广播更像是:

In [280]: y[np.array([0,2,4])[...,None], np.array([1,2])]
Out[280]: 
array([[ 1,  2],
       [15, 16],
       [29, 30]])

我在[0,2,4]中添加了一个尺寸，使其成为2d. broadcast_arrays可用于查看广播的数组的外观:

I added a dimension to [0,2,4] making it 2d. broadcast_arrays can be used to see what the broadcasted arrays look like:

In [281]: np.broadcast_arrays(np.array([0,2,4])[...,None], np.array([1,2]))
Out[281]: 
[array([[0, 0],
        [2, 2],
        [4, 4]]), 
 array([[1, 2],
        [1, 2],
        [1, 2]])]

np.broadcast_arrays([[0],[2],[4]], [1,2])与没有array包装器的情况相同. np.meshgrid([0,2,4], [1,2], indexing='ij')是产生这些索引数组的另一种方法.

np.broadcast_arrays([[0],[2],[4]], [1,2]) is the samething without the array wrappers. np.meshgrid([0,2,4], [1,2], indexing='ij') is another way of producing these indexing arrays.

(由meshgrid或broadcast_arrays产生的列表可以用作y[_]的参数.)

(the lists produced by meshgrid or broadcast_arrays could be used as the argument for y[_].)

所以说[1,2]与索引数组一起广播是正确的，但是它省略了有关调整尺寸的内容.

So it's right to say [1,2] is broadcast with the index array, but it omits the bit about adjusting dimensions.

早些时候，他们有这个例子:

A little earlier they have this example:

y[np.array([0,2,4])]

等效于y[np.array([0,2,4]), :].它选择3行，并从中选择所有项目. 1:3情况可以看作是对此的扩展，选择3行，然后2列.

which is equivalent to y[np.array([0,2,4]), :]. It picks 3 rows, and all items from them. The 1:3 case can be thought of as an extension of this, picking 3 rows, and then 2 columns.

y[[0,2,4],:][:,1:3]

如果广播过于混乱，这可能是思考索引的一种更好的方法.

This might be a better way of thinking about the indexing if broadcasting is too confusing.

还有另一个文档页面可能会更好地处理此问题

There's another docs page that might handle this better

http://docs.scipy.org/doc/numpy/reference/arrays.indexing.html

在此文档中，基本索引涉及切片和整数

In this docs, basic indexing involves slices and integers

y[:,1:3], y[1,:], y[1, 1:3]

高级索引涉及一个数组(或列表)

Advanced indexing involves an array (or list)

y[[0,2,4],:]

这产生与y[::2,:]相同的结果，除了列表大小写会产生一个副本(切片(基本)视图).

This produces the same result as y[::2,:], except the list case produces a copy, the slice (basic) a view.

y[[0,2,4], [1,2,3]]是纯高级索引数组索引的情况，结果是3个项目，分别位于(0,1)，(2,2)和(4,3).

y[[0,2,4], [1,2,3]] is a case of pure advance index array indexing, the result is 3 items, ones at (0,1), (2,2), and (4,3).

y[[0,2,4], 1:3]是本文档调用Combining advanced and basic indexing的一种情况，从'[0,2,4]'开始'高级'，从'1:3'开始基本.

y[[0,2,4], 1:3] is a case that this docs calls Combining advanced and basic indexing, 'advanced' from `[0,2,4]', basic from '1:3'.

http://docs.scipy.org/doc/numpy/reference/arrays.indexing.html#combining-advanced-and-basic-indexing

查看更复杂的索引数组可能会增加一些见识.

Looking at a more complex index array might add some insight.

In [222]: i=[[0,2],[1,4]]

与另一个列表一起使用，它是纯"高级的，并且广播结果:

Used with another list, it is 'pure' advanced, and the result is broadcasted:

In [224]: y[i, [1,2]]
Out[224]: 
array([[ 1, 16],
       [ 8, 30]])

索引数组是:

In [234]: np.broadcast_arrays(i, [1,2])
Out[234]: 
[array([[0, 2],
        [1, 4]]), 
 array([[1, 2],
        [1, 2]])]

[1,2]列表刚刚扩展为一个(2,2)数组.

The [1,2] list is just expanded to a (2,2) array.

将它与切片一起使用是这种混合高级/基本示例，结果为3d (2,2,2).

Using it with a slice is an example of this mixed advanced/basic, and the result is 3d (2,2,2).

In [223]: y[i, 1:3]
Out[223]: 
array([[[ 1,  2],
        [15, 16]],

       [[ 8,  9],
        [29, 30]]])

与广播等效的是

y[np.array(i)[...,None], [1,2]]