Vectorized operation of sparse matrix

I have the following code

a = randi(903,10293,1800);
c = cell(1,10293);
for n = 1:10293
    b = a(n,:);
    
    i = b(1:length(b)-1);
    j = b(2:length(b));
    k = ones(1,length(b)-1);
    
    s = sparse(i,j,k,903,903);
    total = nansum(s,2);
    p = s./total;
    c{n} = p;
    
end

After it runs for sometime, matlab runs out of memory. Without the """ total = nansum(s,2); p = s./total; """ part, and setting c{n} =s, it runs fine. Can anyone tell me how to fix this and make the code run efficiently?

Answer 1

[Matt J]

The problem is that you have lots of 0/0 operations occuring whenever total=0. These result in lots of NaNs in p rendering it highly non-sparse. I don't know how you wish to define the results of 0/0 in this situation, but here is one possibility for avoiding them:

a = randi(903,10293,1800);
c = cell(1,10293);
for n = 1:10293
    b = a(n,:);
    
    i = b(1:length(b)-1);
    j = b(2:length(b));
    k = ones(1,length(b)-1);
    
    s = sparse(i,j,k,903,903);
    total = nansum(s,2);
    total(~total)=1;  %<---Matt J inserted
    p=s./total; 
    c{n} = p;
    
end

[Esegboria Osarhemen]

Thank you very much, it works now, just how i wanted