更新时间:2022-11-09 23:32:21
您可以使用数组的更高维度视图,并沿额外维度取平均值:
You can use a higher dimensional view of your array and take the average along the extra dimensions:
In [12]: a = np.arange(36).reshape(6, 6)
In [13]: a
Out[13]:
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],
[24, 25, 26, 27, 28, 29],
[30, 31, 32, 33, 34, 35]])
In [14]: a_view = a.reshape(3, 2, 3, 2)
In [15]: a_view.mean(axis=3).mean(axis=1)
Out[15]:
array([[ 3.5, 5.5, 7.5],
[ 15.5, 17.5, 19.5],
[ 27.5, 29.5, 31.5]])
通常,如果要为(rows, cols)
的数组分配形状为(a, b)
的容器,则将其重塑为.reshape(rows // a, a, cols // b, b)
.还请注意.mean
的顺序很重要,例如a_view.mean(axis=1).mean(axis=3)
会引发错误,因为a_view.mean(axis=1)
仅具有三个维度,尽管a_view.mean(axis=1).mean(axis=2)
可以正常工作,但是却很难理解发生了什么.
In general, if you want bins of shape (a, b)
for an array of (rows, cols)
, your reshaping of it should be .reshape(rows // a, a, cols // b, b)
. Note also that the order of the .mean
is important, e.g. a_view.mean(axis=1).mean(axis=3)
will raise an error, because a_view.mean(axis=1)
only has three dimensions, although a_view.mean(axis=1).mean(axis=2)
will work fine, but it makes it harder to understand what is going on.
照原样,上面的代码仅在您可以在数组中容纳整数个bin的情况下才有效,即a
除以rows
并且b
除以cols
.有很多方法可以处理其他情况,但是您将必须定义自己想要的行为.
As is, the above code only works if you can fit an integer number of bins inside your array, i.e. if a
divides rows
and b
divides cols
. There are ways to deal with other cases, but you will have to define the behavior you want then.