what is another logic that has execution time faster than randi?

here,part of my following code.

clc
clear all
A=8;
B=12;
C=10;

for ii=1:C;    
    for x=1:2:A-1;
        %make first random matrice
        [vals RandomMatriceA(x,:,ii)]=sort(rand(1,B),2);   
        done=false;
        while ~done,
              %make row of second random matrice
              NewRowRandomMatriceB=randi([0 2],1,B);
              %Make sure sum all element per row in RandomMatriceB <=11
              done=sum(NewRowRandomMatriceB)<12;
        end
        RandomMatriceB(x,:,ii)=NewRowRandomMatriceB;
        ...
    end
    for h=2:2:A;
        doneA=false;
        while ~doneA,
          [vals RandomMatriceA(h,:,ii)]=sort(rand(1,B),2);
          doneB=false;
          while ~doneB,
                 NewRowRandomMatriceB=randi([0 2],1,B);
                 doneB= sum(NewRowRandomMatriceB)<12;
          end
          RandomMatriceB(h,:,ii)=NewRowRandomMatriceB;
        ...
    end
    ...

when I tried to use profile viewer to check my code. above part that makes my code took longer time. so, I want suggestion to speed my above code, specially in "randi"

[Matt Fig]

RANDI is a built-in so it is fast. What are you trying to do overall? Perhaps we can help with a different approach.

[Febri]

well @Matt Fig :

Actually, I want to make a matrices(A,85,C) that is result of combining two matrices (RandomMatriceA and RandomMatriceB). RandomMatriceA is a Matrices (8,12) which has random permutation value [1 12] per row. RandomMatriceB(8,12) which has random value [0 2]. every row In RandomMatriceB must have value [0 2] with length 12 that has sum of them <12. To combine two above matrice, I have to convert both of them to another matrices : ProcessRandomMatriceA,EndProcess,StartProcess, and then StartPost. ProcessRandomMatriceA is Matrices that show value RandomMatriceA converted by ProcessTime Matrices.

ProcessTime= [ 11 11 11 11 4 4 4 4 4 4 4 2 ]; %Converting Matrice show that per column is value in RandoMatriceA( example: in RandomMatrice A has value 3 so, 3 will have value 11).

%Pre-Allocating Size.
RandomMatriceA=zeros(A,B,C);
RandomMatriceB=zeros(A,B,C);
ProcessRandomMatriceA=zeros(A,B,C);
StartProcess=zeros(A,B,C);
EndProcess=zeros(A,B,C);
TimeOff=zeros(A,B,C);
StartPost=zeros(A,B,C);

for ii=1:C;    
    for x=1:2:A-1;
        timeoff=false;
        while ~timeoff,
        %make first random matrice
        [vals RandomMatriceA(x,:,ii)]=sort(rand(1,B),2); 
        doneA=false;
        while ~doneA,
              %make row of second random matrice
              NewRowRandomMatriceB=randi([0 2],1,B);
              %Make sure sum all element per row in RandomMatriceB <=11
              %weeks
              %Make sure maximum consecutive of hoilday <= 12 weeks
              idxA= bitand(RandomMatriceA(x,:,ii) >= 1,...
                    RandomMatriceA(x,:,ii) <= 4);
              NewRowRandomMatriceB([false(size(idxA, 1), 1)...
              idxA(:,1:end-1)]) = 1;
              NewRowRandomMatriceB(idxA)=1;
              doneA=(sum(NewRowRandomMatriceB)<12) &&...
              (~any(diff([0,find(NewRowRandomMatriceB),B+1])>3));
        end
        RandomMatriceB(x,:,ii)=NewRowRandomMatriceB;
        %After making RandomMatriceA and RandomMatriceB, then Make New
        %Matrice
        %To know varible of new matrice which is result of combining
        %RandomMatriceA,RandomMatriceB and ProcessTime
        ProcessRandomMatriceA(x,:,ii)=...
                 ProcessTime(RandomMatriceA(x,:,ii))+...
                 RandomMatriceB(x,:,ii);
        %secondVaribale:to know in which column 
        %The Consecutive value RandomMatrice will be end
        EndProcess(x,:,ii)=cumsum(ProcessRandomMatriceA(x,:,ii),2);
        %ThirdVariable:to know in which column 
        %The Consecutive Value will be start
        StartProcess(x,:,ii)=EndProcess(x,:,ii)-...
       (ProcessTime(RandomMatriceA(x,:,ii))-1);

          %To make sure 43th column only has zeros
          TimeOff(x,:,ii)=(StartProcess(x,:,ii)-...
          RandomMatriceB(x,:,ii)).*(RandomMatriceB(x,:,ii)~=0);
          timeoff=any(TimeOff(x,:,ii)==43);       
          end
          for yy=1:B;
              N=RandomMatriceA(x,yy,ii);
              StartPost(x,N,ii)=StartProcess(x,yy,ii);
          end 
      end
      for h=2:2:A;       
          doneB=false;
          while ~doneB,
            [vals RandomMatriceA(h,:,ii)]=sort(rand(1,B),2);
            doneC=false;
            while ~doneC,
                   NewRowRandomMatriceB=randi([0 2],1,B);
                   %Make sure maximum consecutive holiday <=12 Weeks
                   idxB= bitand(RandomMatriceA(h,:,ii) >= 1,...
                         RandomMatriceA(h,:,ii) <= 4);
                   NewRowRandomMatriceB([false(size(idxB, 1), 1)...
                         idxB(:,1:end-1)]) = 1;
                   NewRowRandomMatriceB(idxB)=1;
                   doneC= (sum(NewRowRandomMatriceB)<12)&&...
                   (~any(diff([0,find(NewRowRandomMatriceB),B+1])>3));
            end
            RandomMatriceB(h,:,ii)=NewRowRandomMatriceB;
            ProcessRandomMatriceA(h,:,ii)=...
             ProcessTime(RandomMatriceA(h,:,ii))+...
             RandomMatriceB(h,:,ii);

             %secondVaribale:to know in which column 
             %The Consecutive value RandomMatrice will be end
             EndProcess(h,:,ii)=cumsum(ProcessRandomMatriceA(h,:,ii),2);
             %ThirdVariable:to know in which column 
             %The Consecutive Value will be start
             StartProcess(h,:,ii)=EndProcess(h,:,ii)-...
             (ProcessTime(RandomMatriceA(h,:,ii))-1);
            TimeOff(h,:,ii)=(StartProcess(h,:,ii)-...
            RandomMatriceB(h,:,ii)).*(RandomMatriceB(h,:,ii)~=0);
            for z=1:B;
              N=RandomMatriceA(h,z,ii);
              StartPost(h,N,ii)=StartProcess(h,z,ii);
            end 
             doneB =all(ismembc(StartPost(h,1:4,ii),...
             sort(StartPost(h-1,1:4,ii)))) && any(TimeOff(h,:,ii)==43);
          end
          %arrange value in column 1 until 4 
          %in order that it will have same
          %value
          StartPost(h,1:4,ii)=StartPost(h-1,1:4,ii);
      end     
  end

[Matt Fig]

To run the above code, I have to know what you use for (A,B,C) or the general relation between them. As it is now, the code you pasted above is not running because these are missing....

Answer 1

[Sven]

I agree with Matt - I bet there are faster ways to run your loop, and we can probably help out if you can describe what the loop is meant to do...

Nevertheless one way to keep your current code structure but speed it up a bit would be to pre-compute a buffer of randi integers between 0 and 2, rather than calling randi multiple times in the loop like this:

A=8;
B=12;
C=10;
bufferSz = 5000;
randBuffer = randi([0 2], bufferSz, B);
randUpTo = 1;

RandomMatriceB = zeros(A,B,C);

for ii=1:C;    
    for x=1:2:A-1;
        %make first random matrice
        [vals RandomMatriceA(x,:,ii)]=sort(rand(1,B),2);   
        done=false;
        while ~done,
              %make row of second random matrice
              if randUpTo>bufferSz
                  randBuffer = randi([0 2], bufferSz, B);
                  randUpTo = 1;
              end
              NewRowRandomMatriceB = randBuffer(randUpTo,:);
              randUpTo = randUpTo + 1;
              %Make sure sum all element per row in RandomMatriceB <=11
              done=sum(NewRowRandomMatriceB)<12;
        end
        RandomMatriceB(x,:,ii)=NewRowRandomMatriceB;
        ...
    end
    for h=2:2:A;
        doneA=false;
        while ~doneA,
          [vals RandomMatriceA(h,:,ii)]=sort(rand(1,B),2);
          doneB=false;
          while ~doneB,
                 if randUpTo>bufferSz
                     randBuffer = randi([0 2], bufferSz, B);
                     randUpTo = 1;
                 end
                 NewRowRandomMatriceB = randBuffer(randUpTo,:);
                 randUpTo = randUpTo + 1;
                 doneB= sum(NewRowRandomMatriceB)<12;
          end
          RandomMatriceB(h,:,ii)=NewRowRandomMatriceB;
        ...
    end
    ...

If you know the absolute maximum number of randi() integers you'll need, you won't even need the "check" that I put in that makes sure you don't try to access integers beyond the buffer size.

Did this speed things up for you Febri?

Oh, and do you initialize RandomMatriceB to a particular size? It doesn't look so..., so I bet that you'll get even more speedup by putting this line at the start of your code:

RandomMatriceB = zeros(A,B,C);

In fact, now that I look at your code a bit more, I think this would be the main issue.

[Matt Fig]

One could even just compute all possible arrays of size 1-by-12 that have the correct sum range, and set them into random order from the get-go. This requires npermutek.

T = npermutek([0 1 2],12); % All perms of [0 1 2] of length 12.
T = T(sum(T,2)<12,:);  % Only those with sum less than 12
T = T(randperm(length(T)),:);  % In a random order.

Now just draw from T. However, I suspect there is more at play and things could maybe be done even better. We won't know until OP comes back....

[Febri]

I think i cannot use npermutek in my old version matlab (R2009a). so what is the similar logic with it Mr.?

[Matt Fig]

Did you click the link to download the file?? I gave the link in my previous comment. The function works in r2007b so it should work in r2009a!