Multidimensional array reshape with numpy

Question

I've been stuck on this numpy operation for a while now. I have an np.array of np.shape (x, y, z) that I want to make into an array of np.shape (y, x, z). I am having a hard time understanding the order on which np.reshape is done. For instance, I would like the values to be grouped by their index in the 2nd dim, and not on the first dimension.

For example, with this array of np.shape (3, 2, 9):

[[[-25 -25  15 -26 -26   0 -26 -26   3]
  [ -8  -2   0 -21  -9  39 -14 -11   2]]

 [[-25   2  18 -26 -10  10 -26 -15  14]
  [ -8  -2   0 -21  -9  39 -14 -11   2]]

 [[-25   2  18 -26 -10  14 -26 -15   5]
  [ -8  -2   3 -21  -9  18 -14 -11   3]]]

I would like the output of np.shape (2,3,9) to be:

[[[-25 -25  15 -26 -26   0 -26 -26   3]
  [-25   2  18 -26 -10  10 -26 -15  14]
  [-25   2  18 -26 -10  14 -26 -15   5]]

 [[ -8  -2   0 -21  -9  39 -14 -11   2]
  [ -8  -2   0 -21  -9  39 -14 -11   2]
  [ -8  -2   3 -21  -9  18 -14 -11   3]]]

And not :

[[[-25 -25  15 -26 -26   0 -26 -26   3]
  [ -8  -2   0 -21  -9  39 -14 -11   2]
  [-25   2  18 -26 -10  10 -26 -15  14]]

 [[ -8  -2   0 -21  -9  39 -14 -11   2]
  [-25   2  18 -26 -10  14 -26 -15   5]
  [ -8  -2   3 -21  -9  18 -14 -11   3]]]

I know this is basic numpy but so far every combination of order='F', array.T, etc got me no luck. Appreciate any help!

Could you provide an example with integers instead of floats? It's very difficult to quickly recognize which floats correspond to which others. — Linuxios
– Linuxios, Commented Jul 21, 2019 at 23:32
Separately, is your goal to exchange the first and second dimensions, rather than reshaping the array? — Linuxios
– Linuxios, Commented Jul 21, 2019 at 23:33
Sure I edited with integers. That's the goal yes. Is there another function than np.reshape to do this kind of operation ? — ldoe
– ldoe, Commented Jul 21, 2019 at 23:39

Akaisteph7 · Accepted Answer · 2019-07-21 23:39:37Z

2

I think what you are looking for is swapaxes:

np.swapaxes(x,0,1) #x is your array

answered Jul 21, 2019 at 23:39

Akaisteph7

6,9222 gold badges39 silver badges53 bronze badges

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

3 Comments

ldoe Over a year ago

Thanks a lot that's it.

Daniel F Over a year ago

For readablity, I'd almost even recommend np.einsum for cases like this, even though you're not doing any summing. np.einsum('ijk->jik', x)

Akaisteph7 Over a year ago

@DanielF I think it's much clearer with np.swapaxes what the goal is here. But it's great to have alternatives!

Collectives™ on Stack Overflow

Multidimensional array reshape with numpy

1 Answer 1

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related