deeplake.core.sample

class deeplake.core.sample.Sample
__init__(path: Optional[str] = None, array: Optional[ndarray] = None, buffer: Optional[Union[bytes, memoryview]] = None, compression: Optional[str] = None, verify: bool = False, shape: Optional[Tuple[int]] = None, dtype: Optional[str] = None, creds: Optional[Dict] = None, storage: Optional[StorageProvider] = None, timeout: Optional[float] = None)

Represents a single sample for a tensor. Provides all important meta information in one place.

Note

If self.is_lazy is True, this Sample doesn’t actually have any data loaded. To read this data, simply try to read it into a numpy array (sample.array)

Parameters
  • path (str) – Path to a sample stored on the local file system that represents a single sample. If path is provided, array should not be. Implicitly makes self.is_lazy == True.

  • array (np.ndarray) – Array that represents a single sample. If array is provided, path should not be. Implicitly makes self.is_lazy == False.

  • buffer – (bytes): Byte buffer that represents a single sample. If compressed, compression argument should be provided.

  • compression (str) – Specify in case of byte buffer.

  • verify (bool) – If a path is provided, verifies the sample if True.

  • shape (Tuple[int]) – Shape of the sample.

  • dtype (optional, str) – Data type of the sample.

  • creds (optional, Dict) – Credentials for s3, gcp and http urls.

  • storage (optional, StorageProvider) – Storage provider.

  • timeout (optional, float) – Timeout in seconds for reading the file. Applicable only for http(s) urls.

property array: ndarray

Return numpy array corresponding to the sample. Decompresses the sample if necessary.

Example

>>> sample = deeplake.read("./images/dog.jpg")
>>> arr = sample.array
>>> arr.shape
(323, 480, 3)
compressed_bytes(compression: Optional[str]) bytes

Returns this sample as compressed bytes.

Note

If this sample is pointing to a path and the requested compression is the same as it’s stored in, the data is returned without re-compressing.

Parameters

compression (Optional[str]) – self.array will be compressed into this format. If compression is None, return uncompressed_bytes().

Returns

Bytes for the compressed sample. Contains all metadata required to decompress within these bytes.

Return type

bytes

Raises

ValueError – On recompression of unsupported formats.

property pil: Image

Return PIL image corresponding to the sample. Decompresses the sample if necessary.

Example

>>> sample = deeplake.read("./images/dog.jpg")
>>> pil = sample.pil
>>> pil.size
(480, 323)
uncompressed_bytes() Optional[bytes]

Returns uncompressed bytes.