MATLAB: Alternate to the circshift function in a 3d matix for an agent based model

3 dimensional matrixagent based modelcircshiftefficient solution

Hey there,
I have a 3 dimensional matrix randworld and need to perform certain calculations on each the row and column cell around a randomly selected cell holding the third dimension constant. I used the circshift function for this. However as I need to do 121000000 iterations, this method is very slow. I noticed the circshift function consumes most of the time (from profile viewer). Could you please suggest an alternative to the circshift function.
I have copied the code below so that you have an idea of what I'm looking to do.
Thank you.
   nside=11; %rows and coloumns
   u=120; %neighbours
    randworld=randi(10,nside,nside,5); %create a random 3-d world
%select a random agent and feature
   randomagentr=randi(10,1);
   randomagentc=randi(10,1);
   randomfeat=randi(5,1);
   nsize=0;
%neighbourhood
              for q=-5:5
                  for w=-5:5
                      if abs(q)+abs(w)~=0
                      nsize=nsize+1;
                      neigh(nsize,:)=[q,w];
                      end
                  end
              end
   for k=1:u
              vm=circshift(randworld,[neigh(k,1),neigh(k,2)]);
              v=vm(randomagentr,randomagentc,randomfeat);
              %perform certain computations on this new variable v
  end

Best Answer

If a circular shift by +1 done with indexing is significantly faster than your 'circshift' operations, then you can do the equivalent of your circular shifts in the last nested for-loops as a succession of such "index" shifts by +1 and it would then presumably be faster. Don't create the array 'neigh'. Just do this:
 vm = circshift(squeeze(randworld(:,:,randomfeat)),[-5,-5]);
 p = [11,1:10]; % <-- Another way to do circular shift by +1
 for ir = 1:11
   for ic = 1:11
     v=vm(randomagentr,randomagentc);
     %perform certain computations on this new variable v
     vm = vm(:,p); % Circular column shift by +1
   end
   vm = vm(p,:); % Circular row shift by +1
 end
You will have to test for the case ir == 6 and ic == 6 to avoid the case when there is to be no overall shift in order to get just 120 instead of 121 steps.
Related Question