跨步将2D数组转换为3D数组 [英] Transform 2D array to a 3D array with overlapping strides

问题描述

我可以通过使用NumPy或本机函数将2d数组转换为具有先前行的3d.

I would convert the 2d array into 3d with previous rows by using NumPy or native functions.

输入:

[[1,2,3],
 [4,5,6],
 [7,8,9],
 [10,11,12],
 [13,14,15]]

输出:

[[[7,8,9],    [4,5,6],    [1,2,3]],
 [[10,11,12], [7,8,9],    [4,5,6]],
 [[13,14,15], [10,11,12], [7,8,9]]]

任何人都可以帮忙吗? 我已经在网上搜索了一段时间，但找不到答案.

Any one can help? I have searched online for a while, but cannot got the answer.

推荐答案

方法1

使用 np.lib.stride_tricks.as_strided 的一种方法view放入输入2D数组中，因此不再占用存储空间-

One approach with np.lib.stride_tricks.as_strided that gives us a view into the input 2D array and as such doesn't occupy anymore of the memory space -

L = 3  # window length for sliding along the first axis
s0,s1 = a.strides

shp = a.shape
out_shp = shp[0] - L + 1, L, shp[1]
strided = np.lib.stride_tricks.as_strided
out = strided(a[L-1:], shape=out_shp, strides=(s0,-s0,s1))

样本输入，输出-

In [43]: a
Out[43]: 
array([[ 1,  2,  3],
       [ 4,  5,  6],
       [ 7,  8,  9],
       [10, 11, 12],
       [13, 14, 15]])

In [44]: out
Out[44]: 
array([[[ 7,  8,  9],
        [ 4,  5,  6],
        [ 1,  2,  3]],

       [[10, 11, 12],
        [ 7,  8,  9],
        [ 4,  5,  6]],

       [[13, 14, 15],
        [10, 11, 12],
        [ 7,  8,  9]]])

---

方法2

或者，使用 broadcasting 在生成所有行索引时-

Alternatively, a bit easier one with broadcasting upon generating all of row indices -

In [56]: a[range(L-1,-1,-1) + np.arange(shp[0]-L+1)[:,None]]
Out[56]: 
array([[[ 7,  8,  9],
        [ 4,  5,  6],
        [ 1,  2,  3]],

       [[10, 11, 12],
        [ 7,  8,  9],
        [ 4,  5,  6]],

       [[13, 14, 15],
        [10, 11, 12],
        [ 7,  8,  9]]])

这篇关于跨步将2D数组转换为3D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！