Numpy: Reshape array along a specified axis

Question

I have the following array:

x = np.arange(24).reshape((2,3,2,2))
array([[[[ 0,  1],
     [ 2,  3]],

    [[ 4,  5],
     [ 6,  7]],

    [[ 8,  9],
     [10, 11]]],


   [[[12, 13],
     [14, 15]],

    [[16, 17],
     [18, 19]],

    [[20, 21],
     [22, 23]]]])

I would like to reshape it to a (3,4,2) array like below:

array([[[ 0,  1],
    [ 2,  3],
    [12, 13],
    [14, 15]],

   [[ 4,  5],
    [ 6,  7],
    [16, 17],
    [18, 19]],

   [[ 8,  9],
    [10, 11],
    [20, 21],
    [22, 23]]])

I've tried to use reshape but it gave me the following which is not what I want.

array([[[ 0,  1],
    [ 2,  3],
    [ 4,  5],
    [ 6,  7]],

   [[ 8,  9],
    [10, 11],
    [12, 13],
    [14, 15]],

   [[16, 17],
    [18, 19],
    [20, 21],
    [22, 23]]])

Can someone please help?

Divakar · Accepted Answer · 2016-08-13 05:50:46Z

8

Use transpose and then reshape like so -

shp = x.shape
out = x.transpose(1,0,2,3).reshape(shp[1],-1,shp[-1])

answered Aug 13, 2016 at 5:50

Divakar

222k19 gold badges273 silver badges374 bronze badges

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

Comments

Julien · Accepted Answer · 2016-08-13 05:36:11Z

1

x = np.arange(24).reshape((2,3,2,2))
y = np.dstack(zip(x))[0]
print y

result:

[[[ 0  1]
  [ 2  3]
  [12 13]
  [14 15]]

 [[ 4  5]
  [ 6  7]
  [16 17]
  [18 19]]

 [[ 8  9]
  [10 11]
  [20 21]
  [22 23]]]

answered Aug 13, 2016 at 5:36

Julien

15.3k6 gold badges33 silver badges58 bronze badges

3 Comments

Allen Qin Over a year ago

Thanks Julien. This works fine. However, I'm more after a native numpy solution for performance reasons.

Allen Qin Over a year ago

This function is going to be used to transform large numpy arrays many times. I'm not sure if zip is optimized for these kind of array operations.

Allen Qin Over a year ago

Nothing wrong with both yours and Divakar's solution. I just spent a bit time testing the performance of both solution. It turned out Divakar's solution is twice faster so I marked his solution as the answer. Thank you both.

Supamee · Accepted Answer · 2019-08-06 18:46:22Z

1

You can also use concatenate like so-

out=np.concatenate((x),axis=1)

I will note those since you mentioned this is for performance, this doesn't seem faster than Divakar suggestion:

shp = x.shape
out = x.transpose(1,0,2,3).reshape(shp[1],-1,shp[-1])

If anyone does a bench mark or finds something faster I would love to know.

answered Aug 6, 2019 at 18:46

Supamee

6355 silver badges16 bronze badges

Collectives™ on Stack Overflow

Numpy: Reshape array along a specified axis

3 Answers 3

Comments

3 Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

Comments

3 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related