Tensorflow共享变量

使用Variable声明变量，同名变量的name后会自动加_1，可以赋初始值，但是需要在session初始化后才会生效。

import tensorflow as tf var1 = tf.Variable(1.0,name='firstVar') print("var1:",var1.name) var1 = tf.Variable(2.0,name='firstVar') print("var1:",var1.name) var2 = tf.Variable(3.0) print("var2:",var2.name) var2 = tf.Variable(4.0) print("var2:",var2.name) with tf.Session() as sess: sess.run(tf.global_variables_initializer()) print("var1=",var1.eval()) print("var2=",var2.eval())

运行结果：

var1: firstVar:0 var1: firstVar_1:0 var2: Variable:0 var2: Variable_1:0 var1= 2.0 var2= 4.0

使用get_variable()也可以创建变量，但是不能生成两个name相同的变量（在同一scope中），

getVar1 = tf.get_variable("firstVar",[1],initializer=tf.constant_initializer(0.3)) print("getVar:",getVar1.name) getVar1 = tf.get_variable("firstVar1",[1],initializer=tf.constant_initializer(0.4)) print("getVar:",getVar1.name)

声明不同scope的变量：

with tf.variable_scope("t1"): getVar1 = tf.get_variable("firstVar",[1],initializer=tf.constant_initializer(0.3)) print("getVar:",getVar1.name) with tf.variable_scope("t2"): getVar1 = tf.get_variable("firstVar",[1],initializer=tf.constant_initializer(0.4)) print("getVar:",getVar1.name) getVar: t1/firstVar:0 getVar: t2/firstVar:0

使用作用域的reuse参数实现共享变量的功能，因为加了reuse所以这里获取的变量必须是已经创建过的，否则会报错。

with tf.variable_scope("t1",reuse=True): get1=getVar1 = tf.get_variable("firstVar", [1], initializer=tf.constant_initializer(0.3)) print(get1) # t1/firstVar:0 auto_reuse在第一调用这个作用域是传入false，第二次为true；也就是说如果变量没有被创建过，则会帮你创建，创建过的话就不创建。 with tf.variable_scope("t1",reuse=tf.AUTO_REUSE): get1 = tf.get_variable("firstVar", [1], initializer=tf.constant_initializer(0.3)) print(get1.name)

tf.name_scope（）可以限制操作符的作用域，空字符串可以把作用域扩大到顶层。

with tf.variable_scope("scope",reuse=tf.AUTO_REUSE) as sp: get1 = tf.get_variable("firstVar", [1], initializer=tf.constant_initializer(0.3)) print(get1.name) with tf.name_scope("bar"): v = tf.get_variable("v",[1]) x = 10+v with tf.name_scope(""): y = 10+v with tf.variable_scope("scope2"): var2 = tf.get_variable("V",[1]) with tf.variable_scope(sp) as sp1: var3 = tf.get_variable("V3",[1]) with tf.variable_scope(""): var4 = tf.get_variable("v4",[1]) print("var4:",var4.name) print("y:",y.op.name)