tfds.features.TensorInfo

Stay organized with collections Save and categorize content based on your preferences.

Structure containing info on the tf.Tensor shape/dtype.

tfds.features.TensorInfo(
    shape: Shape,
    dtype: type_utils.TfdsDType,
    default_value=None,
    sequence_rank: Optional[int] = None,
    dataset_lvl: int = 0
)

shape tuple[int], shape of the tensor.
dtype Tensor DType that will be converted to NumPy DType.
default_value Used for retrocompatibility with previous files if a new field is added to provide a default value when reading the file.
sequence_rank int, Number of tfds.features.Sequence dimension.
dataset_lvl int, if >0, nesting level of a tfds.features.Dataset.

dataset_lvl

default_value

dtype Return the TensorFlow DType of this TensorInfo.
np_dtype

numpy_dtype

sequence_rank

shape

tf_dtype

Methods

copy_from

View source

@classmethod
copy_from(
    tensor_info: 'TensorInfo'
) -> 'TensorInfo'

Copy constructor.

from_tensor_spec

View source

@classmethod
from_tensor_spec(
    tensor_spec: tf.TensorSpec
) -> 'TensorInfo'

to_tensor_spec

View source

to_tensor_spec() -> tf.TensorSpec

Converts this TensorInfo instance to a tf.TensorSpec.

Note that there is a bug (b/227584124) around RaggedTensorSpec, so the output for sequences of sequences may not be correct.

Returns
The tf.TensorSpec corresponding to this instance.

__eq__

View source

__eq__(
    other
)

Equality.