如何冻结张量流中变量的某些部分 [英] How to freeze some part of the variable in tensorflow

问题描述

我注意到可以通过设置 trainable=False 来冻结变量.我想知道如何只冻结其中的一部分而不是将其拆分为两个变量.

I've noticed that a variable can be frozen by setting trainable=False. I'm wondering how to freeze only part of it instead of splitting it into two variables.

例如我只想冻结v = tf.Variable(tf.zeros([2])中的第一个元素.看起来grad_new = tf.multiply(grad, np.array([0,1])) 不起作用.

For example, I only want to freeze the first element in v = tf.Variable(tf.zeros([2]). It seems that grad_new = tf.multiply(grad, np.array([0,1])) doesn't work.

推荐答案

问题已解决.

grad_new = tf.multiply(grad, np.array([0,1])) 实际上效果很好.是把代码弄乱的保护程序.

grad_new = tf.multiply(grad, np.array([0,1])) actually works quite well. It's the saver who mess up the code.

这篇关于如何冻结张量流中变量的某些部分的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！