python人工智能tensorflow函数tf.assign使用方法

参数数量及其作用

该函数共有五个参数，分别是：

• 被赋值的变量 ref

• 要分配给变量的值 value、

• 是否验证形状 validate_shape

• 是否进行锁定保护 use_locking

• 名称 name

def assign(ref, value, validate_shape=None, use_locking=None, name=None)
Update 'ref' by assigning 'value' to it.
This operation outputs a Tensor that holds the new value of 'ref' after
the value has been assigned. This makes it easier to chain operations  
that need to use the reset value.  
Args:
 ref: A mutable `Tensor`.  
Should be from a `Variable` node. May be uninitialized.  
 value: A `Tensor`. Must have the same type as `ref`.  
   The value to be assigned to the variable.  
 validate_shape: An optional `bool`. Defaults to `True`.  
   If true, the operation will validate that the shape  
   of 'value' matches the shape of the Tensor being assigned to.  If false,  
   'ref' will take on the shape of 'value'.  
 use_locking: An optional `bool`. Defaults to `True`.  
   If True, the assignment will be protected by a lock;  
   otherwise the behavior is undefined, but may exhibit less contention.  
 name: A name for the operation (optional).  
Returns:
 A `Tensor` that will hold the new value of 'ref' after  
 the assignment has completed.    

该函数的作用是将一个要分配给变量的值value赋予被赋值的变量ref，用于tensorflow各个参数的变量赋值。

例子

该例子将举例如何进行变量之间的数据赋值和如何进行集合间的数据赋值。

import tensorflow as tf;  
import numpy as np;  
c1 = ['c1', tf.GraphKeys.GLOBAL_VARIABLES]
c2 = ['c2', tf.GraphKeys.GLOBAL_VARIABLES]
#常量初始化器
v1_cons = tf.get_variable('v1_cons',dtype = tf.float32,shape=[1,4], initializer=tf.constant_initializer(), collections = c1)
v2_cons = tf.get_variable('v2_cons',dtype = tf.float32,shape=[1,4], initializer=tf.constant_initializer(9), collections = c1)
#正太分布初始化器
v1_nor = tf.get_variable('v1_nor',dtype = tf.float32, shape=[1,4], initializer=tf.random_normal_initializer(mean=0, stddev=5), collections = c2)
v2_nor = tf.get_variable('v2_nor',dtype = tf.float32, shape=[1,4], initializer=tf.random_normal_initializer(mean=0, stddev=5), collections = c2)
assign1 = tf.assign(v1_cons,v2_cons)    #将v2_cons赋予v1_cons
c1_get = tf.get_collection('c1')        #获得c1集合
c2_get = tf.get_collection('c2')        #获得c2集合
assign2 = [tf.assign(cg1,cg2) for cg1,cg2 in zip(c1_get,c2_get) ]   #将c2赋予c1
with tf.Session() as sess:
   sess.run(tf.global_variables_initializer())
   print("v1_cons:",sess.run(v1_cons))
   print("v2_cons:",sess.run(v2_cons))
   print(sess.run(assign1))            #显示赋值后的结果
   print("将v2_cons赋予v1_cons:",sess.run(v1_cons))
   print("c1_get_collection:",sess.run(c1_get))
   print("c2_get_collection:",sess.run(c2_get))
   print(sess.run(assign2))            #显示赋值后的结果
   print("将c2赋予c1:",sess.run(c1_get))

其输出为：

v1_cons: [[0. 0. 0. 0.]]
v2_cons: [[9. 9. 9. 9.]]
[[9. 9. 9. 9.]]
将v2_cons赋予v1_cons: [[9. 9. 9. 9.]]
c1_get_collection: [array([[9., 9., 9., 9.]], dtype=float32), array([[9., 9., 9., 9.]], dtype=float32)]
c2_get_collection: [array([[-3.9746916, -7.5332146,  2.4480317, -1.3282107]], dtype=float32), array([[10.687443 ,  3.6653206,  1.7079141, -4.524155 ]], dtype=float32)]
[array([[-3.9746916, -7.5332146,  2.4480317, -1.3282107]], dtype=float32), array([[10.687443 ,  3.6653206,  1.7079141, -4.524155 ]], dtype=float32)]
将c2赋予c1: [array([[-3.9746916, -7.5332146,  2.4480317, -1.3282107]], dtype=float32), array([[10.687443 ,  3.6653206,  1.7079141, -4.524155 ]], dtype=float32)]

来源：https://blog.csdn.net/weixin_44791964/article/details/96145026