NumPy array permute columns 3D matrix with argsort

Question

I need to permute elements of columns in the matrix A (3D matrix by axis 0) by 2D permutation matrix pi obtained from argsort, that contains new indices for all columns.

By application permutation matrix pi on the matrix A (A[pi]) I will get a 4D matrix with new shape. For example, the shape of A is (2,3,4) and the shape of A[pi] is (2,3,3,4).

I am able to extract the required sorted matrix from A[pi] using the command:

swapaxes (diagonal(A[pi], axis1=2, axis2=1),1,2)

But it seems to be too complicated and slow.

Is there another elegant solution?

Example:

print(A)
[[[   73   701  2411  2414]
  [ 5515  8292  8414 16135]
  [  100  1241  2146  2931]]

 [[ 1335  1747  3418  6312]
  [ 3788  5449  5753  9738]
  [  565  3038  3800  5430]]]

pi=argsort(Norm_order(A),0)

print(pi)
[[1, 0, 1],
 [0, 1, 0]]

print(swapaxes(diagonal(A[pi],axis1=2,axis2=1),1,2))
[[[ 1335  1747  3418  6312]
  [ 5515  8292  8414 16135]
  [  565  3038  3800  5430]]

 [[   73   701  2411  2414]
  [ 3788  5449  5753  9738]
  [  100  1241  2146  2931]]]

Thank's for help Willem :). I am new here.

Roman Hajtmanek
– Roman Hajtmanek

2017-03-27 01:56:51 +00:00
Commented Mar 27, 2017 at 1:56 — Roman Hajtmanek
– Roman Hajtmanek, Commented Mar 27, 2017 at 1:56

Paul Panzer · Accepted Answer · 2017-03-27 00:51:36Z

0

Maybe a matter of taste, but I find the following a bit more readable:

i, j = np.ogrid[:3, :4]
A[pi[..., None], i, j]

Output:

array([[[ 1335,  1747,  3418,  6312],
        [ 5515,  8292,  8414, 16135],
        [  565,  3038,  3800,  5430]],

       [[   73,   701,  2411,  2414],
        [ 3788,  5449,  5753,  9738],
        [  100,  1241,  2146,  2931]]])

answered Mar 27, 2017 at 0:51

Paul Panzer

53.3k3 gold badges60 silver badges103 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

NumPy array permute columns 3D matrix with argsort

1 Answer 1

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related