Re: Sparse Matrix non zero elements in each row

[Joao F. D. Rodrigues]

On 06-08-15 14:45, Shamika Mohanan wrote:
> I'm creating a sparse matrix using Octave in c++ code. The code is-
>
> octave_value_list out = feval (pstData[0], in, 1);
> if(out(0).is_sparse_type())
> {
>   SparseMatrix sparse_matrix=out(0).sparse_matrix_value ();
>   int iNbItem= sparse_matrix.nnz(); //Get number of non zero elements 
> in sparse matrix
> }
>
> How do I get the number of non zero elements in each row and the 
> column indices of the non zero elements?
>
> Shamika

A basic (and probably inefficient) alternative to the suggestion of Kai 
is to use:

[rowind, colind, val] = find(sparse_matrix);

then row indices should be sorted. If not:

[rowind, isort] = sort(rowind);
colind = colind(isort);

colind is now the sorted list of column indices. You still need to find 
out which row they belong to.

You can use a row sum to find out the total number of elements in each row

nnzrow = full(sum((spare_matrix)!=0),2));

In principle there might be empty rows, so the optimal way to proceed 
depends strongly on how you want to access this information later on. 
One possible way is to have two vectors, whose length is the number of 
non-empty rows in the original matrix, one with the index of the row:

posrow = nonzeros(nnzrow);

and the other with the final element in colind belonging to row i, 
assuming the first element is poscol(i-1)+1. Again, there are different 
ways of building poscol, depending on its size. The more obvious option is:

poscol = find(nnzrow);
for i  = 2 : length (poscol)
    poscol(i) = poscol(i-1) + poscol(i);
endfor

I am sure there are simple ways to circumvent the loop, but right now I 
can only think of repmat, triu and sum, which would require creating 
potentially large dense matrix objects.