tf.compat.v1.placeholder_with_default  |  TensorFlow v2.16.1 (original) (raw)

tf.compat.v1.placeholder_with_default

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

A placeholder op that passes through input when its output is not fed.

tf.compat.v1.placeholder_with_default(
    input, shape, name=None
)

Migrate to TF2

This API is strongly discouraged for use with eager execution andtf.function. The primary use of this API is for testing computation wrapped within a tf.function where the input tensors might not have statically known fully-defined shapes. The same can be achieved by creating aconcrete functionfrom the tf.function with a tf.TensorSpec input which has partially defined shapes. For example, the code

@tf.function def f(): x = tf.compat.v1.placeholder_with_default( tf.constant([[1., 2., 3.], [4., 5., 6.]]), [None, 3]) y = tf.constant([[1.],[2.], [3.]]) z = tf.matmul(x, y) assert z.shape[0] == None assert z.shape[1] == 1

f()

can easily be replaced by

@tf.function def f(x): y = tf.constant([[1.],[2.], [3.]]) z = tf.matmul(x, y) assert z.shape[0] == None assert z.shape[1] == 1

g = f.get_concrete_function(tf.TensorSpec([None, 3]))

You can learn more about tf.function at Better performance with tf.function.

Description