For Loop taking too long to execute.

My for loops are taking too long to execute. I am writing my code here. Is there any possibility of improving my code so that it takes less time or can I completely bypass the for loops?

   Transitionbwd = zeros(2048,11) ;
   StateTransitionbwd = zeros(2048,2048);

 for k = 1:2048
    for l = 1:11
 for i = 1:2048
    for j = 1:11

        if inputfwd(k,l) == 0

            Transitionfwd(i,j) =  1 - 0.001;

        elseif norm(Statesfwd(k,l) - sjfwd(i,j)) == Statesfwd(k,l)

                        Transitionfwd(i,j)   = 0.5  -   0.5*tanh(0.5 * inputfwd(k,l));     
             else

                        Transitionfwd(i,j) = 0.5  +   0.5*tanh(0.5 * inputfwd(k,l));
        end
    end
end

        dim = 2;
        StateTransitionfwd(k,:) = prod(Transitionfwd,2); 

    end

    end

Answer 1

[Walter Roberson]

You could remove your

elseif norm(Statesfwd(k,l) - sjfwd(i,j)) == Statesfwd(k,l)

and the associated action. Due to numeric roundoff in finite precision binary floating point, values computed in even slightly different ways will seldom compare as equal for the purposes of "==". The "==" comparison checks for bit-wise identical (non-NaN) numbers. As you will only get equality by accident, you might as well remove that test.

I am assuming here that you consider your existing loops to be correct but just too slow. There is an alternative interpretation, which is that your existing code is not correct, and that instead of comparing using "==" you want to check to see if the norm is "close to" the stored value, for some definition of "close to".

http://matlab.wikia.com/wiki/FAQ#Why_is_0.3_-_0.2_-_0.1_.28or_similar.29_not_equal_to_zero.3F

[Salman Saeed]

Actually Statesfwd(k,l) and sjfwd(i,j) are 0 or 1. I want to use the mentioned formula if sjfwd(i,j) is 0. Do you think it can still have the same problem of floating points?

[Walter Roberson]

Why not just test if sjfwd(i,j) == 0 ?