tff.types.tensorflow_to_type

View source on GitHub

Returns a tff.Type for an obj containing TensorFlow type specs.

tff.types.tensorflow_to_type(
    obj: object
) -> tff.types.Type

Used in the notebooks

This function extends tff.types.to_type to handle TensorFlow type specs and Python structures containing TensorFlow type specs:

• tf.dtypes.DType
• tensor-like objects (e.g. (tf.int32, tf.TensorShape([2, 3])))
• tf.TensorSpec
• tf.data.DatasetSpec

For example:

tensorflow_to_type(tf.int32)
tff.TensorType(np.int32)

tensorflow_to_type((tf.int32, tf.TensorShape([2, 3])))
tff.TensorType(np.int32, (2, 3))

spec = tf.TensorSpec(shape=[2, 3], dtype=tf.int32)
tensorflow_to_type(spec)
tff.TensorType(np.int32, (2, 3))

spec = tf.data.DatasetSpec(tf.TensorSpec([2, 3], dtype=tf.int32))
tensorflow_to_type(spec)
tff.SequenceType(tff.TensorType(np.int32, (2, 3)))