如何在有和没有重叠的固定大小块中拆分 numpy 数组? [英] How to split a numpy array in fixed size chunks with and without overlap?
问题描述
假设我有一个数组:
<预><代码>>>>arr = np.array(range(9)).reshape(3, 3)>>>阿尔数组([[0, 1, 2],[3, 4, 5],[6, 7, 8]])我想创建一个函数 f(arr, shape=(2, 2))
它接受数组和一个形状,并将数组分成给定形状的块 没有填充.因此,如有必要,可以重叠某些部分.例如:
我设法使用 np.lib.stride_tricks.as_strided(arr, shape=(2, 2, 2, 2), strides=(24, 8, 24, 8)) 创建到上面的输出代码>.但我不知道如何将其推广到所有数组和所有块大小.
最好用于 3D 阵列.
如果不需要重叠,则应避免重叠.另一个例子:
<预><代码>>>>arr = np.array(range(16).reshape(4,4)>>>阿尔数组([[ 0, 1, 2, 3],[ 4, 5, 6, 7],[ 8, 9, 10, 11],[12, 13, 14, 15]])>>>f(arr, shape=(2,2))数组([[[[[0, 1],[4, 5]],[[2, 3],[6, 7]]],[[[8, 9],[12, 13]],[[10, 11],[14, 15]]]])skimage.util.view_as_blocks
接近,但要求数组和块形状兼容.
scikit-image as view_as_windows
正是为了做到这一点 -
from skimage.util.shape import view_as_windowsview_as_windows(arr, (2,2))
样品运行 -
在[40]中:arr出[40]:数组([[0, 1, 2],[3, 4, 5],[6, 7, 8]])在 [41]: view_as_windows(arr, (2,2))出[41]:数组([[[[[0, 1],[3, 4]],[[1, 2],[4, 5]]],[[[3, 4],[6, 7]],[[4, 5],[7, 8]]]]])
<小时>
对于第二部分,使用来自同一家族/模块的表亲view_as_blocks
-
from skimage.util.shape import view_as_blocksview_as_blocks(arr, (2,2))
Lets say I have an array:
>>> arr = np.array(range(9)).reshape(3, 3)
>>> arr
array([[0, 1, 2],
[3, 4, 5],
[6, 7, 8]])
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]]]])
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.
Preferably, for 3D arrays.
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
comes close, but requires that the array and block shape are compatible.
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))
Sample run -
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]]]])
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))
这篇关于如何在有和没有重叠的固定大小块中拆分 numpy 数组?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!