Based on this efficient concept for generating a matrix of binary combinations:
you can efficiently create a matrix of the required combinations, by incrementally adding to a matrix and removing the superfluous rows at each loop iteration, thus avoiding any out-of-memory issues (as would happen if you tried to generate all combinations at once).
Generate matrix of all combinations:
n = numel(x);
m = [0;1];
for k = 2:n*n
r = size(m,1);
m = [zeros(r,1),m;ones(r,1),m];
m(sum(m,2)>s,:) = [];
end
m(sum(m,2)~=s,:) = [];
For s=3 I get 2300 rows (i.e. each row is a unique combination), and for s=5 I get 53130 rows. You can easily check that each row sums to s (i.e. 3 or 5):
>> all(sum(m,2)==3)
ans = 1
and that the rows are unique:
>> size(unique(m,'rows'),1)==size(m,1)
ans = 1
Then you can simply loop over those rows, reshape each row into a matrix, and do whatever operations you want:
r = size(m,1);
B = cell(r,1);
for k = 1:r
A = reshape(m(k,:),n,n);
B{k} = A*x;
end
The first few outputs are:
>> B{1:3}
ans =
0
0
5
5
5
ans =
0
5
0
5
5
ans =
0
5
5
0
5
Do this once for s=3, and once for s=5.
