A `Dataset` comprising records from one or more TFRecord files.
| 271 | |
| 272 | @tf_export("data.TFRecordDataset", v1=[]) |
| 273 | class TFRecordDatasetV2(dataset_ops.DatasetV2): |
| 274 | """A `Dataset` comprising records from one or more TFRecord files.""" |
| 275 | |
| 276 | def __init__(self, filenames, compression_type=None, buffer_size=None, |
| 277 | num_parallel_reads=None): |
| 278 | """Creates a `TFRecordDataset` to read one or more TFRecord files. |
| 279 | |
| 280 | Args: |
| 281 | filenames: A `tf.string` tensor or `tf.data.Dataset` containing one or |
| 282 | more filenames. |
| 283 | compression_type: (Optional.) A `tf.string` scalar evaluating to one of |
| 284 | `""` (no compression), `"ZLIB"`, or `"GZIP"`. |
| 285 | buffer_size: (Optional.) A `tf.int64` scalar representing the number of |
| 286 | bytes in the read buffer. If your input pipeline is I/O bottlenecked, |
| 287 | consider setting this parameter to a value 1-100 MBs. If `None`, a |
| 288 | sensible default for both local and remote file systems is used. |
| 289 | num_parallel_reads: (Optional.) A `tf.int64` scalar representing the |
| 290 | number of files to read in parallel. If greater than one, the records of |
| 291 | files read in parallel are outputted in an interleaved order. If your |
| 292 | input pipeline is I/O bottlenecked, consider setting this parameter to a |
| 293 | value greater than one to parallelize the I/O. If `None`, files will be |
| 294 | read sequentially. |
| 295 | |
| 296 | Raises: |
| 297 | TypeError: If any argument does not have the expected type. |
| 298 | ValueError: If any argument does not have the expected shape. |
| 299 | """ |
| 300 | filenames = _create_or_validate_filenames_dataset(filenames) |
| 301 | |
| 302 | self._filenames = filenames |
| 303 | self._compression_type = compression_type |
| 304 | self._buffer_size = buffer_size |
| 305 | self._num_parallel_reads = num_parallel_reads |
| 306 | |
| 307 | def creator_fn(filename): |
| 308 | return _TFRecordDataset(filename, compression_type, buffer_size) |
| 309 | |
| 310 | self._impl = _create_dataset_reader(creator_fn, filenames, |
| 311 | num_parallel_reads) |
| 312 | variant_tensor = self._impl._variant_tensor # pylint: disable=protected-access |
| 313 | super(TFRecordDatasetV2, self).__init__(variant_tensor) |
| 314 | |
| 315 | def _clone(self, |
| 316 | filenames=None, |
| 317 | compression_type=None, |
| 318 | buffer_size=None, |
| 319 | num_parallel_reads=None): |
| 320 | return TFRecordDatasetV2(filenames or self._filenames, |
| 321 | compression_type or self._compression_type, |
| 322 | buffer_size or self._buffer_size, |
| 323 | num_parallel_reads or self._num_parallel_reads) |
| 324 | |
| 325 | def _inputs(self): |
| 326 | return self._impl._inputs() # pylint: disable=protected-access |
| 327 | |
| 328 | @property |
| 329 | def element_spec(self): |
| 330 | return tensor_spec.TensorSpec([], dtypes.string) |