为什么tf.variable_scope具有default_name参数？

Question

The first two arguments of tf.variable_scope 's __init__ method are

• name_or_scope : string or VariableScope : the scope to open.
• default_name : The default name to use if the name_or_scope argument is None , this name will be uniquified. If name_or_scope is provided it won't be used and therefore it is not required and can be None .

If I understand correctly, this argument is equivalent to (and therefore could be easily replaced with)

if name_or_scope is None:
  name_or_scope = default_name
with tf.variable_scope(name_or_scope, ...):
  ...

Now, I am not sure I understand why it was deemed necessary to have this special treatment for the scope name — after all, many parameters could use a parameterizable default argument.

So what is the rationale behind the introduction of this argument?

Answer 1

You are right. It is just a convenience.

Take the case of TensorFlow models defined here . If you take a specific look at InceptionV4.py , you will see that it has a scope argument in its definition. Just below you will see that InceptionV4 has been passed as a default scope. Therefore it was entirely not required to even has a scope argument in the definition. But it makes sense, if somebody gives scope=None .

Think about it. Model definitions can get very comples very quickly. Therefore, a default_scope argument, helps in reinforcing the wisdom of the model definition writer to introduce some sort of deliberate structure in the model definition, even if the end user is very naive about it.