Python - Rearrange elements in a 3D array

Question

I have the following Numpy array of shape (4, 4, 3):

a = [[[ 0  1  2]
      [ 3  4  5]
      [ 6  7  8]
      [ 9 10 11]]

      [[12 13 14]
       [15 16 17]
       [18 19 20]
       [21 22 23]]

      [[24 25 26]
       [27 28 29]
       [30 31 32]
       [33 34 35]]

      [[36 37 38]
       [39 40 41]
       [42 43 44]
       [45 46 47]]]

I am looking for an elegant solution to re-arrange the elements in that array to get the following 3D array of shape (3, 4, 4):

a_new = [[[ 0  3  6  9]
          [12 15 18 21]
          [24 27 30 33]
          [36 39 42 45]]

         [[ 1  4  7 10]
          [13 16 19 22]
          [25 28 31 34]
          [37 40 43 46]]

         [[ 2  5  8 11]
          [14 17 20 23]
          [26 29 32 35]
          [38 41 44 47]]]

So basically you want to put all elements from all lists into one and then split them into three step-wise? — Hugo G
– Hugo G, Commented Jan 13, 2017 at 14:59

Divakar · Accepted Answer · 2017-01-13 14:59:13Z

4

Use np.transpose -

a.transpose(2,0,1)

Or use np.rollaxis -

np.rollaxis(a,2,0) # Or np.rollaxis(a,-1,0)

answered Jan 13, 2017 at 14:59

Divakar

222k19 gold badges273 silver badges374 bronze badges

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

Comments

Hugo G · Accepted Answer · 2017-01-13 16:10:01Z

2

In case somebody asks the same question for pure Python:

mylist = [[[1,2,3], [4,5,6]], [[7,8,9], [10, 11, 12]]]
flat = sum(sum(mylist, []), [])
groups = 3
print [flat[r::groups] for r in range(groups)]

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

answered Jan 13, 2017 at 16:10

Hugo G

16.7k9 gold badges62 silver badges82 bronze badges

Comments

alexblae · Accepted Answer · 2017-01-13 15:07:06Z

1

The fastest way I can think of is to use numpy's swapaxes function in combination with the transpose function.

anew=np.swapaxes(a,0,1).T

answered Jan 13, 2017 at 15:07

alexblae

7461 gold badge5 silver badges13 bronze badges

5 Comments

Ehsan Khaveh Over a year ago

This works too but np.transpose seems to be much faster

alexblae Over a year ago

Interesting. You would think that array.T is just calling np.transpose on self, but apparently not. I need to test this some other time.

Ehsan Khaveh Over a year ago

Okay. I had 0.00405ms for array.T and 0.00095ms for np.transpose. Tested it on a bigger array of shape (256, 256, 3)

Divakar Over a year ago

@Ehsan With np.swapaxes(a,0,1).T we are making two passes of permuting axes, once with the swapaxes and then with the transpose (.T), whereas a.transpose(2,0,1) is just one pass of permuting axes. Hence, the performance difference.

Ehsan Khaveh Over a year ago

@Divakar that makes sense. I totally missed that. thanks :)

Collectives™ on Stack Overflow

Python - Rearrange elements in a 3D array

3 Answers 3

Comments

Comments

5 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

Comments

Comments

5 Comments

Your Answer

Sign up or log in

Post as a guest

Related