tf.type_spec_from_value

Returns a tf.TypeSpec that represents the given value.

tf.type_spec_from_value(
    value
) -> tf.TypeSpec

Used in the notebooks

Used in the guide 

>>> tf.type_spec_from_value(tf.constant([1, 2, 3]))
TensorSpec(shape=(3,), dtype=tf.int32, name=None)
>>> tf.type_spec_from_value(np.array([4.0, 5.0], np.float64))
TensorSpec(shape=(2,), dtype=tf.float64, name=None)
>>> tf.type_spec_from_value(tf.ragged.constant([[1, 2], [3, 4, 5]]))
RaggedTensorSpec(TensorShape([2, None]), tf.int32, 1, tf.int64)

example_input = tf.ragged.constant([[1, 2], [3]])
@tf.function(input_signature=[tf.type_spec_from_value(example_input)])
def f(x):
  return tf.reduce_sum(x, axis=1)

value A value that can be accepted or returned by TensorFlow APIs. Accepted types for value include tf.Tensor, any value that can be converted to tf.Tensor using tf.convert_to_tensor, and any subclass of CompositeTensor (such as tf.RaggedTensor).

A TypeSpec that is compatible with value.

TypeError If a TypeSpec cannot be built for value, because its type is not supported.