Timeline for Can I fix a non-positive definite correlation matrix to study partial correlations?
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 26, 2016 at 22:13 | vote | accept | Firebug | ||
| Aug 30, 2016 at 1:31 | history | edited | Firebug | CC BY-SA 3.0 |
added 61 characters in body
|
| Aug 30, 2016 at 1:26 | answer | added | Mark L. Stone | timeline score: 2 | |
| Aug 29, 2016 at 23:49 | comment | added | Firebug | @MarkL.Stone $\mathbb C$ is a correlation matrix. It's the pearson coefficient denoting pairwise correlations between some input vectors. The smallest eigenvalues are indeed $\mathcal{O}(1E-8)$. | |
| Aug 29, 2016 at 21:00 | comment | added | Mark L. Stone | I presume you mean C are covariance matrices. How are your non-positive definite covariance matrices generated (computed) and how far do they deviate from positive definiteness? For example, is smallest eigenvalue negative with magnitude less than 1e-8, which might be roundoff error away from positive definiteness? See stats.stackexchange.com/questions/63817/… for some ways of making positive definite. Your results (inverse diagonal elements) will be very sensitive to what you do - not good. | |
| Aug 29, 2016 at 20:33 | history | asked | Firebug | CC BY-SA 3.0 |