Timeline for LASSO and ridge from the Bayesian perspective: what about the tuning parameter?
Current License: CC BY-SA 4.0
23 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| S Dec 13, 2018 at 20:49 | history | bounty ended | CommunityBot | ||
| S Dec 13, 2018 at 20:49 | history | notice removed | user227843 | ||
| Dec 12, 2018 at 12:19 | vote | accept | Richard Hardy | ||
| Dec 9, 2018 at 0:35 | comment | added | amoeba | @statslearner2 I think it does address Richard's question very well. Your bounty seems to be focused on a more narrow aspect (about a hyperprior) than Richard's Q. | |
| Dec 8, 2018 at 21:44 | comment | added | user227843 | @amoeba I've read them, but they do not address this question. | |
| Dec 7, 2018 at 11:00 | comment | added | Sextus Empiricus | @statslearner2 related andrewgelman.com/2004/11/08/crossvalidation stats.stackexchange.com/questions/343420/… | |
| Dec 7, 2018 at 8:49 | comment | added | amoeba | @statslearner2 Did you see the link I gave in the 1st comment above? This might be useful for you. | |
| Dec 7, 2018 at 7:08 | review | Suggested edits | |||
| Dec 7, 2018 at 14:21 | |||||
| Dec 7, 2018 at 6:49 | comment | added | user227843 | PS: to those aiming for the bounty, note my comment: I want to see an explicit answer that shows a prior that induces a MAP estimate equivalent to frequentist cross-validation. | |
| Dec 7, 2018 at 6:05 | answer | added | Ben | timeline score: 24 | |
| Dec 7, 2018 at 5:46 | comment | added | user227843 | @guy can you explain better the connection between a hyper-prior and k-fold CV? Is there a prior that would induce a similar behavior? | |
| Dec 7, 2018 at 5:41 | review | Suggested edits | |||
| Dec 7, 2018 at 6:46 | |||||
| S Dec 7, 2018 at 5:39 | history | bounty started | CommunityBot | ||
| S Dec 7, 2018 at 5:39 | history | notice added | user227843 | Canonical answer required | |
| Sep 21, 2018 at 21:01 | history | tweeted | twitter.com/StackStats/status/1043243874003677184 | ||
| Sep 21, 2018 at 14:35 | history | edited | Richard Hardy | CC BY-SA 4.0 |
deleted 1 character in body
|
| Sep 21, 2018 at 14:34 | comment | added | Richard Hardy | @kjetilbhalvorsen, great question. Not sure if it should be appended here or posted separately, though. | |
| Sep 21, 2018 at 14:31 | comment | added | Richard Hardy | @amoeba, thank you, this is roughly what I expected. The link to the other thread was helpful, too. | |
| Sep 21, 2018 at 13:00 | answer | added | Dimitris Rizopoulos | timeline score: 6 | |
| Sep 21, 2018 at 12:49 | comment | added | guy | Bayesians can put a prior on the tuning parameter, as it usually corresponds to a variance parameter. This is usually what is done to avoid CV in order to stay fully-Bayes. Alternatively, you can use REML to optimize the regularization parameter. | |
| Sep 21, 2018 at 12:21 | comment | added | kjetil b halvorsen♦ | Additional question (could be part of main Q): Do there exist some prior on the regularization parameter that somehow replaces the cross-validation process, somehow? | |
| Sep 21, 2018 at 12:10 | comment | added | amoeba | I imagine that a fully Bayesian approach would start with a given prior and not modify it, yes. But there is also an empirical-bayes approach that optimizes over hyperparameter values: e.g. see stats.stackexchange.com/questions/24799. | |
| Sep 21, 2018 at 12:05 | history | asked | Richard Hardy | CC BY-SA 4.0 |