Timeline for Computation of the marginal likelihood from MCMC samples
Current License: CC BY-SA 3.0
10 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 30, 2016 at 18:37 | vote | accept | lacerbi | ||
| Apr 30, 2016 at 18:00 | answer | added | Xi'an | timeline score: 32 | |
| Apr 29, 2016 at 15:48 | comment | added | lacerbi | (+1) Thanks for the reference, looks spot on -- I will check it out. I agree that all model-based approaches can be problematic (the good thing with Bayesian quadrature is that you get an estimate of uncertainty, although not sure how calibrated it is). For the moment my modest goal is to do something that is "better than a Laplace approximation". | |
| Apr 29, 2016 at 9:31 | comment | added | Florian Hartig | I think Chib, S. and Jeliazkov, I. 2001 "Marginal likelihood from the Metropolis--Hastings output" generalizes to normal MCMC outputs - would be interested to hear experiences with this approach. As for the GP - basically, this boils down to emulation of the posterior, which you could also consider for other problems. I guess the problem is that you are never sure about the quality of the approximation. What I also wonder is if an MCMC sample is ideal for a GP model, or whether you should invest more in the tails. | |
| Apr 28, 2016 at 23:14 | history | tweeted | twitter.com/StackStats/status/725825535910285313 | ||
| Apr 28, 2016 at 19:48 | history | edited | lacerbi | CC BY-SA 3.0 |
added 144 characters in body
|
| Apr 28, 2016 at 17:12 | history | edited | lacerbi | CC BY-SA 3.0 |
deleted 20 characters in body
|
| Apr 28, 2016 at 14:12 | history | edited | lacerbi | CC BY-SA 3.0 |
added 76 characters in body
|
| Apr 28, 2016 at 14:02 | history | edited | lacerbi | CC BY-SA 3.0 |
added 2 characters in body
|
| Apr 28, 2016 at 13:57 | history | asked | lacerbi | CC BY-SA 3.0 |