speed up the calculation in multidimensional array

Array:RP_bin

Matrix: matrix

[~,c]=size(matrix);


xx=find(RP_bin);
q=length(xx);

[~,c]=size(matrix);

cor=zeros(c,c,q);


for i=1:q
  for gg=1:c
       for yy=1:c
              a=matrix(1:xx(i),gg);
              b=matrix(1:xx(i),yy);
              cc=sum((a>0 & b>0)|(a<0 & b<0));
              cc1=sum(a~=0 & b~=0);  
                cor(yy,gg,i)=round(100 * cc/cc1) / 100; %cc/cc1; %%array multidimension
              end            
       end
   end 


>> size(cor)

ans =

    82    82   188


to execution this code is :
Elapsed time is 21.425685 seconds.

is possibile to speed loop?

[James Tursa]

Can you provide a small example of matrix and RP_bin?

[aldo]

yes ..attachment it

Answer 1

[Dyuman Joshi]

You can reduce a for loop by vectorizing the code -

load('RP_BIN.mat')
load('matlab.mat')

[~,c]=size(matrix);

xx=find(RP_bin);


q=numel(xx);

%Preallocation
[cor,COR]=deal(zeros(c,c,q));


%%Original approach
tic
for i=1:q
    for gg=1:c
        for yy=1:c
            a=matrix(1:xx(i),gg);
            b=matrix(1:xx(i),yy);
            cc=sum((a>0 & b>0)|(a<0 & b<0));
            cc1=sum(a~=0 & b~=0);
            cor(yy,gg,i)=round(100 * cc./cc1)/100; %cc/cc1; %%array multidimension    
            
        end
    end
end
toc

Elapsed time is 18.790586 seconds.

%%Modified approach
tic
gg=1:c;
for i=1:q
    %As the array a is only dependent on the outer loop, bring it out of
    %the inner loop, so that it isn't re-calculated in every iteration of inner loop
    %thereby increasing the speed of the code
    a=matrix(1:xx(i),gg);
    for yy=1:c
        b=matrix(1:xx(i),yy);
        cc=sum((a>0 & b>0)|(a<0 & b<0),1);
        cc1=sum(a~=0 & b~=0,1);
        COR(yy,gg,i)=round(100 * cc./cc1)/100; %cc/cc1; %%array multidimension
    end
end
toc

Elapsed time is 9.287863 seconds.

%%Comparing the output obtained
%As the calculations have NaN data, use isequaln()
isequaln(cor,COR)

ans = logical
   1

As you can see that there is significant improvement in speed (more than 50%)

[aldo]

Thank you but the above code makes me the correlation between 2 arrays calculated differently by the "corrcoef" function. If I calculate the original matlab function it takes me 0.3 seconds

for i=1:q

cor(:,:,i)=corrcoef(matrix(1:xx(i),:));

end

Elapsed time is 0.360681 seconds.

(0.36 vs 9.28)

there is always a lot of difference in duration

[Dyuman Joshi]

Then why not use that only?

You asked if the loop can be sped up, I responded accordingly. You did not say that you were comparing it to an inbuilt MATLAB function.

[aldo]

it is a correlation calculated in a different way and the result is different .. I am interested in the result of my code not that of matlab

anyway thanks for your code