tensorflow scope理解

import tensorflow as tf
def scoping(fn, scope1, scope2, vals):
    with fn(scope1):
        a = tf.Variable(vals[0], name='a')
        b = tf.get_variable('b', initializer=vals[1])
        c = tf.constant(vals[2], name='c')
        with fn(scope2):
            d = tf.add(a * b, c, name='res')

        print '\n  '.join([scope1, a.name, b.name, c.name, d.name]), '\n'
    return d

d1 = scoping(tf.variable_scope, 'scope_vars', 'res', [1, 2, 3])
d2 = scoping(tf.name_scope,     'scope_name', 'res', [1, 2, 3])

with tf.Session() as sess:
    writer = tf.summary.FileWriter('logs', sess.graph)
    sess.run(tf.global_variables_initializer())
    print sess.run([d1, d2])
    writer.close()

从上面的可以看出来，tf.variable_scope()为所有变量（不管你是怎么创建的）、操作（ops）、常量（constant）添加一个前缀，而tf.name_scope()会忽视使用tf.get_variable()创建的变量，因为它假设你知道你使用的变量位于哪个scope中。