Why the tf.name_scope with same name is different?

Question

Look at the code snippet：

import tensorflow as tf

with tf.name_scope('y'):
    a1 = tf.Variable(1,name='a')

with tf.name_scope('y'):
    a2 = tf.Variable(1,name='b')

print(a1.name)
print(a2.name)

The output is

y/a:0
y_1/b:0

Why the the name_scope of variable a2 is y_1?

Answer 1

On github there is an interesting discussion regarding this topic.

What you could do it append a '/' at the end, making it an absolute identifier:

import tensorflow as tf

with tf.name_scope('y'):
    a1 = tf.Variable(1,name='a')

with tf.name_scope('y/'):
    a2 = tf.Variable(1,name='b')

print(a1.name)
print(a2.name)

Which yields:

y/a:0
y/b:0

The same also applies to tf.variable_scope().

The bottomline to your question is likely that Tensorflow can't know whether you explicitely want to append something to the scope or whether somewhere else somebody created a different scope and wants to protect you against unintentional re-use. By appending a '/' at the end you turn the name into an absolute identifier.

Answer 2

The tf.name_scope returns a new context manager everytime when it's called with a string as the name parameter, it doesn't matter if you supply a name that has been seen before, in that case, the name of the scope will simply be made unique by calling unique_name(name).

If you want to re-entry the same name scope, you have to capture it somewhere, and use that scope as the parameter for name_scope.

Example taken from ops.py:

  # Creates a scope called "nested"
  with g.name_scope("nested") as scope:
    nested_c = tf.constant(10.0, name="c")
    assert nested_c.op.name == "nested/c"
    # Creates a nested scope called "inner".
    with g.name_scope("inner"):
      nested_inner_c = tf.constant(20.0, name="c")
      assert nested_inner_c.op.name == "nested/inner/c"
    # Create a nested scope called "inner_1".
    with g.name_scope("inner"):
      nested_inner_1_c = tf.constant(30.0, name="c")
      assert nested_inner_1_c.op.name == "nested/inner_1/c"
      # Treats `scope` as an absolute name scope, and
      # switches to the "nested/" scope.
      with g.name_scope(scope):
        nested_d = tf.constant(40.0, name="d")
        assert nested_d.op.name == "nested/d"
        with g.name_scope(""):
          e = tf.constant(50.0, name="e")
          assert e.op.name == "e"

And apparently you can put a slash ('/') at the end of the name to avoid it to be unique_named.