numpy.swapaxes如何工作? [英] How does numpy.swapaxes work?

问题描述

我创建了一个示例数组:

I created a sample array:

a = np.arange(18).reshape(9,2)

在打印时，我将其作为输出:

On printing, I get this as output:

[[ 0  1]
[ 2  3]
[ 4  5]
[ 6  7]
[ 8  9]
[10 11]
[12 13]
[14 15]
[16 17]]

在执行此重塑操作时:

b = a.reshape(2,3,3).swapaxes(0,2)

我得到:

[[[ 0  9]
[ 3 12]
[ 6 15]]

[[ 1 10]
[ 4 13]
[ 7 16]]

[[ 2 11]
[ 5 14]
[ 8 17]]]

我遇到了这个问题，但是并不能解决我的问题.

I went through this question, but it does not solve my problem.

在NumPy中重塑数组

该文档也没有用.

https://docs.scipy.org/doc/numpy/reference/generation/numpy.swapaxes.html

我需要知道交换的工作方式(x轴，y轴，z轴).进行图解说明将是最有帮助的.

I need to know how the swapping is working(which is x-axis, y-axis, z-axis). A diagrammatic explanation would be most helpful.

推荐答案

从重塑开始

In [322]: a = np.arange(18).reshape(2,3,3)
In [323]: a
Out[323]: 
array([[[ 0,  1,  2],
        [ 3,  4,  5],
        [ 6,  7,  8]],

       [[ 9, 10, 11],
        [12, 13, 14],
        [15, 16, 17]]])

这将显示为2个平面，每个平面为3x3.那部分清楚吗?数组在某一点处成形为(9,2)的事实并不重要.重塑不会改变元素的顺序.

This displays as 2 planes, and each plane is a 3x3. Is that part clear? The fact that the array was shaped (9,2) at one point isn't significant. Reshaping doesn't change the order of elements.

应用swapaxes.形状现在是(3,3,2). 3个平面，每个平面为3x2.这种特殊的交换与转置相同

Apply the swapaxes. Shape is now (3,3,2). 3 planes, each is 3x2. This particular swap is the same as a transpose

np.arange(18).reshape(2,3,3).transpose(2,1,0)

中轴不变.仍然有[0,3,6]，[9,12,15]等列.

The middle axis is unchanged. There are still columns of [0,3,6], [9,12,15], etc.

使用3种不同尺寸的轴可视化更改可能更容易

It may be easier to visualize the change with 3 different sized axes

In [335]: a=np.arange(2*3*4).reshape(2,3,4)
In [336]: a
Out[336]: 
array([[[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11]],

       [[12, 13, 14, 15],
        [16, 17, 18, 19],
        [20, 21, 22, 23]]])
In [337]: a.swapaxes(0,2)
Out[337]: 
array([[[ 0, 12],
        [ 4, 16],
        [ 8, 20]],

       [[ 1, 13],
        [ 5, 17],
        [ 9, 21]],

       [[ 2, 14],
        [ 6, 18],
        [10, 22]],

       [[ 3, 15],
        [ 7, 19],
        [11, 23]]])

请注意当我展平数组时会发生什么情况

Notice what happens when I flatten the array

In [338]: a.swapaxes(0,2).ravel()
Out[338]: 
array([ 0, 12,  4, 16,  8, 20,  1, 13,  5, 17,  9, 21,  2, 14,  6, 18, 10,
       22,  3, 15,  7, 19, 11, 23])

条款的顺序已经改组.创建时为[0,1,2,3 ...].现在1是第六项(2x3).

the order of terms has been shuffled. As created it was [0,1,2,3...]. Now the 1 is the 6th term (2x3).

在封面下numpy实际上是通过更改shape，strides和order来执行交换或转置的，而不更改数据缓冲区(即它是视图).但是，进一步的重塑(包括穿行)迫使其进行复制.但这在现阶段可能比帮助更令人困惑.

Under the covers numpy actually performs the swap or transpose by changing shape, strides and order, without changing the data buffer (i.e. it's a view). But further reshaping, including raveling, forces it to make a copy. But that might be more confusing than helpful at this stage.

在numpy中已编号.诸如x，y，z或平面，行，列之类的术语可以帮助您将其映射到可以可视化的结构上，但它们不是内置"的.用文字描述交换或转置非常棘手.

In numpy axes are numbered. Terms like x,y,z or planes, rows, columns may help you map those on to constructs that you can visualize, but they aren't 'built-in'. Describing the swap or transpose in words is tricky.

这篇关于numpy.swapaxes如何工作?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！