Why the same code works for the matrix of size 5 but does not work for the matrix of size 100?

Hi,

I have a sparse matrix L of zeros and ones and I want to change the non-zero entries to some fractions. This is my code:

    beta = 0.5
    n = size(L,1); %set n = dim(L)
    rsv=ones(1,n)*L; %row sum vector of L 
    H =L;
    for i=1:n
       for j=1:n
          if H(i,j) == 1
            H(i,j)=1/((1-beta)*rsv(i)+ beta*rsv(j));
          end
        end
    end

I found this code works for the matrix of size 5 but does not work for the matrix of size 100 (all entries stay unchanged).

Do you know why and how to change the code?

Many thanks,

Weijian

[John Doe]

What is it that doesn't work? It looks good to me? Do you get an error?

[Matt J]

rsv=ones(1,n)*L;

It should be more efficient to implement this as

 rsv=sum(L,1);

[Matt J]

It should also be more efficient to avoid the for-loops as follows

 [i,j,s]=find(L);

 H=sparse(i,j,1./((1-beta)*rsv(i)+ beta*rsv(j)));

[Weijian]

Thanks for your suggestion. Your code solves the problem :)

[Matt J]

It is not really clear why this should have made a difference. It is equivalent to your original code, if L is type double. I can only suppose that L was logical and that my other Answer below accounts for the difference.

Answer 1

[Matt J]

If L has a single 1 on all columns then rsv(i)=1 for all i. If that happens then

1/((1-beta)*rsv(i)+ beta*rsv(j))

will equal 1 for all i and j.

[Weijian]

No, this is not the case, most of the columns have more than 1 non-zeros. But never mind, your code solves my problem. Thanks a lot.

Answer 2

[Matt J]

Another possibility. If L is type logical, then creating H as H=L will make H type logical as well. This means that all assignments

 H(i,j) = something_nonzero

will convert "something_nonzero" to logical 1.