Reshape an array in NumPy

Question

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...

eumiro · Accepted Answer · 2013-01-23 09:37:31Z

70

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]]])

answered Jan 23, 2013 at 9:37

eumiro

214k36 gold badges307 silver badges264 bronze badges

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

3 Comments

Jaime Over a year ago

Note that b now is not contiguous, which means it cannot be reshaped in place: b.reshape(9, 2) returns a copy, not a view of the same data, and b.shape = (9, 2) will raise and error.

deepelement Over a year ago

Very Very important comment by @Jaime, as the point of Shape is to allow optimistic resizing without a clone. Big deal with massive datasets

Jürgen K. Over a year ago

Why do you need to swap the axes?

Or_K · Accepted Answer · 2019-04-25 11:35:53Z

1

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)

answered Apr 25, 2019 at 11:35

Or_K

617 bronze badges

Comments

Alleo · Accepted Answer · 2019-07-01 00:13:46Z

0

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]]])

answered Jul 1, 2019 at 0:13

Alleo

8,6382 gold badges46 silver badges34 bronze badges

Collectives™ on Stack Overflow

Reshape an array in NumPy

3 Answers 3

3 Comments

Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

3 Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related