scope 命名方法

tf.name_scope()

Tensorflow 中有两种途径生成变量 variable, tf.get_variable(), tf.Variable().

import tensorflow as tf

with tf.name_scope("a_name_scope"):
    initializer = tf.constant_initializer(value=1)
    var1 = tf.get_variable(name='var1', shape=[1], dtype=tf.float32, initializer=initializer)
    var2 = tf.Variable(name='var2', initial_value=[2], dtype=tf.float32)
    var21 = tf.Variable(name='var2', initial_value=[2.1], dtype=tf.float32)
    var22 = tf.Variable(name='var2', initial_value=[2.2], dtype=tf.float32)


with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(var1.name)        # var1:0
    print(sess.run(var1))   # [ 1.]
    print(var2.name)        # a_name_scope/var2:0
    print(sess.run(var2))   # [ 2.]
    print(var21.name)       # a_name_scope/var2_1:0
    print(sess.run(var21))  # [ 2.0999999]
    print(var22.name)       # a_name_scope/var2_2:0
    print(sess.run(var22))  # [ 2.20000005]

tf.Variable() 定义的时候, 为了不重复变量名, Tensorflow 输出的变量名并不是一样的.

var2, var21, var22 并不是一样的变量.

tf.get_variable()定义的变量不会被tf.name_scope()当中的名字所影响.

tf.variable_scope()

达到重复利用变量的效果, 要使用 tf.variable_scope(), 并搭配 tf.get_variable() 这种方式产生和提取变量.

tf.get_variable() 如果遇到了同样名字的变量时, 它会单纯的提取这个同样名字的变量(避免产生新变量). 而在重复使用的时候, 一定要

在代码中强调 scope.reuse_variables(), 否则系统将会报错, 以为你只是单纯的不小心重复使用到了一个变量.

with tf.variable_scope("a_variable_scope") as scope:
    initializer = tf.constant_initializer(value=3)
    var3 = tf.get_variable(name='var3', shape=[1], dtype=tf.float32, initializer=initializer)
    scope.reuse_variables()
    var3_reuse = tf.get_variable(name='var3',)
    var4 = tf.Variable(name='var4', initial_value=[4], dtype=tf.float32)
    var4_reuse = tf.Variable(name='var4', initial_value=[4], dtype=tf.float32)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(var3.name)            # a_variable_scope/var3:0
    print(sess.run(var3))       # [ 3.]
    print(var3_reuse.name)      # a_variable_scope/var3:0
    print(sess.run(var3_reuse)) # [ 3.]
    print(var4.name)            # a_variable_scope/var4:0
    print(sess.run(var4))       # [ 4.]
    print(var4_reuse.name)      # a_variable_scope/var4_1:0
    print(sess.run(var4_reuse)) # [ 4.]