reduce time taken to execute nested for loop

Hi,

I am trying to subtract two matrix. and my program is as follows :
for i=1:length(B)
for j=1:length(A)
 Y(j,:)=B(i,:)-A(j,:); %(takes one row of B and subtracts it with each row of A)
end
end

Now the inner for loop takes about 0.37 seconds. But my matrices are huge A[7000,100] & B[4000,1000]. Hence the above nested for loop takes about half hour+ to execute completely. Can anyone please suggest a much faster way to do the same. (hopefully in a minute or two)

[krs.rb1]

well i tried ,

[i1, j1] = meshgrid(1:size(B,1),1:size(A,1));

Y = minus(B(i1,:),A(j1,:));

but it gave an error :

??? Out of memory. Type HELP MEMORY for your

options.

did i do something wrong?

and,I had made a mistake, it's -> A[7000,1000] & B[4000,1000]

Answer 1

[Jan]

1. Pre-allocate the output! 2. Relying on LENGTH to reply the wanted dimension is prone to errors. Use SIZE(X, dim) instead.

lenA = size(A, 1);
lenB = size(B, 1);
Y = zeros(lenA, lenB);
for i = 1:lenB
  for j = 1:lenA
    Y(j,:) = B(i,:) - A(j,:);
  end
end

Now the values of Y are overwritten completely in every iteration of the for i loop. Most likely this is not intented. If this problem is fixed, you can try if BSXFUN or manual expansion by ONES is faster instead of the inner loop:

Y = bsxfun(@minus, B(i, :), A);

% Or:
Bi = B(i, :);
Y = Bi(ones(1, lenA), :) - A;