如何有效地从numpy数组中提取由其索引给出的元素列表? [英] How to extract a list of elements given by their indices from a numpy array efficiently?
问题描述
我有一个多维numpy数组,我想利用它的某些元素来构建一维数组.我需要采用的元素由它们的索引给出,例如:
I have a multidimensional numpy array and I would like to take some of its elements to construct a one dimensional array. The elements that I need to take are given by their indices, for example:
inds = [(0,0), (0,1), (1,1), (1,0), (0,2)]
我以一种简单的方式解决它:
I solve it in a straightforward way:
ls = []
for i, j in inds:
ls += [a[i,j]]
它给出期望的结果.但是,我已经意识到此解决方案对于我的目的而言太慢了.是否有可能以更有效的方式做同样的事情?
It gives the desired result. However, I have realized that this solution is too slow for my purposes. Are there a possibility to do the same in a more efficient way?
推荐答案
numpy数组可以用序列(通常是numpy数组)建立索引.
numpy arrays can be indexed with sequences (and, more generally, numpy arrays).
例如,这是我的数组a
In [19]: a
Out[19]:
array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14]])
i
和j
保存inds
数组的第一个和第二个坐标的序列:
i
and j
hold the sequences of the first and second coordinates of your inds
array:
In [20]: i
Out[20]: [0, 0, 1, 1, 0]
In [21]: j
Out[21]: [0, 1, 1, 0, 2]
您可以使用它们从a
中拉出相应的值:
You can use these to pull the corresponding values out of a
:
In [22]: a[i, j]
Out[22]: array([0, 1, 6, 5, 2])
如果代码中已经包含inds
,则可以使用zip
将元组列表分为i
和j
:
If you already have inds
in your code, you can separate the list of tuples into i
and j
using zip
:
In [23]: inds
Out[23]: [(0, 0), (0, 1), (1, 1), (1, 0), (0, 2)]
In [24]: i, j = zip(*inds)
In [25]: i
Out[25]: (0, 0, 1, 1, 0)
In [26]: j
Out[26]: (0, 1, 1, 0, 2)
或者,如果inds
是形状为(n,2)的数组,如下所示:
Or, if inds
is an array with shape (n, 2), like so:
In [27]: inds = np.array(inds)
In [28]: inds
Out[28]:
array([[0, 0],
[0, 1],
[1, 1],
[1, 0],
[0, 2]])
您可以简单地将inds
的转置分配给i, j
:
you can simply assign the transpose of inds
to i, j
:
In [33]: i, j = inds.T
In [34]: i
Out[34]: array([0, 0, 1, 1, 0])
In [35]: j
Out[35]: array([0, 1, 1, 0, 2])
In [36]: a[i, j]
Out[36]: array([0, 1, 6, 5, 2])
这篇关于如何有效地从numpy数组中提取由其索引给出的元素列表?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!