how to vectorize nested for loops

Hello everyone

I want to vectorize the following code to reduce operation time. Can you guys help me out?

thanks

% X is a double matrix
s=size(X);
RepSol=repmat(X,2,2);

% Y is a binary matrix
Y=repmat(Y,2,2);

% Preallocation
p=zeros(s);

for m=1:s(1)
    for n=1:s(2)
        for i=1:s(1)
            for j=1:s(2)
                if RepSol(i,j)==RepSol(m+i-1,n+j-1)
                    p(m,n)=p(m,n)+Y(i,j)*Y(m+i-1,n+j-1);
                end
            end
        end
    end
end

Answer 1

[darova]

IS it correct?

for m=1:s(1)
    for n=1:s(2)
        cond = RepSol(1:s(1),1:s(2)) == RepSol((1:s(1))+m-1,(1:s(2))+n-1);
%         cond = X - RepSol((1:s(1))+m-1,(1:s(2))+n-1);
        P = Y(1:s(1),1:s(2)) .* Y((1:s(1))+m-1,(1:s(2))+n-1);
        res = cond .* P;
        p(m,n) = sum(res(:));                
    end
end

[Bob Thompson]

cond = RepSol(1:s(1),1:s(2)) == RepSol((1:s(1))+m-1,(1:s(2))+n-1);

A couple of things:

RepSol(1:s(1),1:s(2))

This is just the entirety of RepSol, you shouldn't need to index it

RepSol((1:s(1))+m-1,(1:s(2))+n-1)

Your indexing here doesn't make sense. (1:s(1)) is just calling all rows, but then you're trying to adjust that range to +m-1. So, suppose you have ten rows, and you are looking at m = 5, then you are saying you want to look at rows 5:15, which isn't possible because you only have 10 rows.

Unfortunately, I don't have a way to solve your problem off the top of my head, but it might help us come up with a solution if you explain what you're trying to do with your loops. There might be a command which does it already.

[darova]

s(1) and s(2) are size of RepSol before repmat (two times less)

s=size(X);
RepSol=repmat(X,2,2);

[AliHg]

Dear darova

Thank you for your prompt response. Your code produce the same answer as mine, yet it does not improve the speeding as much. The reason the speed is incredibly important to me is that this code is a function of a much larger code and it needs to be computed more than 30.000 times; hence I need it to be as simple as possible. My own solution for improving the speed was to vectorize the nested "for" loops and avoid using Repmat cause according to MATLAB Profiler, the "if" part in my code consumes most of the runtime.

I am wondering if it is feasible to vectorize the two remaining "for" loops in your code.

[AliHg]

Dear Bob

Thank you for putting your time and energy into my question. As I mentioned in my previous comment, derova’s code produces identical answers but does not improve the speed.

To answer to your question I have to say in my knowledge there is no function in MATLAB that does the same thing. Also, I cannot go through details because the code is somehow simple and hard to explain at the same time.

Answer 2

[AliHg]

Can anyone suggest anything regarding boosting my code? I have tried everything but nothing seems to workout.

[darova]

Can you tell something more about your problem? What are yuo working on now?

Maybe there is some solution

[AliHg]

Thank you for your response. Repsol matrix contains cluster numbers and matrix p do some calculation for identical cluster numbers.

[darova]

Unfortunately is explains nothing

The script does the follwoing if i understood correctly:

[Rik]

I agree with darova that your explanation is minimal. Please edit your question, adding what you're trying to achieve. Maybe your problem can be solved with a well-chosen convolution, but with the current level of detail it is impossible to give you a better solution than you already have been given.