Can this for loop be vectorized for speed

Question

I need to find out the difference of every element of H_A with every element in H_B, here is my approach

              H_A=reshape(H_A,1,[]);
              H_B=reshape(H_B,1,[]);
              for i=1:SIZE^2
              D(i,:)=((H_A(i)-H_B)).^2;
              end

can this be vectorized for speed.

The dupe is for 3D points, this question is for 1D points. But the answers over there actually work for any number of dimensions, including 1. — Cris Luengo, Apr 26 '20 at 20:11

score 1 · Accepted Answer · answered Apr 26 '20 at 19:20

1

Try bsxfun.

H_A = reshape(H_A, [], 1);
H_B = reshape(H_B, 1, []);
D = bsxfun(@minus, H_A, H_B).^2;

answered Apr 26 '20 at 19:20

weiyin

6,819
4
47
58

score 1 · Answer 2 · answered Apr 26 '20 at 19:37

You could use broadcasting:

H_A = reshape(H_A, [], 1);
H_B = reshape(H_B, 1, []);
(H_A-H_B).^2;

Probably the fastest option. There is also pdist2 which allows you to calculate different distance metrics:

H_A = reshape(H_A, [], 1);
H_B = reshape(H_B, [],1);
pdist2(H_A,H_B,'squaredeuclidean')

Can this for loop be vectorized for speed

2 Answers2