Vectorization of multiple embedded for loops

I have the following code that includes 3 iterated for loops in order to create an upper diagonal matrix, I plan on performing on large data set many times and want to make as computationally efficient as possible.

data = magic(3);
n = size(data,1);
W = zeros(n,n);
for i = 1:n
  for j = i:n
    if i==j
      www(i,j)=0;
    else
      for k = 1:n
        temp(1,k) = (data(i,k)-data(j,k))^2;
        sumTemp = sumTemp + temp(1,k);
      end
      W(i,j)=sqrt(sumTemp);
    end
    temp = 0;
    sumTemp = 0;        
  end
end

Answer should look like:

[0 6.4807 9.7980
0 0 6.4807
0 0 0]

I am working it hard right now, but figure I would throw it out there in case anyone has any suggestions that would save me hours of fiddling around. Thanks in advance.

Keith

Answer 1

[Andrei Bobrov]

W = triu(squeeze(sqrt(sum(bsxfun(@minus,data,permute(data,[3 2 1])).^2,2))));

[Keith]

Andrei - quite impressive. I would like to get to the point where I can vectorize like this. Can you point me to any good learning references. Thanks for quick and accurate response!

[Andrei Bobrov]

Hi Keith! Full vectoring is not always the best solution. About efficient training: I think, this site and documentation for MATLAB.