Is there any solution to make positive semi-definite covariance matrix?

Question

K M Ibrahim Khalilullah 2016-7-5

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/293735-is-there-any-solution-to-make-positive-semi-definite-covariance-matrix

评论： Walter Roberson 2016-7-5

I have image data. row of the data represents pixel and column represents rgb values. that is, the size of the data is N by 3. However, I would like to apply kmeans and GMM for clustering. But the covariance matrix is not positive semi-definite.

Why and what is solution?

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。

Answer 1

Walter Roberson 2016-7-5

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/293735-is-there-any-solution-to-make-positive-semi-definite-covariance-matrix#answer_227680

在 MATLAB Online 中打开

Covariance matrices of real variables are always positive semi-definite. See http://math.stackexchange.com/questions/114072/what-is-the-proof-that-covariance-matrices-are-always-semi-definite

If you calculated a covariance matrix but it is being reported as not positive semi-definite then you might have encountered numeric round-off. In such a case use

(X.' + X) / 2

in place of X: this will eliminate any round-off error that is keeping the matrix from being completely symmetric.