choosing K elements out of N elements

Hi,

Is there any function which acts as the following: given a set of n elements, it returns all the k-chosen combinations such that these k elements have at a distance of at most P? For example out of 1:100 we want to chose 5 elements such that they have a maximum distance of 10. how can we do it? btw I need a fast function

Thanks a lot

[John D'Errico]

Do you seriously expect that such a code already exists?

First, define what you mean by the word distance. I can think of at least 2 meanings.

As for being fast, everybody wants fast code. I have never actually seen someone say that they explicitly want slow code, or that they don't really care if the code is fast or incredibly slow.

[Sepehr Janghorbani]

:D I mean that if for example you have 3 numbers and you want their distance to be at most 10, they should be chosen such that any two of them are at most 10 numbers apart

Answer 1

[Jan]

Depending on the fact, if you want repetitions or not, you can use FEX: VCHooseKR or FEX: VChooseK.

If the maximum distance between the elements should be 10, you can solve the problem for the chunk 1:11 at first and in a second step add 0:89 to the result. bsxfun will be useful for the last step.

[Sepehr Janghorbani]

Thanks for your answer. This is actually what I was thinking of, but it requires yout function FEX: VChooseK to be called in a MATLAB loop which may make it much slower. I was wondering if someone has had the same problem before and maybe developed the whole code for it. Btw, I tried to use your code, but the compiling seemed really complicated. (I downloaded SDK 7.1 and installed it but it still didn't work.)

Thanks a gain for ur time

[Jan]

Yes, unfortunately compiling is not trivial.

If you explain, if you need repetions or not, I could perhaps offer a specific code. You have to call the code for the permutation once only:

B      = VChooseK(1:11, 5);
Result = bsxfun(@plus, reshape(B, [1, size(B)], (0:89).');
Result = reshape(Result, [], 5);

In this case Matlab's slower nchoosek would solve youtr problem also. Using uint8 values would accelerate the computations. If you cannot get the compilation to work, FEX: Combinator works, but slower.

[Sepehr Janghorbani]

Thanks, I just used Combinator in a loop as you mentioned. Btw don't you have a compiled version of Vchoosek?