Help speeding up matrix algebra in loop...

Hi everyone, I'm simply looking for some suggestions for speeding up the following block of code:

    bhat  = (X'*X)\(X'*y);                  
    yhat  = X*bhat;                         
    resid = y - yhat;                       
    bclu=zeros(size(X,2));                  
    for i=idx.unique
        idmat=find(idx.sample==i);
        x_g=X(idmat,:);
        u_g=resid(idmat,:);
        bclu_i=x_g'*u_g*(u_g)'*x_g;
        bclu = bclu+bclu_i;
    end
    vhat=((X'*X)\bclu)/(X'*X);

I suppose there might be some scope for the use of "sparse" here, or perhaps vectorization, but it's not obvious to me how to optimize this. Any suggestions would be great. Thanks!

Answer 1

[Walter Roberson]

You could probably use parfor, if you have the Parallel Computing Toolkit.

[John]

Oops! Didn't mean to accept so quickly!

This code is part of an optimization routine which itself is run in parallel, so I can't easily use parallel here. Any other suggestions?

[John]

Did I miss my chance??