def _RawRecordToRecordBatchInternal(self,
batch_size: Optional[int] = None
) -> beam.PTransform:
@beam.typehints.with_input_types(List[bytes])
@beam.typehints.with_output_types(pa.RecordBatch)
def _PTransformFn(raw_records_pcoll: beam.pvalue.PCollection):
"""Returns RecordBatch of csv lines."""
# Decode raw csv lines to record batches.
record_batches = (
raw_records_pcoll
| "CSVToRecordBatch" >> csv_decoder.CSVToRecordBatch(
column_names=self._column_names,
delimiter=self._delimiter,
skip_blank_lines=self._skip_blank_lines,
schema=self._schema,
desired_batch_size=batch_size,
multivalent_columns=self._multivalent_columns,
secondary_delimiter=self._secondary_delimiter,
produce_large_types=self._can_produce_large_types,
raw_record_column_name=self._raw_record_column_name))
return record_batches
return beam.ptransform_fn(_PTransformFn)()
def RawRecordBeamSource(self) -> beam.PTransform:
"""Returns a PTransform that produces a PCollection[bytes].
Used together with RawRecordToRecordBatch(), it allows getting both the
PCollection of the raw records and the PCollection of the RecordBatch from
the same source. For example:
record_batch = pipeline | tfxio.BeamSource()
raw_record = pipeline | tfxio.RawRecordBeamSource()
would result in the files being read twice, while the following would only
read once:
raw_record = pipeline | tfxio.RawRecordBeamSource()
record_batch = raw_record | tfxio.RawRecordToRecordBatch()
"""
@beam.typehints.with_input_types(Any)
@beam.typehints.with_output_types(bytes)
def _PTransformFn(pcoll_or_pipeline: Any):
return (pcoll_or_pipeline
| "ReadRawRecords" >> self._RawRecordBeamSourceInternal()
|
"CollectRawRecordTelemetry" >> telemetry.ProfileRawRecords(
self._telemetry_descriptors, self._logical_format,
self._physical_format))
return beam.ptransform_fn(_PTransformFn)()
def _SerializedExamplesSource(self) -> beam.PTransform:
@beam.typehints.with_input_types(bytes)
@beam.typehints.with_output_types(bytes)
def _PTransformFn(raw_records_pcoll: beam.pvalue.PCollection):
return raw_records_pcoll
return beam.ptransform_fn(_PTransformFn)()
def RawRecordToRecordBatch(self,
batch_size: Optional[int] = None
) -> beam.PTransform:
"""Returns a PTransform that converts raw records to Arrow RecordBatches.
The input PCollection must be from self.RawRecordBeamSource() (also see
the documentation for that method).
Args:
batch_size: if not None, the `pa.RecordBatch` produced will be of the
specified size. Otherwise it's automatically tuned by Beam.
"""
@beam.typehints.with_input_types(bytes)
@beam.typehints.with_output_types(pa.RecordBatch)
def _PTransformFn(pcoll: beam.pvalue.PCollection):
return (pcoll
| "RawRecordToRecordBatch" >>
self._RawRecordToRecordBatchInternal(batch_size)
| "CollectRecordBatchTelemetry" >>
telemetry.ProfileRecordBatches(self._telemetry_descriptors,
self._logical_format,
self._physical_format))
return beam.ptransform_fn(_PTransformFn)()
def _RawRecordBeamSourceInternal(self) -> beam.PTransform:
@beam.typehints.with_input_types(bytes)
@beam.typehints.with_output_types(bytes)
def _PTransformFn(raw_records_pcoll: beam.pvalue.PCollection):
return raw_records_pcoll
return beam.ptransform_fn(_PTransformFn)()
def testOverridableRecordBasedTFXIO(self):
tmp_dir = tempfile.mkdtemp(dir=FLAGS.test_tmpdir)
file1 = os.path.join(tmp_dir, "tfrecord1")
file1_records = [b"aa", b"bb"]
_WriteTfRecord(file1, file1_records)
def _CheckRecords(actual, expected):
for a, e in zip(actual, expected):
self.assertDictEqual(a.to_pydict(), e)
@beam.typehints.with_input_types(Any)
@beam.typehints.with_output_types(bytes)
def _RawRecordBeamSource(pipeline: Any):
return pipeline | beam.io.ReadFromTFRecord(file1 + "*")
@beam.typehints.with_input_types(bytes)
@beam.typehints.with_output_types(pa.RecordBatch)
def _RawRecordsToRecordBatch(pcoll, batch_size):
batch_size = 1 if not batch_size else batch_size
class _CreateRBDoFn(beam.DoFn):
def process(self, examples):
return [
pa.RecordBatch.from_arrays([pa.array(examples)], ["column_name"])
]
return (pcoll | beam.BatchElements(batch_size)
| beam.ParDo(_CreateRBDoFn()))
tfxio = record_based_tfxio.OverridableRecordBasedTFXIO(
telemetry_descriptors=None,
logical_format="tfrecord",
physical_format="tf_example",
raw_record_beam_source=beam.ptransform_fn(_RawRecordBeamSource),
raw_record_to_record_batch=beam.ptransform_fn(_RawRecordsToRecordBatch))
expected = [{"column_name": [b"aa"]}, {"column_name": [b"bb"]}]
with beam.Pipeline() as p:
record_pcoll = p | tfxio.BeamSource()
beam_test_util.assert_that(
record_pcoll,
lambda actual: _CheckRecords(actual, expected))
def BeamSource(self, batch_size: Optional[int] = None) -> beam.PTransform:
@beam.typehints.with_input_types(Any)
@beam.typehints.with_output_types(pa.RecordBatch)
def _PTransformFn(pcoll_or_pipeline: Any):
"""Converts raw records to RecordBatches."""
return (pcoll_or_pipeline
| "RawRecordBeamSource" >> self.RawRecordBeamSource()
| "RawRecordToRecordBatch" >>
self.RawRecordToRecordBatch(batch_size))
return beam.ptransform_fn(_PTransformFn)()
def _RawRecordBeamSourceInternal(self) -> beam.PTransform:
@beam.typehints.with_input_types(beam.Pipeline)
@beam.typehints.with_output_types(bytes)
def _PTransformFn(pipeline: beam.pvalue.PCollection):
return pipeline | "ReadFromTFRecord" >> beam.io.ReadFromTFRecord(
self._file_pattern,
coder=beam.coders.BytesCoder(),
# TODO(b/114938612): Eventually remove this override.
validate=False)
return beam.ptransform_fn(_PTransformFn)()
def _RawRecordToRecordBatchInternal(self,
batch_size: Optional[int] = None
) -> beam.PTransform:
@beam.typehints.with_input_types(bytes)
@beam.typehints.with_output_types(pa.RecordBatch)
def _PTransformFn(raw_record_pcoll: beam.pvalue.PCollection):
return (raw_record_pcoll
| "Batch" >> beam.BatchElements(
**batch_util.GetBatchElementsKwargs(batch_size))
| "ToRecordBatch" >> beam.Map(_BatchedRecordsToArrow,
self.raw_record_column_name))
return beam.ptransform_fn(_PTransformFn)()
def _RawRecordToRecordBatchInternal(
self, batch_size: Optional[int]) -> beam.PTransform:
@beam.typehints.with_input_types(bytes)
@beam.typehints.with_output_types(pa.RecordBatch)
def _PTransformFn(raw_records_pcoll: beam.pvalue.PCollection):
return (raw_records_pcoll
| "BatchElements" >> beam.BatchElements(
**batch_util.GetBatchElementsKwargs(batch_size))
| "Decode" >> beam.ParDo(
_RecordsToRecordBatch(self._saved_decoder_path,
self.raw_record_column_name,
self._can_produce_large_types)))
return beam.ptransform_fn(_PTransformFn)()
def _split_generators(self, dl_manager, pipeline):
del dl_manager
examples = (pipeline | beam.Create(range(1000))
| beam.Map(_gen_example))
# Wrap the pipeline inside a ptransform_fn to add `'label' >> ` to avoid
# duplicated PTransform nodes names.
generate_examples = beam.ptransform_fn(self._generate_examples)
return {
'train':
examples | 'train' >> generate_examples(num_examples=1000),
'test': examples | 'test' >> generate_examples(num_examples=725),
}
def _RawRecordToRecordBatchInternal(self,
batch_size: Optional[int] = None
) -> beam.PTransform:
@beam.typehints.with_input_types(bytes)
@beam.typehints.with_output_types(pa.RecordBatch)
def ptransform_fn(raw_records_pcoll: beam.pvalue.PCollection):
return (raw_records_pcoll
| "Batch" >> beam.BatchElements(
**batch_util.GetBatchElementsKwargs(batch_size))
| "Decode" >> beam.ParDo(
_DecodeBatchExamplesDoFn(self._GetSchemaForDecoding(),
self.raw_record_column_name)))
return beam.ptransform_fn(ptransform_fn)()
def _RawRecordToRecordBatchInternal(
self, batch_size: Optional[int]) -> beam.PTransform:
@beam.typehints.with_input_types(bytes)
@beam.typehints.with_output_types(pa.RecordBatch)
def _PTransformFn(raw_records_pcoll: beam.pvalue.PCollection):
return (
raw_records_pcoll
| "BatchElements" >> beam.BatchElements(
**batch_util.GetBatchElementsKwargs(batch_size))
| "Decode" >> beam.ParDo(
_RecordsToRecordBatch(
self._saved_decoder_path, self.telemetry_descriptors,
shared.Shared() if self._use_singleton_decoder else
None, self.raw_record_column_name,
self._record_index_column_name)))
return beam.ptransform_fn(_PTransformFn)()
def BeamSource(self, batch_size: Optional[int] = None) -> beam.PTransform:
@beam.typehints.with_input_types(Union[beam.PCollection,
beam.Pipeline])
@beam.typehints.with_output_types(pa.RecordBatch)
def _PTransformFn(pcoll_or_pipeline: Any):
"""Reads Parquet tables and converts to RecordBatches."""
return (pcoll_or_pipeline
| "ParquetBeamSource" >> beam.io.ReadFromParquetBatched(
file_pattern=self._file_pattern,
min_bundle_size=self._min_bundle_size,
validate=self._validate,
columns=self._column_names) | "ToRecordBatch" >>
beam.FlatMap(self._TableToRecordBatch, batch_size)
| "CollectRecordBatchTelemetry"
>> telemetry.ProfileRecordBatches(
self._telemetry_descriptors, _PARQUET_FORMAT,
_PARQUET_FORMAT))
return beam.ptransform_fn(_PTransformFn)()
def normalize_legacy_split_generators(
self,
split_generators: Union[Dict[str, SplitGenerator],
List[SplitGeneratorLegacy]],
generator_fn: Callable[..., Any],
is_beam: bool,
) -> Dict[str, SplitGenerator]:
"""Normalize legacy split API into new dict[split_name, generator].
This function convert the legacy `List[tfds.core.SplitGenerator]` into
the new `{'split_name': generator}` structure.
Could be removed if all datasets were updated.
Args:
split_generators: Either legacy or new split_generators
generator_fn: The `GeneratorBasedBuilder._generate_examples` function.
is_beam: `True` if using legacy `tfds.core.BeamBasedBuilder`
Returns:
split_generators: New split generator structure.
"""
if isinstance(split_generators, dict): # New structure
return split_generators
if isinstance(split_generators, list): # Legacy structure
if is_beam: # Legacy `tfds.core.BeamBasedBuilder`
beam = lazy_imports_lib.lazy_imports.apache_beam
generator_fn = beam.ptransform_fn(generator_fn)
return {
s.name: generator_fn(
**s.gen_kwargs) # Create the `beam.PTransform`
for s in split_generators
}
else:
return {
split_generator.name:
generator_fn(**split_generator.gen_kwargs)
for split_generator in split_generators
}
else:
raise TypeError(
f'Invalid `_split_generators` returned value: {split_generators}'
)
def _RawRecordBeamSourceInternal(self) -> beam.PTransform:
return (beam.ptransform_fn(lambda x: x)().with_input_types(
bytes).with_output_types(bytes))
def get_models(self):
"""PTransform that outputs a PCollection with all models in the stub."""
return beam.ptransform_fn(lambda pcoll: (
pcoll.pipeline
| beam.Create(m for m in self._models_by_id.values())))
def _CSVSource(self) -> beam.PTransform:
return (beam.ptransform_fn(lambda x: x)().with_input_types(
bytes).with_output_types(bytes))
