How to replace for loop with vectorization?

Is there any way I can replace this for loop with a vectorized approach?

v = zeros(4, 3);
ii = [1 ; 2 ; 3 ; 1 ; 3 ; 4];
res = [-1 -1 1 ; -1 -1 1 ; -1 -1 1 ; 1 1 -1 ; 1 1 -1 ; -1 1 -1];

for i = 1:3
     v(:,i) = accumarray(ii , res(:,i));
end

I know if res was just scalar values, I could use this:

v= accumarray(ii , res);

But is it posible to remove the loop knowing that res has more than one columns?

[Walter Roberson]

You can repmat(ii, size(res,2),1) and use res(:)

[Sai Bhargav Avula]

repmat may not be the right way as the accumarray is used here.

Answer 1

[Sai Bhargav Avula]

Hi,

You can eliminate the for loop by using the code below.

ii = [1 ; 2 ; 3 ; 1 ; 3 ; 4];
res = [-1 -1 1 ; -1 -1 1 ; -1 -1 1 ; 1 1 -1 ; 1 1 -1 ; -1 1 -1];
[x, y] = ndgrid(ii,1:size(res,2));
v=accumarray([x(:) y(:)],res(:));

But I would recommend to use the for loop if the size of the matrix is small.

In the above case

The elapsed time in the case of for loop is 0.000248 sec. Where as in the case of where for loop is eliminated the elapsed time is 0.000750 sec.

Hope this helps!

Answer 2

[Bruno Luong]

[iii,jj]=ndgrid(ii,1:size(res,2));
v=accumarray([iii(:) jj(:)], res(:))