（反向）按列对二维 numpy 数组进行排序

Question

The following code sorts an 2d numpy array column-wise forth and back

import numpy as np
#Column-wise sort and inverse sort of image (2d array)
nrows = 10
ncols = 5
a = np.random.randint(nrows, size=(nrows, ncols))
a_sorted = np.sort(a, axis=0)

ori_indices = np.zeros_like(a)
for c in range(ncols):
    ori_indices[:,c] = np.argsort(np.argsort(a[:,c]))
#Do some work on sorted array, like e.g row-wise filtering
#After processing sorted array, move it back to original order
a_backsorted = np.zeros_like(a)
for c in range(ncols):
    a_backsorted[:,c] = a_sorted[:,c][ori_indices[:,c]]

print (a); print ()
print (a_backsorted); print ()
print (a_sorted); print ()

The code work as is but I guess there is a more efficient implementation without for loop (using fancy indexing)

Answer 1

You can try a_sorted[::-1] to reverse the array

print (a_sorted); print ()
print (a_sorted[::-1])

[[0 0 0 2 0]
 [2 0 0 2 2]
 [4 0 2 6 4]
 [4 2 3 7 5]
 [4 4 4 7 6]
 [5 5 4 8 7]
 [6 5 4 8 7]
 [7 6 8 9 8]
 [8 7 9 9 9]
 [8 8 9 9 9]]

[[8 8 9 9 9]
 [8 7 9 9 9]
 [7 6 8 9 8]
 [6 5 4 8 7]
 [5 5 4 8 7]
 [4 4 4 7 6]
 [4 2 3 7 5]
 [4 0 2 6 4]
 [2 0 0 2 2]
 [0 0 0 2 0]]

Answer 2

#Column-wise sort and inverse sort of image (2d array)
import numpy as np
#Define random array and sort it
nrows = 10
ncols = 5
a = np.random.randint(nrows, size=(nrows, ncols))
a_sorted = np.sort(a, axis=0)
#Save original order of columns
ori_indices = np.argsort(np.argsort(a, axis=0), axis=0)
#Do some work on sorted array, like e.g row-wise filtering.
#....
#After processing sorted array, move it back to original order:
c=np.array([[i] for i in range(ncols)]).T
a_backsorted = a_sorted[ori_indices, c]
#Check results
print (a); print ()
print (a_backsorted); print ()
print (a_sorted); print ()

Answer 3

import numpy as np
nrows = 10; ncols = 5
a = np.random.randint(nrows, size=(nrows, ncols))
a_sorted = np.sort(a, axis=0)
a_backsorted = np.zeros_like(a)
c = np.array([[i] for i in range(ncols)]).T
a_backsorted[np.argsort(a, axis=0), c] = a_sorted

The reverting of the column-wise sorting is done by inserting the values of the sorted array at the argsorted positions in the backsorted array. Since this is done columnwise, the argsorted positions are paired with the columns represented in the c array