跨步将2D数组转换为3D数组


问题内容

我可以使用NumPy或本机函数将2d数组转换为具有前一行的3d。

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]]]

有人可以帮忙吗?我已经在网上搜索了一段时间,但找不到答案。


问题答案:

方法1

一种方法np.lib.stride_tricks.as_strided使我们可以view进入输入2D数组,因此不再占用存储空间-

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

另一种方法是,broadcasting在生成所有行索引时更容易一些-

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]]])