Main diagonal operations problem
조회 수: 6 (최근 30일)
이전 댓글 표시
Hi guys, I need your help.
I want to create a matrix(4,4) in which the main diagonal have values between 0.3 and 1 and the other cells assume values such as to have a horizontal sum equal to 1.
By now I'm using the following code but the only result is to have a main diagonal composed by the same numbers:
x = eye(4)
x(1,1) = 1+(0.3-1)*rand(1,1)
x(2,2) = x(1,1)
x(3,3) = x(1,1)
x(4,4) = x(1,1)
Any suggestion?
PS : I've tried even with diag
댓글 수: 2
Stephen23
2019년 9월 13일
편집: Stephen23
2019년 9월 13일
"...the other cells assume values such as to have a horizontal sum equal to 1"
Are there any other requirements on the other elements? Positive, negative, fractional values, integer, >1, >100, >1e100 ... what values are allowed?
What is the "horizontal sum": do you mean to sum along the 2nd dimension (i.e. along each row) ?
채택된 답변
Stephen23
2019년 9월 13일
편집: Stephen23
2019년 9월 13일
>> N = 4; % matrix size
>> M = nan(N,N); % preallocate
>> V = 0.3+(1-0.3)*rand(1,1) % diagonal value
V =
0.47505
>> M(~eye(N)) = randfixedsum(N-1,N,1-V,0,1); % other values
>> M = M.'; % transpose
>> M(1:N+1:end) = V % assign diagonal value
M =
0.47505 0.40657 0.0087969 0.10958
0.12917 0.47505 0.21287 0.1829
0.35794 0.15519 0.47505 0.011825
0.41335 0.032696 0.078907 0.47505
Checking the sum of each row:
>> sum(M,2)
ans =
1
1
1
1
and diagonal:
>> diag(M)
ans =
0.47505
0.47505
0.47505
0.47505
댓글 수: 6
John D'Errico
2019년 9월 13일
편집: John D'Errico
2019년 9월 13일
Yes. I had come to that conclusion after some thought too. But there still seems to be the question of what is the true goal, since only Giuseppe can know that. I think Stephen's solution comes closer than mine.
On the other hand, HAD I generated the diagonal elements using a better distribution than uniform, then my solution would be an avenue to not needing to use a rejection while loop at all.
Bruno Luong
2019년 9월 13일
편집: Bruno Luong
2019년 9월 14일
Oh I see your point. I think you are right by forcing the right PDF to generate D, it's equivalent to rejection method. This I'm pretty sure it's right because of the special property of simplex and barycentric coordinates.
So the right way (I simplify the procedure to a single row without loosing the generality) is
N = 4;
dmin = 0.3;
dmax = 1;
d = dmax-(dmax-dmin)*rand().^(1/(N-1)); % rather than dmin+(dmax-dmin)*rand;
v = (1-d) * randfixedsum(N-1,1,1,0,1);
then insert d to v....
추가 답변 (3개)
John D'Errico
2019년 9월 13일
편집: John D'Errico
2019년 9월 13일
Easy enough, it seems. First, determine the diagonal elements.
x = diag(rand(1,4)*.7 + .3);
Next, you need to choose the other row elements randomly so the sum will be 1. But that sum will now depend on the diagonal element you just chose. Stilll simple, as long as you use randfixedsum, by Roger Stafford, found on the file exchange.
for i = 1:4
x(i,setdiff(1:4,i)) = randfixedsum(3,1,1 - x(i,i),0,1)';
end
Did it work? Of course.
x
x =
0.83586 0.075979 0.057706 0.030454
0.012356 0.85664 0.11425 0.016757
0.13748 0.21163 0.43081 0.22009
0.15838 0.037488 0.16129 0.64284
>> sum(x,2)
ans =
1
1
1
1
Find randfixedsum here:
댓글 수: 0
Bruno Luong
2019년 9월 13일
편집: Bruno Luong
2019년 9월 13일
Here is a method that has two advantages:
- without the need of Roger's FEX randfixedsum
- Produce matrix with rigourous uniform conditional probability
N = 4; % matrix size
% diagonal lo/up bounds
dmin = 0.3;
dmax = 1;
% random (common) diagonal value
d = dmax-(dmax-dmin)*rand().^(1/(N-1)); % Edit see comment above, equiv to rejection method
% d = dmin+(dmax-dmin)*rand;
% Generate N random vectors of length N-1 required sum == (1-d)
V = -log(rand(N-1,N)); % Marsaglia's [1961] method
V = V .* ((1-d)./sum(V,1));
% Arrange in the final matrix
A = zeros(N);
isdiag = sparse(1:N,1:N,true);
A(isdiag) = d;
A(~isdiag) = V(:);
A = A.';
% Check result
disp(A)
sum(A,2)
댓글 수: 0
참고 항목
카테고리
Help Center 및 File Exchange에서 Creating and Concatenating Matrices에 대해 자세히 알아보기
제품
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!