Problems specifying nzmax in sparse command from MATLAB -

Problems specifying nzmax in sparse command from MATLAB -

i create sparse matrix of size n × n. usual i, j row , column indices, s values. command

s=sparse(i,j,s,n,n);

works fine, uses lot of memory during creation of s. according documentation matlab uses nzmax=length(s) default. save memory, want specify nzmax s can compute in advance. testing purpose (i created s above command) calculate

nonzeros=nzmax(s);

and call

s=sparse(i,j,s,n,n,nonzeros);

note in illustration nonzeros much smaller length(s). also, nnz(s)=nzmax(s). error message

error using sparse: index exceeds matrix dimensions.

can explain behavior me, please? possible remedy/workaround?

let me add together code snippet example

i=randi(10,1000,1); j=randi(10,1000,1); s=rand(1000,1); ell=size((unique([i j],'rows')),1); s=sparse(i,j,s,10,10,ell); error using sparse: index exceeds matrix dimensions. s=sparse(i,j,s,10,10); nnz(s) ans = 100 ell ell = 100 nzmax(s) ans = 100

how using accumarray, specifying you're working sparse matrices (the 6th argument):

n = 100; num = 10000; = randi(n,num,1); j = randi(n,num,1); s = ones(num, 1); = accumarray([i, j], s, [n,n], [], 0, true);

this uses next construction accumarray:

a = accumarray(subs,val,sz,fun,fillval,issparse)

an alternative utilize spalloc first. way should avoid default initialization.

a = spalloc(n, n, non_z); a(sub2ind(size(a),i,j)) = s;

yet alternative utilize unique 'rows' sec input count number of unique index pairs.

uni_ind = numel(unique([i,j], 'rows')); = sparse(i, j, s, n, n, uni_ind);

matlab sparse-matrix