.Reader(self.tmp_dir, 'some_spec')
def _write_tfrecord(self, split_name, shards_number, records):
path = os.path.join(self.tmp_dir, 'mnist-%s.tfrecord' % split_name)
writer = tfrecords_writer._TFRecordWriter(path, len(records),
shards_number)
for rec in records:
writer.write(six.b(rec))
with absltest.mock.patch.object(tfrecords_writer,
'_get_number_shards',
return_value=shards_number):
writer.finalize()
def _write_tfrecords(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
self._write_tfrecord('test', 3, 'mnopqrs')
def test_nodata_instruction(self):
# Given instruction corresponds to no data.
with self.assertRaisesWithPredicateMatch(AssertionError,
'corresponds to no data!'):
self.reader.read('mnist', 'train[0:0]', self.SPLIT_INFOS)
def test_noskip_notake(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
ds = self.reader.read('mnist', 'train', self.SPLIT_INFOS)
read_data = list(tfds.as_numpy(ds))
self.assertEqual(read_data, [six.b(l) for l in 'abcdefghijkl'])
def test_overlap(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
ds = self.reader.read('mnist', 'train+train[:2]', self.SPLIT_INFOS)
read_data = list(tfds.as_numpy(ds))
self.assertEqual(read_data, [six.b(l) for l in 'abcdefghijklab'])
def test_complex(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
self._write_tfrecord('test', 3, 'mnopqrs')
ds = self.reader.read('mnist', 'train[1:-1]+test[:-50%]',
self.SPLIT_INFOS)
read_data = list(tfds.as_numpy(ds))
self.assertEqual(read_data, [six.b(l) for l in 'bcdefghijkmno'])
def test_4fold(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
instructions = [
tfrecords_reader.ReadInstruction('train',
from_=k,
to=k + 25,
unit='%')
for k in range(0, 100, 25)
]
tests = self.reader.read('mnist', instructions, self.SPLIT_INFOS)
instructions = [
(tfrecords_reader.ReadInstruction('train', to=k, unit='%') +
tfrecords_reader.ReadInstruction('train', from_=k + 25, unit='%'))
for k in range(0, 100, 25)
]
trains = self.reader.read('mnist', instructions, self.SPLIT_INFOS)
read_tests = [list(r) for r in tfds.as_numpy(tests)]
read_trains = [list(r) for r in tfds.as_numpy(trains)]
self.assertEqual(read_tests, [[b'a', b'b', b'c'], [b'd', b'e', b'f'],
[b'g', b'h', b'i'], [b'j', b'k', b'l']])
self.assertEqual(
read_trains,
[[b'd', b'e', b'f', b'g', b'h', b'i', b'j', b'k', b'l'],
[b'a', b'b', b'c', b'g', b'h', b'i', b'j', b'k', b'l'],
[b'a', b'b', b'c', b'd', b'e', b'f', b'j', b'k', b'l'],
[b'a', b'b', b'c', b'd', b'e', b'f', b'g', b'h', b'i']])
class ReaderTest(testing.TestCase):
SPLIT_INFOS = [
splits.SplitInfo(name='train', shard_lengths=[2, 3, 2, 3,
2]), # 12 ex.
splits.SplitInfo(name='test', shard_lengths=[2, 3, 2]), # 7 ex.
]
def setUp(self):
super(ReaderTest, self).setUp()
with absltest.mock.patch.object(example_parser, 'ExampleParser',
testing.DummyParser):
self.reader = tfrecords_reader.Reader(self.tmp_dir, 'some_spec')
self.reader.read = functools.partial(
self.reader.read,
read_config=read_config_lib.ReadConfig(),
shuffle_files=False,
)
def _write_tfrecord(self, split_name, shards_number, records):
path = os.path.join(self.tmp_dir, 'mnist-%s.tfrecord' % split_name)
num_examples = len(records)
with absltest.mock.patch.object(tfrecords_writer,
'_get_number_shards',
return_value=shards_number):
shard_specs = tfrecords_writer._get_shard_specs(
num_examples, 0, [num_examples], path)
serialized_records = [six.b(rec) for rec in records]
for shard_spec in shard_specs:
tfrecords_writer._write_tfrecord_from_shard_spec(
shard_spec, lambda unused_i: iter(serialized_records))
def _write_tfrecords(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
self._write_tfrecord('test', 3, 'mnopqrs')
def test_nodata_instruction(self):
# Given instruction corresponds to no data.
with self.assertRaisesWithPredicateMatch(AssertionError,
'corresponds to no data!'):
self.reader.read('mnist', 'train[0:0]', self.SPLIT_INFOS)
def test_noskip_notake(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
ds = self.reader.read('mnist', 'train', self.SPLIT_INFOS)
read_data = list(tfds.as_numpy(ds))
self.assertEqual(read_data, [six.b(l) for l in 'abcdefghijkl'])
def test_overlap(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
ds = self.reader.read('mnist', 'train+train[:2]', self.SPLIT_INFOS)
read_data = list(tfds.as_numpy(ds))
self.assertEqual(read_data, [six.b(l) for l in 'abcdefghijklab'])
def test_complex(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
self._write_tfrecord('test', 3, 'mnopqrs')
ds = self.reader.read('mnist', 'train[1:-1]+test[:-50%]',
self.SPLIT_INFOS)
read_data = list(tfds.as_numpy(ds))
self.assertEqual(read_data, [six.b(l) for l in 'bcdefghijkmno'])
def test_shuffle_files(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
ds = self.reader.read('mnist',
'train',
self.SPLIT_INFOS,
shuffle_files=True)
shards = [ # The shards of the dataset:
[b'a', b'b'],
[b'c', b'd', b'e'],
[b'f', b'g'],
[b'h', b'i', b'j'],
[b'k', b'l'],
]
# The various orders in which the dataset can be read:
expected_permutations = [
tuple(sum(shard, [])) for shard in itertools.permutations(shards)
]
ds = ds.batch(12).repeat(100)
read_data = set(tuple(e) for e in tfds.as_numpy(ds))
for batch in read_data:
self.assertIn(batch, expected_permutations)
# There are theoritically 5! (=120) different arrangements, but we would
# need too many repeats to be sure to get them.
self.assertGreater(len(set(read_data)), 10)
def test_shuffle_deterministic(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
read_config = read_config_lib.ReadConfig(shuffle_seed=123, )
ds = self.reader.read('mnist',
'train',
self.SPLIT_INFOS,
read_config=read_config,
shuffle_files=True)
ds_values = list(tfds.as_numpy(ds))
# Check that shuffle=True with a seed provides deterministic results.
self.assertEqual(ds_values, [
b'a', b'b', b'k', b'l', b'h', b'i', b'j', b'c', b'd', b'e', b'f',
b'g'
])
def test_4fold(self):
self._write_tfrecord('train', 5, 'abcdefghijkl')
instructions = [
tfrecords_reader.ReadInstruction('train',
from_=k,
to=k + 25,
unit='%')
for k in range(0, 100, 25)
]
tests = self.reader.read('mnist', instructions, self.SPLIT_INFOS)
instructions = [
(tfrecords_reader.ReadInstruction('train', to=k, unit='%') +
tfrecords_reader.ReadInstruction('train', from_=k + 25, unit='%'))
for k in range(0, 100, 25)
]
trains = self.reader.read('mnist', instructions, self.SPLIT_INFOS)
read_tests = [list(r) for r in tfds.as_numpy(tests)]
read_trains = [list(r) for r in tfds.as_numpy(trains)]
self.assertEqual(read_tests, [[b'a', b'b', b'c'], [b'd', b'e', b'f'],
[b'g', b'h', b'i'], [b'j', b'k', b'l']])
self.assertEqual(
read_trains,
[[b'd', b'e', b'f', b'g', b'h', b'i', b'j', b'k', b'l'],
[b'a', b'b', b'c', b'g', b'h', b'i', b'j', b'k', b'l'],
[b'a', b'b', b'c', b'd', b'e', b'f', b'j', b'k', b'l'],
[b'a', b'b', b'c', b'd', b'e', b'f', b'g', b'h', b'i']])
def split_info_for(name: str, shard_lengths, template) -> splits.SplitInfo:
return splits.SplitInfo(
name=name,
shard_lengths=shard_lengths,
num_bytes=0,
filename_template=template)