vectorisation of a loop

6 次查看（过去 30 天）

显示更早的评论

Zaid B 2021-11-30

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1599564-vectorisation-of-a-loop

编辑： Matt J 2021-12-6

在 MATLAB Online 中打开

the matrix is "Sea" of m*n dimension .. this is the code i would like to vectorize :

=======================

S = size(Sea); 
i = 1;
while i < S(1)
    indices = find ( pdist2( Sea( i , : ),Sea( i+1:S(1) , :)) <0.05 );
    Sea(indices , : ) = [];
    S = size(Sea);
    i = i+1;
end

===========

desc : i'm trying to calculate the distance between each row and all the other rows in the matrix and delete the ones that are closer than 0.05

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

回答（2 个）

Matt J 2021-11-30

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1599564-vectorisation-of-a-loop#answer_844069

编辑：Matt J 2021-12-1

在 MATLAB Online 中打开

[m,n]=size(Sea);
D=pdist2(Sea,Sea);
D(1:m+1:end)=inf;
indices=any(triu(D<0.05),1);
Sea(indices,:)=[];

12 个评论
显示 10更早的评论隐藏 10更早的评论

Zaid B 2021-12-2

编辑：Zaid B 2021-12-2

thank you for answering , but why ( i - 1 ) is added on that line?!

("Moreover, I don't think there is any more optimization that can be done" ... i thought so , but wanted to check if anyone has a better idea that would help)

Zaid B 2021-12-2

编辑：Zaid B 2021-12-2

i got it why is nedded there but i think "i + find(...)" not "(i-1)+find(..)" because it should keep the current i , thank you that was a great help there ^^

请先登录，再进行评论。

Matt J 2021-12-2

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1599564-vectorisation-of-a-loop#answer_845980

编辑：Matt J 2021-12-2

在 MATLAB Online 中打开

I find this version to be about 30% faster:

    S = size(Sea,1); 
    i = 1;
    t=0.05^2;
    
    while i < S
        indices=true(1,S);
        indices(i+1:end) =  sum( abs( Sea( i , : )-Sea( i+1:S , :) ).^2,2) >= t ;
        
        Sea=Sea(indices , : );  
        
        S = size(Sea,1);
        i = i+1;
        
    end

13 个评论
显示 11更早的评论隐藏 11更早的评论

Zaid B 2021-12-5

在 MATLAB Online 中打开

    Sea  = rand(10000,14);
tic
P1 = version1(Sea,1); %my version always faster 
ans = 1×2
        9647          14
checkmin(P1)
ans = 0.5000
V1 = toc
V1 = 7.7813
tic
P2 = version2(Sea,1);% your version slighly slower but doing the job needed
checkmin(P2)
ans = 0.5000
V2 = toc
V2 = 11.0393
tic 
P3 = version3(Sea,1); %verision three you proposed 
checkmin(P3)
ans = 0.3358
V3 = toc
V3 = 4.1707
TRUE  = isequal(P1,P3) %cheking if its the same matrix returned from previous versions
TRUE = logical
   0
function sea = version1(m,step)
S = size(m);
SeaCopy= [];
i = 1;
while i < S(1)
    SeaCopy = [SeaCopy ; m(i,:)];
    indices = i + find(pdist2(m(i,:),m(i+1:S(1),:))<0.5);
    if(length(indices) > 1)
        SeaCopy = [SeaCopy ; mean(m(indices,:))];
        m(indices,:)=[];
    else
        m(indices,:)=[];
    end
    S = size(m);
    i = i +step;
end
size(SeaCopy)
sea = m;
end
function sea = version2(m,step)
SeaCopy= [];
S = size(m);
i = 1;
t=0.5^2;
while i < S(1)
    SeaCopy = [SeaCopy ; m(i,:)];
    indices=true(1,S(1));
    indices(i+1:end) =  sum( abs(m( i , : )-m( i+1:S(1) , :) ).^2,2) >= t ;
    SeaCopy = [SeaCopy ; mean(m(indices,:))];
    m = m(indices , : );
    S = size(m,1);
    i = i+1;
end
sea = m;
end
function sea = version3(m,step)
    S = size(m,1);
    i = 1;
    t=0.05^2;
    while i < S
    indices=false(1,S);
    indices(i+1:end) =  sum( abs(m( i , : )-m( i+1:S , :) ).^2,2) < t ;
    m(indices , : )=[];
    S = size(m,1);
    i = i+1;
    end
    sea = m;
end
function val=checkmin(Sea)
D=pdist2(Sea,Sea);
val=min(D(D>0));
end

Matt J 2021-12-5

编辑：Matt J 2021-12-5

在 MATLAB Online 中打开

Something must have been miscopied, because I'm getting identical results. Below is a comparison of all 3 versions, showing they are equal.

Sea  = rand(8000,14);
tic
P1 = version1(Sea,1);
time_V1 = toc
time_V1 = 2.5249
tic
P2 = version2(Sea,1); 
time_V2 = toc
time_V2 = 2.5527
tic
P3 = version3(Sea,1);
time_V3 = toc
time_V3 = 1.2148
isequal(P2,P1)
ans = logical
   1
isequal(P3,P1)
ans = logical
   1
function sea = version1(m,step)
    S = size(m);
%     SeaCopy= [];
    i = 1;
    while i < S(1)
%         SeaCopy = [SeaCopy ; m(i,:)];
        indices = i + find(pdist2(m(i,:),m(i+1:S(1),:))<0.05);
        if(length(indices) >= 1)
%           SeaCopy = [SeaCopy ; mean(m(indices,:))];
            m(indices,:)=[];
        end
        S = size(m);
        i = i +step;
    end
    sea = m;
end
function sea = version2(m,step)
    S = size(m);
    i = 1;
    t=0.05^2;
    while i < S(1)
    indices=true(1,S(1));
    indices(i+1:end) =  sum( abs(m( i , : )-m( i+1:S(1) , :) ).^2,2) >= t ;
    m=m(indices , : );
    S = size(m,1);
    i = i+1;
    end
    sea = m;
end
function sea = version3(m,step)
    S = size(m,1);
    i = 1;
    t=0.05^2;
    while i < S
    indices=false(1,S);
    indices(i+1:end) =  sum( abs(m( i , : )-m( i+1:S , :) ).^2,2) < t ;
     m(indices , : )=[];
    S = size(m,1);
    i = i+1;
    end
    sea = m;
end