Timeline for Estimating correlation matrix using numeric likelihood maximization
Current License: CC BY-SA 4.0
17 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 16, 2020 at 0:33 | answer | added | Anirban Mukherjee | timeline score: 0 | |
| Apr 22, 2019 at 18:41 | answer | added | dave fournier | timeline score: 2 | |
| Apr 22, 2019 at 16:55 | vote | accept | Felipe D. | ||
| Apr 22, 2019 at 16:55 | answer | added | Felipe D. | timeline score: 1 | |
| Aug 5, 2018 at 21:12 | comment | added | Felipe D. | Since I'm not providing the algorithm with an analytical gradient, it's just using a numerical gradient. I just found a quick-and-dirty reference on how to correct non PSD matrices by using the eigenvalue method you mentioned. I'll try to implement that and see if implementing it in the pairwise approach works better! | |
| Aug 5, 2018 at 20:45 | comment | added | Mark L. Stone | That link provides some references on dealing with negative eigenvalues. Is your optimization using finite difference gradient? I don;t see where gradient is specified. Not sure what optimizer did - says it did one gradient evaluation, but a lot of (objective) function evaluations, so maybe failed to descend and gave up after a lot of flailing, but in such case, it shouldn't report success. Can you turned on a more detailed level of reporting? What is the objective value at the starting point. BTW, what they call "function value" is the objective function value. | |
| Aug 5, 2018 at 20:18 | comment | added | Felipe D. | The thing is that this is a small part in other larger estimation problems I have to solve. The main problem I deal with is the Generalized Ordered Probit Model, where I have multiple (discrete) ordered outcomes and the error terms are jointly distributed. In this problem, we estimate the influence of a bunch of exogenous covariates as well as the correlation terms between the errors (more info here). So I tried to translate the simplest version of the problem to a clean-cut context to present it here. | |
| Aug 5, 2018 at 20:16 | comment | added | Felipe D. | The output I get is this: "Optimization terminated successfully. Current function value: 23025.850930 Iterations: 0 Function evaluations: 1277 Gradient evaluations: 1" | |
| Aug 5, 2018 at 20:00 | comment | added | Mark L. Stone | What output is displayed when the optimizer terminates after one iteration? By virtue of your approach (even if no mistakes), you may have introduced spurious saddle points. .Is there a reason why you can;t just form the empirical covariance matrix - do you have unclean, inconsistent data (not all variable components measured together)? If not, then other than roundoff errors, empirical covariance should be psd. You can adjust eigenvalues or use other methods to adjust an almost psd "covariance" or correlation matrix to be psd, or have minimum eigenvalue. | |
| Aug 5, 2018 at 19:55 | history | edited | Felipe D. | CC BY-SA 4.0 |
added 2 characters in body
|
| Aug 5, 2018 at 19:32 | history | edited | Felipe D. | CC BY-SA 4.0 |
edited body
|
| Aug 5, 2018 at 19:25 | history | edited | Felipe D. | CC BY-SA 4.0 |
added 3754 characters in body
|
| Aug 5, 2018 at 19:18 | history | edited | Felipe D. | CC BY-SA 4.0 |
added 3754 characters in body
|
| Aug 5, 2018 at 18:40 | comment | added | Felipe D. | Good point. I'll edit the original post to reflect that. Thx for the heads up! | |
| Aug 5, 2018 at 17:32 | comment | added | Mark L. Stone | You don't seem to have told us what optimization problem you are solving with scipy.optimize. Despite your long presentation, I have essentially no idea what you've done, other than relying on a Choelsky factorization to ensure positive semidefiniteness. You apparently populate a Choelsky factor with a random entries, but is that just an initialization (starting value) for numerical optimization via some unstated optimization problem formulation? | |
| Aug 5, 2018 at 17:17 | history | edited | Felipe D. | CC BY-SA 4.0 |
Issues while trying to numerically estimate correlation matrix through likelihood maximization.
|
| Aug 5, 2018 at 4:59 | history | asked | Felipe D. | CC BY-SA 4.0 |