在ND4J中使用广播 [英] Using broadcast in ND4J

问题描述

我最近从numpy切换到ND4J，但是很难理解ND4J中的广播如何工作.

I recently switched from numpy to ND4J but had a hard time understanding how broadcasting in ND4J works.

说我有两个ndarray，a的形状为[3,2,4,5]，b的形状为[2,4,5].我想逐个元素地将它们加起来并向每个a[i] for i = 0 to 2广播b.在numpy中，可以简单地通过a + b完成，而在ND4J中，a.add(b)会引发异常.我尝试了a.add(b.broadcast(3))，但还是没有运气.

Say I have two ndarray, a of shape [3,2,4,5] and b of shape [2,4,5]. I would like to element-wise add them up and broadcast b to each a[i] for i = 0 to 2. In numpy it can simply be done via a + b, while in ND4J a.add(b) throws an exception. I tried a.add(b.broadcast(3)) but still no luck.

在ND4J中执行此操作的正确方法是什么?

What is the correct way of doing this in ND4J?

推荐答案

到目前为止，我发现的唯一方法如下

The only method I have found so far is as following

var a = Nd4j.createUninitialized(Array(3,2,4,4))
var b = Nd4j.createUninitialized(Array(2,4,4))
b = b.reshape(1,32)
b = b.broadcast(3,32)
b = b.reshape(3, 2, 4, 4)
a.add(b)

请告诉我是否有更好的方法

Please kindly let me know if there is a better way to do this

这篇关于在ND4J中使用广播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！