34

I would like to generate all the possible combinations of the elements of a given number of vectors.

For example, for [1 2], [1 2] and [4 5] I want to generate the elements:

[1 1 4; 1 1 5; 1 2 4; 1 2 5; 2 1 4; 2 1 5; 2 2 4; 2 2 5]

The problem is that I don't know the number of vectors for which I need to calculate the combinations. There might be 3 as in this case, or there may be 10, and I need a generalization. Can you please help me to this in MATLAB? Is there already a predefined function that can do this task?

chappjc
  • 30,359
  • 6
  • 75
  • 132
Simon
  • 4,999
  • 21
  • 69
  • 97

4 Answers4

52

Consider this solution using the NDGRID function:

sets = {[1 2], [1 2], [4 5]};
[x y z] = ndgrid(sets{:});
cartProd = [x(:) y(:) z(:)];

cartProd =
     1     1     4
     2     1     4
     1     2     4
     2     2     4
     1     1     5
     2     1     5
     1     2     5
     2     2     5

Or if you want a general solution for any number of sets (without having to create the variables manually), use this function definition:

function result = cartesianProduct(sets)
    c = cell(1, numel(sets));
    [c{:}] = ndgrid( sets{:} );
    result = cell2mat( cellfun(@(v)v(:), c, 'UniformOutput',false) );
end

Note that if you prefer, you can sort the results:

cartProd = sortrows(cartProd, 1:numel(sets));

Also, the code above does not check if the sets have no duplicate values (ex: {[1 1] [1 2] [4 5]}). Add this one line if you want to:

sets = cellfun(@unique, sets, 'UniformOutput',false);
Amro
  • 123,847
  • 25
  • 243
  • 454
17

Try ALLCOMB function at FileExchange.

If you store you vectors in a cell array, you can run it like this:

a = {[1 2], [1 2], [4 5]};
allcomb(a{:})
ans =

     1     1     4
     1     1     5
     1     2     4
     1     2     5
     2     1     4
     2     1     5
     2     2     4
     2     2     5
yuk
  • 19,098
  • 13
  • 68
  • 99
  • 6
    Note that `ALLCOMB` uses `NDGRID` in essentially the same way as in Amro's answer, with error-proofing on top. – lodhb May 21 '13 at 22:42
13

This late answers provides two additional solutions, where the second is the solution (in my opinion) and an improvement on Amro's answer solution with ndgrid by applying MATLAB's powerful comma-separated lists instead of cell arrays for high performance,

  1. If you have the Neural Network Toolbox: use combvec
  2. If you do not have the toolbox, as is usually the case: below is another way to generalize the Cartesian product for any number of sets.

Just as Amro did in his answer, the comma-separated lists syntax (v{:}) supplies both the inputs and outputs of ndgrid. The difference (fourth line) is that it avoids cellfun and cell2mat by applying comma-separated lists, again, now as the inputs to cat:

N = numel(a);
v = cell(N,1);
[v{:}] = ndgrid(a{:});
res = reshape(cat(N+1,v{:}),[],N);

The use of cat and reshape cuts execution time almost in half. This approach was demonstrated in my answer to an different question, and more formally by Luis Mendo.

Community
  • 1
  • 1
chappjc
  • 30,359
  • 6
  • 75
  • 132
  • THIS is the way to go in my opinion. Noticed that it can easily be extended to get all the permutation of length `m` for a vector of length `n` with: `[v{1:m}] = ndgrid(1:n);` and `res = reshape(cat(m+1,v{:}),[],m)` – obchardon Aug 27 '19 at 15:14
0

we can also use the 'combvec' instruction in matlab

    no_inp=3 % number of inputs we want...in this case we have 3 inputs                  
    a=[1 2 3]
    b=[1 2 3]
    c=[1 2 3]

    pre_final=combvec(c,b,a)';
    final=zeros(size(pre_final));

    for i=1:no_inp
    final(:,i)=pre_final(:,no_inp-i+1);
    end
    final

Hope it helps. Good luck.

Akhil Appu Shetty
  • 1