I would convert the 2d array into 3d with previous rows by using NumPy or native functions.
Input:
[[1,2,3],
[4,5,6],
[7,8,9],
[10,11,12],
[13,14,15]]
Output:
[[[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]]]
Any one can help? I have searched online for a while, but cannot got the answer.
Approach #1
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))
Sample input, output -
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]]])
Approach #2
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]]])
How about a list comprehension?
In [144]: np.array([l[i:i + 3][::-1] for i in range(0, len(l) - 2)])
Out[144]:
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]]])