如何将numpy数组拆分为固定大小的块(有无重叠)? [英] How to split a numpy array in fixed size chunks with and without overlap?
问题描述
让我说我有一个数组:
>>> arr = np.array(range(9)).reshape(3, 3)
>>> arr
array([[0, 1, 2],
[3, 4, 5],
[6, 7, 8]])
我想创建一个函数f(arr, shape=(2, 2))
,该函数采用数组和形状,并将数组拆分为给定形状的块,而无需填充.因此,如有必要,通过重叠某些部分.例如:
I would like to create a function f(arr, shape=(2, 2))
that takes the array and a shape, and splits the array into chunks of the given shape without padding. Thus, by overlapping certain parts if necessary. For example:
>>> f(arr, shape=(2, 2))
array([[[[0, 1],
[3, 4]],
[[1, 2],
[4, 5]]],
[[[3, 4],
[6, 7]],
[[4, 5],
[7, 8]]]])
我设法用np.lib.stride_tricks.as_strided(arr, shape=(2, 2, 2, 2), strides=(24, 8, 24, 8))
在上面创建了输出.但是我不知道如何将其推广到所有数组和所有块大小.
I managed to creates to output above with np.lib.stride_tricks.as_strided(arr, shape=(2, 2, 2, 2), strides=(24, 8, 24, 8))
. But I don't know how to generalize this for to all arrays and all chunk sizes.
最好用于3D阵列.
如果没有必要重叠,则应避免重叠.另一个示例:
If no overlap is necessary, it should avoid that. Another example:
>>> arr = np.array(range(16).reshape(4,4)
>>> arr
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]])
>>> f(arr, shape=(2,2))
array([[[[0, 1],
[4, 5]],
[[2, 3],
[6, 7]]],
[[[8, 9],
[12, 13]],
[[10, 11],
[14, 15]]]])
skimage.util.view_as_blocks
接近,但要求数组和块的形状兼容.
skimage.util.view_as_blocks
comes close, but requires that the array and block shape are compatible.
推荐答案
There's a builtin in scikit-image as view_as_windows
for doing exactly that -
from skimage.util.shape import view_as_windows
view_as_windows(arr, (2,2))
样品运行-
In [40]: arr
Out[40]:
array([[0, 1, 2],
[3, 4, 5],
[6, 7, 8]])
In [41]: view_as_windows(arr, (2,2))
Out[41]:
array([[[[0, 1],
[3, 4]],
[[1, 2],
[4, 5]]],
[[[3, 4],
[6, 7]],
[[4, 5],
[7, 8]]]])
对于第二部分,使用其来自同一个家族/模块的表亲 view_as_blocks
-
For the second part, use its cousin from the same family/module view_as_blocks
-
from skimage.util.shape import view_as_blocks
view_as_blocks(arr, (2,2))
这篇关于如何将numpy数组拆分为固定大小的块(有无重叠)?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!