Reshape an array in NumPy

Consider an array of the following form (just an example):

[[ 0  1]
 [ 2  3]
 [ 4  5]
 [ 6  7]
 [ 8  9]
 [10 11]
 [12 13]
 [14 15]
 [16 17]]

It’s shape is [9,2]. Now I want to transform the array so that each column becomes a shape [3,3], like this:

[[ 0  6 12]
 [ 2  8 14]
 [ 4 10 16]]
[[ 1  7 13]
 [ 3  9 15]
 [ 5 11 17]]

The most obvious (and surely “non-pythonic”) solution is to initialise an array of zeroes with the proper dimension and run two for-loops where it will be filled with data. I’m interested in a solution that is language-conform…

Answers:

Thank you for visiting the Q&A section on Magenaut. Please note that all the answers may not help you solve the issue immediately. So please treat them as advisements. If you found the post helpful (or not), leave a comment & I’ll get back to you as soon as possible.

Method 1

a = np.arange(18).reshape(9,2)
b = a.reshape(3,3,2).swapaxes(0,2)

# a: 
array([[ 0,  1],
       [ 2,  3],
       [ 4,  5],
       [ 6,  7],
       [ 8,  9],
       [10, 11],
       [12, 13],
       [14, 15],
       [16, 17]])


# b:
array([[[ 0,  6, 12],
        [ 2,  8, 14],
        [ 4, 10, 16]],

       [[ 1,  7, 13],
        [ 3,  9, 15],
        [ 5, 11, 17]]])

Method 2

numpy has a great tool for this task (“numpy.reshape”) link to reshape documentation

a = [[ 0  1]
 [ 2  3]
 [ 4  5]
 [ 6  7]
 [ 8  9]
 [10 11]
 [12 13]
 [14 15]
 [16 17]]

`numpy.reshape(a,(3,3))`

you can also use the “-1” trick

`a = a.reshape(-1,3)`

the “-1” is a wild card that will let the numpy algorithm decide on the number to input when the second dimension is 3

so yes.. this would also work:
a = a.reshape(3,-1)

and this:
a = a.reshape(-1,2)
would do nothing

and this:
a = a.reshape(-1,9)
would change the shape to (2,9)

Method 3

There are two possible result rearrangements (following example by @eumiro). Einops package provides a powerful notation to describe such operations non-ambigously

>> a = np.arange(18).reshape(9,2)

# this version corresponds to eumiro's answer
>> einops.rearrange(a, '(x y) z -> z y x', x=3)

array([[[ 0,  6, 12],
        [ 2,  8, 14],
        [ 4, 10, 16]],

       [[ 1,  7, 13],
        [ 3,  9, 15],
        [ 5, 11, 17]]])

# this has the same shape, but order of elements is different (note that each paer was trasnposed)
>> einops.rearrange(a, '(x y) z -> z x y', x=3)

array([[[ 0,  2,  4],
        [ 6,  8, 10],
        [12, 14, 16]],

       [[ 1,  3,  5],
        [ 7,  9, 11],
        [13, 15, 17]]])

All methods was sourced from stackoverflow.com or stackexchange.com, is licensed under cc by-sa 2.5, cc by-sa 3.0 and cc by-sa 4.0