I’m working with Tensorflow in Python. I notice that there is @tf_export(v1=[xxx]) decorator before many functions.

By reading source code of tf_export, i found it records the custom name to _NAME_TO_SYMBOL_MAPPING during call. It seems provide a ways to rename function.

My questions:

@tf_export(
    "sets.union", v1=["sets.union", "sets.set_union"])
@dispatch.add_dispatch_support
def set_union(a, b, validate_indices=True):

According to my understanding:

1. Python Decorator(tf_export) will is invoked when I really invoke tf.set_union. If i don’t invoke tf.set_union, tf_export doesn’t be invoked.

2. Because [1], why i can use tf.sets.union directly ?

Could somebody explain? thks !