如何将一个numpy数组分割成固定大小的块，是否有重叠?

Lets say I have an array:

假设我有一个数组:

>>> arr = np.array(range(9)).reshape(3, 3)
>>> arr
array([[0, 1, 2],
       [3, 4, 5],
       [6, 7, 8]])

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:

我想要创建一个函数f(arr, shape=(2,2))，它接受数组和一个形状，并将数组分割成给定形状的块，不填充。因此，如果必要的话，通过重叠某些部分。例如:

>>> 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.

我使用np.lib.stride_tricks创建到上面的输出。as_strided(arr, shape=(2,2,2,2, 2)， stride =(24,8,24, 8))但我不知道如何将它推广到所有数组和所有块大小。

Preferably, for 3D arrays.

最好是,3 d数组。

If no overlap is necessary, it should avoid that. Another example:

如果不需要重叠，就应该避免重叠。另一个例子:

>>> 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.

skimage.util。view_as_blocks非常接近，但要求数组和块形状是兼容的。

1 个解决方案

#1

There's a builtin in scikit-image as view_as_windows for doing exactly that -

在scikit-image中有一个作为view_as_windows的内建框，用于实现这一点

from skimage.util.shape import view_as_windows

view_as_windows(arr, (2,2))

Sample run -

样本运行-

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]]]])

For the second part, use its cousin from the same family/module view_as_blocks -

对于第二部分，使用来自同一家族/模块view_as_blocks -的表亲

from skimage.util.shape import view_as_blocks

view_as_blocks(arr, (2,2))

#1