Vector ranking and transformation matrix

Question

Xia 2016-9-21

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/303917-vector-ranking-and-transformation-matrix

评论： Xia 2016-9-22

Hello. Suppose we have a vector [1 4 3], here -x1+x2>0, -x1+x3>0 and also x2-x3>0. How can we transform this ranking information into a matrix like [-1 1 0; -1 0 1; 0 1 -1]? Is there a function to realize it? Thank you in advance for your time and help.

2 个评论
显示无隐藏无

Stalin Samuel 2016-9-21

Once you evaluate the below details you get the answer
What is the values of x1,x2,x3 ?
How do you relate the given vector with ranking information?
What is the logic behind the final matrix?

Xia 2016-9-21

Thank you Stalin and your evaluation questions are of key. I'm coding for a stochastic dominance problem and the x is productivity vector or return vector, while y is weight vector to solve for. The ranking of y should be reverse of that of x, even thought we don't know it yet and wanna a solution.

Thanks for your comment.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Matt J 2016-9-21

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/303917-vector-ranking-and-transformation-matrix#answer_235575

编辑：Matt J 2016-9-21

在 MATLAB Online 中打开

n=length(x);
A=nchoosek(1:n,2);
m=size(A,1); 
B=sparse(1:m, A(:,1),1,m,n) - sparse(1:m, A(:,2),1,m,n); 
result=full(bsxfun(@times, sign(B*x), B))

6 个评论
显示 4更早的评论隐藏 4更早的评论

Xia 2016-9-21

编辑：Matt J 2016-9-21

在 MATLAB Online 中打开

>> x=[2 4 3 1]';
tic
T=length(x);
X=[x [1:T]'];
k=sortrows(X);
V=k(:,2);
s=1;Q=zeros(T*(T-1)/2,T);
for i =1:T 
for j =1:T-i 
  Q(s,V(i))=-1;Q(s,V(i+j))=1;s=s+1;
end
end
Q
toc
n=length(x);
A=nchoosek(1:n,2);
m=size(A,1); 
B=sparse(1:m, A(:,1),1,m,n) - sparse(1:m, A(:,2),1,m,n); 
result=full(bsxfun(@times, sign(B*x), B))
toc
Q =
       1     0     0    -1
       0     0     1    -1
       0     1     0    -1
      -1     0     1     0
      -1     1     0     0
       0     1    -1     0
Elapsed time is 0.028108 seconds.
result =
      -1     1     0     0
      -1     0     1     0
       1     0     0    -1
       0     1    -1     0
       0     1     0    -1
       0     0     1    -1
Elapsed time is 0.060459 seconds.

Strange. My version is 2014b. Maybe the difference of hardware?.. Anyway, thank you very much Matt. My idea for the loop is that, using the ranking positions of x and set value. Now we have

   1                      1   1
   2   after ranking      3   3
   3                      4   2

1 and -1 are a pair to set value according to ranking information. That's why I feel strange that no 1 in your Q.

Matt J 2016-9-21

在 MATLAB Online 中打开

Hmmm. The discrepancy disappeared after I re-pasted the for-loop code. In any case, here is an improved version for which I see a few factors speed-up over the loops.

    x=randperm(1000).';
    tic
    n=length(x);
    [I,J]=ndgrid(1:n);
    idx=J>I;
    m=nnz(idx);
     B=sparse(1:m,J(idx),1,m,n) - sparse(1:m, I(idx),1,m,n); 
    result=bsxfun(@times, sign(B*x), B);
    toc
    %Elapsed time is 0.685717 seconds.
    tic
    T=length(x);
    X=[x [1:T]'];
    k=sortrows(X);
    V=k(:,2);
    s=1;Q=zeros(T*(T-1)/2,T);
    for i =1:T 
    for j =1:T-i 
      Q(s,V(i))=-1;Q(s,V(i+j))=1;s=s+1;
    end
    end
    toc
    %Elapsed time is 2.316114 seconds.

Xia 2016-9-22

Impressive improvement Matt, especially for high dimensional vectors.

Thank you very much, for your help!

请先登录，再进行评论。

Answer 2

Steven Lord 2016-9-21

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/303917-vector-ranking-and-transformation-matrix#answer_235581

If you're asking how to convert the inequalities (like -x1 + x2 < 0) into matrix form, I don't know if there's a function to do exactly that but the equationsToMatrix function comes close. You may be able to slightly modify your inequalities so they are equations then use equationsToMatrix to generate the matrices to use as your A, b, Aeq, and beq inputs to the Optimization Toolbox solvers (which is how I'm assuming you're planning to use those matrices.)

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Xia 2016-9-21

Thanks a lot Steven, for your insight. You are absolutely right that I'm intended for the optimization. In fact, I want to maximize x'y, while I know x but the unknown y should have a reverse ranking of x. For example x=[1 4 3] so y1 should be the largest y2 the smallest. I can't get any clue on restrictions on ranking of the LP unknown, that's why I think using ranking information of x would be an alternative. Any idea on this Steven?

Thank you very much for your answer!

请先登录，再进行评论。