Need help for vectorization?

Hello

This program is running very slow. Can anyone please help me to make it fast like by vectorizing this code.

Thanks

Note that---Pop_Size,n,dr,f,S,T are my inputs. i am giving f,S,T input as matrix like[f S T]

 for u= 1:Pop_Size

 ridx = 1;

 c=n-1;

 output = zeros(n,n*dr);

 for z = 1:n:n*dr

    M = zeros(c);

    f   = inputData(ridx,1); % f= Demand

    S   = inputData(ridx,2); % S= Source node
    T   = inputData(ridx,3); % T= Termination node

    ridx = ridx + 1;

   for k = 1:f

        p = randperm(c);
        for s = 1:c

            M(p(s),s) = M(p(s),s) + 1;
        end

    end
      M = [M(:,1:S-1),zeros(c,1),M(:,S:c)];  % Add column of zeros

      M = [M(1:T-1,:);zeros(1,c+1);M(T:c,:)];% Add row of zeros

      M(1:(c+2):(c+1)*(c+1))= 0;

      output(:, z:z+c) = M;

end

 outp(:,:,u) = output(:,:);      %To save all the chromosomes

 fprintf('Chromosome %d\n',u);

 disp(output);

end

Answer 1

[Roger Stafford]

In your Aug. 20 Answers #151724 article "Please help me to modify the following program" I showed how you could replace all your current code that creates the M matrix with just three lines of code using the 'accumarray' function. It replaces two nested for-loops. This could possibly improve your execution time. Why don't you try it?

[Guillaume]

If output is significantly large, then, yes, the resizing of outp will slow down your program as matlab has to copy the whole content into a new memory slot each time. You can:

Either, predeclare outp:

outp = zeros(n, n*dr, Pop_Size);

Or store output into a cell array:

outp{u} = output;

Answer 2

[Guillaume]

I'm not sure there's much that can be vectorised in there. An alternative for your for k loop could be:

coloffset = 0:c:c^2-1;
for k = 1:f
  indices = randi(c, 1, c) + coloffset;
  M(indices) = M(indices) + 1;
end

But on my machine it's only faster for c>70.

The insertion of a row and column of zero is going to be expensive, so instead I would do:

M = zeros(c+1); %i.e. M = zeros(n);
%randomly add 1 to a row of column 1:n, but never in last row, using whichever method
%with your method change s to go from 1:n (but p still randperm(c))
%with mine change randi to be randi(c, 1, n) and coloffset to be 0:n:n^2-1
M(:, S) = 0; %set column to 0
M(end, :) = M(T, :); %swap row T with last row
M(T, :) = 0;

Finally, you could change setting the diagonal to 0 to:

M(logical(eye(n)) = 0;

Don't know if it's faster.