使用range / arange函数作为参数索引/切片2d numpy数组

I have a basic doubt in numpy. I am using Python 2.7, numpy-1.9.2 on Ubuntu 14.04.

我对numpy有一个基本的怀疑。我在Ubuntu 14.04上使用Python 2.7,numpy-1.9.2。

For example, I initialize a 2d numpy array as a = np.zeros((10,10)).

例如,我将一个2d numpy数组初始化为a = np.zeros((10,10))。

I then try to index a portion of it using the range function as the indices by the following way:

然后我尝试使用范围函数作为索引来索引其中的一部分,方法如下:

a[range(0,5),range(0,5)]. I get an array of shape (5,). What I want is the first 5 rows and columns of the 2d array a.

一个[范围(0,5),范围(0,5)]。我得到一个形状数组(5,)。我想要的是2d数组的前5行和列a。

When I perform a[:5,:5], it seems to give me an array of shape (5,5).

当我执行[:5,:5]时,它似乎给了我一个形状的数组(5,5)。

Can someone explain to me why using the range function for specifying the index is failing? I am still confused about numpy indexing even after working with it for almost an year.

有人可以向我解释为什么使用范围函数来指定索引失败?即使在使用它近一年之后,我仍然对numpy索引感到困惑。

Thanks for your help in advance.

感谢您的帮助。

1 个解决方案

#1

With range you are using integer array indexing as described here:

对于范围,您使用整数数组索引,如下所述:

http://docs.scipy.org/doc/numpy/reference/arrays.indexing.html#integer-array-indexing

To get the equivalent of a[0:5,0:5], you have to take advantage of 'broadcasting'. Here the 1st index is a column vector

为了得到[0:5,0:5]的等价物,你必须利用'广播'。这里第一个索引是列向量

a[np.arange(0,5)[:,None],range(0,5)]

In [137]: np.arange(0,5)[:,None]
Out[137]: 
array([[0],
       [1],
       [2],
       [3],
       [4]])

I could go into more detail, but you could just as well read that doc.

我可以详细介绍一下,但您也可以阅读该文档。

np.ix_ is a utility that helps generate this sort of indexing arrays:

np.ix_是一个帮助生成这种索引数组的实用程序:

In [507]: np.ix_(range(0,5),range(0,5))
Out[507]: 
(array([[0],
        [1],
        [2],
        [3],
        [4]]), array([[0, 1, 2, 3, 4]]))

This (5,1) array broadcasts against a (1,5) array to produce a (5,5) indexing array.

这个(5,1)数组广播(1,5)数组以产生(5,5)索引数组。

MATLAB and numpy have choose alternative advanced indexing approaches:

MATLAB和numpy选择了替代的高级索引方法:

In MATLAB/Octave, a([1,2,3],[1,2,3]) indexes a (3,3) block. In numpy, a[[1,2,3],[1,2,3]] indexes the (3,) diagonal.

在MATLAB / Octave中,a([1,2,3],[1,2,3])索引(3,3)块。在numpy中,[[1,2,3],[1,2,3]]索引(3,)对角线。

a(sub2ind(size(a),[1,2,3],[1,2,3])) is the Octave diagonal; a[np.ix_([1,2,3],[1,2,3])] is the numpy block.

a(sub2ind(size(a),[1,2,3],[1,2,3]))是Octave对角线; a [np.ix _([1,2,3],[1,2,3])]是numpy块。

#1