Discover MakerZone

MATLAB and Simulink resources for Arduino, LEGO, and Raspberry Pi

Learn more

Discover what MATLAB® can do for your career.

Opportunities for recent engineering grads.

Apply Today

To resolve issues starting MATLAB on Mac OS X 10.10 (Yosemite) visit:

Quickly applying a function to all possible combinations of vectors

Asked by Eric on 5 Jan 2013

I have a function f that takes two vectors x and y and

 f(x,y) = Transpose(sin(x-y))*Z*sin(x-y)

where Z is a diagonal positive definite matrix of appropriate size. Now, I have a matrix X where each column is a vector (so X1 = X(:,1), X2 = X(:,2), etc). I want to make a matrix K where K(1,1) = f(X1,X1), K(1,2) = f(X1,X2), etc.

I know i can do this with nested for loops but it runs very slow (X contains over 1000 points and K is calculated a few hundred times and it ends up taking about two hours). Now the code I'm basing it off of has similar functions and can do it in about 2 minutes. Anyone know how to quickly do this?


Eric on 6 Jan 2013

I want it to for X with any number of rows

Matt J on 6 Jan 2013

Yes, but will the number of rows typically be much greater, much less, or comparable to the number of columns? It would affect the approach used.

You can't perform the computation for any number of rows that you want. At some point, you're going to face limits of computer memory.

Eric on 6 Jan 2013

The number of rows will be significantly lower than the number of columns (i plan on using it for about 10 or so rows)



No products are associated with this question.

1 Answer

Answer by Matt J on 5 Jan 2013
Edited by Matt J on 5 Jan 2013
Accepted answer

Using only 1 loop,

    for i=1:size(X,1)
     K=K + sin(bsxfun(@minus,xy.',xy)).^2*Z(i,i);


Matt J on 6 Jan 2013

With what dimensions of X?

When I do it on my machine with size(X)=[2000, 2000], it completes in about 90 sec. If your data sizes are much bigger than that, you may have to face the fact that the computation time you're seeing is mainly dictated by data size and nothing more.

Matt J on 6 Jan 2013

The number of rows will be significantly lower than the number of columns (i plan on using it for about 10 or so rows)

When I run with size(X)=[10,2000], my proposed implementation finishes in less than 0.5 seconds. Even if you have a much poorer machine than me, there's no way it could still be taking 2 hours.

That's assuming of course that 2000 is a reasonable upper bound on the number of columns you're using. If it isn't, you might consider responding to my (third!!!) request for more full information on the size of X that you're benchmarking with.

Eric on 6 Jan 2013

Had to make some adjustments for my code and i made a mistake in translation, works great now, thanks!

Matt J

Contact us