class Foo(object):
def __init__(self):
self.shape = [None, 4]
@tf.function(input_signature=[tf.TensorSpec(shape=self.shape)])
def add_constant(self, a):
print('Building graph')
return a + 5
Any way to make this possible? Currently, I just decorate add_constant
inside Foo class __init__
as:
self.add_constant = tf.function(func=self.add_constant, input_signature=[tf.TensorSpec(shape=self.shape)]).
However, this makes the code less readable comparable to function decorator.