Tensorflow中v.assign(v + 1)和v = v + 1之间的差异

时间:2018-07-04 05:12:31

标签: tensorflow assign arithmetic-expressions

以下Tensorflow代码可以正常工作,并且v1变为[1.,1.,1。]

v1 = tf.get_variable('v1', shape=[3], initializer=tf.zeros_initializer)
v1 = v1 + 1 

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print (v1.eval())

以下代码段也为我们提供了与上述完全相同的结果。如果我们运行v1,则sess.run(inc_v1)会变成[1.,1.,1。]。

v1 = tf.get_variable('v1', shape=[3], initializer=tf.zeros_initializer)
inc_v1 = v1.assign(v1 + 1)


with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    sess.run(inc_v1)
    print (v1.eval())

但是,以下代码会导致错误。

v1 = tf.get_variable('v1', shape=[3], initializer=tf.zeros_initializer)
v1 = v1 + 1 
inc_v1 = v1.assign(v1 + 1)


with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    sess.run(inc_v1)
    print (v1.eval())

错误如下:

AttributeError: 'Tensor' object has no attribute 'assign'

能否请您告诉我为什么会导致错误?

1 个答案:

答案 0 :(得分:3)

张量和变量是TensorFlow中的不同对象

import tensorflow as tf


def inspect(t):
    print('\n %s\n-------' % t.name)
    print(type(t))
    print(t.op.outputs)
    print('has assign method' if 'assign' in dir(t) else 'has no assign method')


v1 = tf.get_variable('v1', shape=[3], initializer=tf.zeros_initializer)
inspect(v1)
v2 = v1 + 1
inspect(v2)

给予

 v1:0
-------
<class 'tensorflow.python.ops.variables.Variable'>
[<tf.Tensor 'v1:0' shape=(3,) dtype=float32_ref>]
has assign method

 add:0
-------
<class 'tensorflow.python.framework.ops.Tensor'>
[<tf.Tensor 'add:0' shape=(3,) dtype=float32>]
has no assign method

因此,v1:0实际上是变量本身,v1具有方法assign。这是有道理的,因为它只是对浮点值的引用。 另一方面,v2 = v1 + 1导致add操作的输出。因此,v2不再是变量,您无法为v2分配新值。您希望在这种情况下更新add的哪个操作数? 每当您使用v1时,请考虑使用read_value()中的v1操作:

v1 = tf.get_variable('v1', shape=[3], initializer=tf.zeros_initializer)
inspect(v1)
w = v1.read_value()
inspect(w)
v2 = v1.read_value() + 1
inspect(v2)

给予

 v1:0
-------
<class 'tensorflow.python.ops.variables.Variable'>
[<tf.Tensor 'v1:0' shape=(3,) dtype=float32_ref>]
has assign method

 read:0
-------
<class 'tensorflow.python.framework.ops.Tensor'>
[<tf.Tensor 'read:0' shape=(3,) dtype=float32>]
has no assign method

 add:0
-------
<class 'tensorflow.python.framework.ops.Tensor'>
[<tf.Tensor 'add:0' shape=(3,) dtype=float32>]
has no assign method