Any danger of using reuse_variables () in version 1.0 +?

Question

In the documentation on upgrading to TF 1.0 here , the following ominous sound appears:

Constructs like tf.get_variable_scope().reuse_variables(), most likely, will not work. We recommend that you delete these lines and replace them with lines such as:

with tf.variable_scope(tf.get_variable_scope(), reuse=True):

I would prefer to stick with it .reuse_variables(), since my code is currently set up, it is quite difficult to use the proposed idiom. So far, my code works fine with v1.0, so I'm not sure if I need to worry about this or not. The documentation does not indicate whether there are plans for deferment or what exactly is the problem with use .reuse_variables().

Answer 1

tf.variable_scope , . TensorFlow 1.0 script.