Why am I receiving this error when trying to implement sparse?

I am trying to modify this code from file exchange. It works perfectly, but it won't allow me to use data that includes zeros. I cannot figure out why or how to modify it so it will allow for zeros and the code is useless unless it can do this.

Here is the error: Error using sparse, Index into matrix must be positive.

And here is the partially modified code.

markovChain = [2 3 3 3 0 4 3 2 4 0 2 1 4];
Norder = 2;

Nstates = max(markovChain);

if Norder == 1,

      %get transition matrix
      transitionMatrix = full(sparse(markovChain(1:end-1),markovChain(2:end),1,Nstates^Norder,Nstates^Norder));

else

      %get Norder-contiguous sequences of the markov chain
      ngrams = [];
      for i = 0:Norder-1
          ngrams = [ngrams, circshift(markovChain,[0 -1*(i)])'];
      end
      ngrams = cellstr(num2str( ngrams));
      ngrams = ngrams(1:end-Norder);

      %create all combinations of Norder-contiguous sequences
      [x{1:Norder}] = ndgrid(1:Nstates);

      %format x to cell
      evalStr = ('xCell = cellstr(num2str([');
      for i = 1:Norder
          evalStr = [evalStr 'x{' num2str(i) '}(:) '];            
      end
      evalStr = [evalStr ']));'];
      eval(evalStr);

      %map ngrams to numbers
      [gn,~,g]=unique([xCell;ngrams]);
      s1 = g(Nstates^Norder+1:end);

      %states following the ngrams
      s2 = markovChain(Norder+1:end);

      %reordered xCell.
      columnStates = gn(1:Nstates^Norder);    

      %get transition matrix
      transitionMatrix = full(sparse(s1,s2,1,Nstates^Norder,Nstates^Norder));

end

transitionMatrix

%plot the transition matrix
    imagesc(transitionMatrix);
    str= 'Transition Matrix of the Markov Chain:\n';
    str=[str sprintf('%d',markovChain) '...'];
    title(sprintf(str));    

      %replace tickLabels with states.
      set(gca,'YTick',1:numel(columnStates));
      set(gca,'YTickLabel',columnStates);
      set(gca,'XTick',1:Nstates);
      set(gca,'XTickLabel',1:Nstates);

[Jonathan]

Another reason for "Error using sparse, Index into matrix must be integer" is if there are NaNs in the vector. I just found that out with a binary search on a very long vector! :)

Answer 1

[James Tursa]

The variable markovChain (or s1 or s2) has 0's in it, and you are attempting to use this for indexes when building the sparse matrix. MATLAB does not allow 0 indexes, hence the error message you are getting.

You will have to reformulate your program to avoid using 0 indexes.

[Walter Roberson]

Perhaps just add 1 to the indices.