numpy resize n-dimensional array with padding

Question

I have two arrays, a and b.

a has shape (1, 2, 3, 4)

b has shape (4, 3, 2, 1)

I would like to make them both (4, 3, 3, 4) with the new positions filled with 0's.

I can do:

new_shape = (4, 3, 3, 4)

a = np.resize(a, new_shape)
b = np.resize(b, new_shape)

..but this repeats the elements of each to form the new elements, which does not work for me.

Instead I thought I could do:

a = a.resize(new_shape)
b = b.resize(new_shape)

..which according to the documentation pads with 0's.

But it doesn't work for multi-dimensional arrays, raising error:

ValueError: resize only works on single-segment arrays

So is there a different way to achieve this? ie. same as np.resize but with 0-padding?

NB: I am only looking for pure-numpy solutions.

EDIT: I'm using numpy version 1.20.2

EDIT: I just found out that is works for numbers, but not for objects, I forgot to mention that it is an array of objects not numbers.

a.resize(4,3,3,4, refcheck=False)? See the different documentations numpy.ndarray.resize and numpy.resize — not_speshal
– not_speshal, Commented Jul 7, 2021 at 17:40
Definitely works for me (numpy v1.20.2). Check your numpy version. — not_speshal
– not_speshal, Commented Jul 7, 2021 at 17:57
@MustafaAydın Not on my pc, just upgraded, I have no idea why — user9413641
– user9413641, Commented Jul 7, 2021 at 18:02

hpaulj · Accepted Answer · 2021-07-07 18:26:58Z

resize method pads with 0s in a flattened sense; the function pads with repeats.

To illustrate how resize "flattens" before padding:

In [108]: a = np.arange(12).reshape(1,4,3)
In [109]: a
Out[109]: 
array([[[ 0,  1,  2],
        [ 3,  4,  5],
        [ 6,  7,  8],
        [ 9, 10, 11]]])
In [110]: a1 = a.copy()
In [111]: a1.resize((2,4,4))
In [112]: a1
Out[112]: 
array([[[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11],
        [ 0,  0,  0,  0]],

       [[ 0,  0,  0,  0],
        [ 0,  0,  0,  0],
        [ 0,  0,  0,  0],
        [ 0,  0,  0,  0]]])

If instead I make a target array of the right shape, and copy, I can maintain the original multidimensional block:

In [114]: res = np.zeros((2,4,4),a.dtype)
In [115]: res[:a.shape[0],:a.shape[1],:a.shape[2]]=a
In [116]: res
Out[116]: 
array([[[ 0,  1,  2,  0],
        [ 3,  4,  5,  0],
        [ 6,  7,  8,  0],
        [ 9, 10, 11,  0]],

       [[ 0,  0,  0,  0],
        [ 0,  0,  0,  0],
        [ 0,  0,  0,  0],
        [ 0,  0,  0,  0]]])

I wrote out the slices explicitly (for clarity). Such a tuple could be created programmatically if needed.

Collectives™ on Stack Overflow

numpy resize n-dimensional array with padding

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related