calculating all combinations using nchoosek. A faster way?

Dear all, I have a cell matrix A which is of dimension 13048 by 4 I need to calculate all the combinations using the following

 J2=nchoosek(1:size(refin_total_input2,1),2);

but it takes a lot of time

I tried to preallocate

 J2=zeros(85118628,2);
J2=nchoosek(1:size(refin_total_input2,1),2);

Is there a faster way of obtaining J2

I use matlaab 2009 64 bit

thanks

Answer 1

[Jan]

The pre-allocation is useless here, because the new values are not inserted, but the original array is overwritten.

Searching the FEX is always a good idea. There you find FEX: VChooseK.mex. I think, this is the fastest implementation, but it needs a compilation. Or you can download the pre-compiled files and install the required runtime libraries of MSVC 2008. An efficient and smart non-Mex version: FEX: combinator.

Beside the speed the C-MEX has the advantage, that it can create the combinations as UINT8/16/32, which occupy less memory than the 8 bytes per double.

[Sabbas]

Hi Jan. what do yuo mean " a compilation" I am not very familiar with these concepts. I downloaded the functions but when I try to run it I obtain the following error

 Error in ==> VChooseK at 1
function Y = VChooseK(X, K)

 ??? Output argument "Y" (and maybe others) not assigned during call to
"D:\Matlab \VChooseK.m>VChooseK".

thanks

[Jan]

The help text of the file contains the corresponding explanations: You have to install a compiler, run "mex -setup" and "mex -O VChooseK.c" afterwards.

I will include pre-compiled files on my home page as soon as possible (today if I find the time, but at least this year is guaranteed).