Comrade jo

♠Tom@sBat 21:56, 10 May 2007 (UTC)Reply

Earlier on Talk:Model selection you posted a link that claimed AIC was not appropriate for bayesian models. I didn't understand it, but I was wondering if you have since come across more documentation on why AIC is not appropriate? EverGreg (talk) 07:40, 30 April 2008 (UTC)Reply

Right - since then I've found out a bit more about it (didn't update, bad me..)
Once again, I must emphasise that I'm not a mathematician and this is based on my limited understanding of the literature as written by maths gurus for phylogeneticists.. Also, I should have clarified further by explaining that by 'Bayesian models' I meant 'Bayesian models of nucleic acid evolution (phylogenetics) as implemented in a Markov-chain Monte Carlo (MCMC) search strategy.
Basically it seems to stem from the different estimation goals of the two methods; while AIC is applied to ML methods that try to find a single best solution so the trade-off between likelihood and parametization is a simple one (the AIC curve is a power-2 one, right?), Bayesian MCMC methods instead estimate the posterior distribution of best models (model parametizations) and so the 'best' model in terms of the maximum likelihood solution may not represent the whole posterior distribution. In fact many posterior distributions are not normal so that intuitively makes sense. In model selection of Bayesian MCMC methods for phylogenetics (which is still a young field) it has become common to use the ratio of harmonic mean posterior likelihoods ('Bayes factor') for model selection.

More references on this:
Suchard et al. (2001) Bayesian selection of continuous-time Markov chain evolutionary models. MBE 18(6):1001-1013 PMID 11371589
Suchard et al. (2005) Models for Estimating Bayes Factors with Applications to Phylogeny and Tests of Monophyly. Biometrics 61(3):665-673 PMID 16135017
--Comrade jo (talk) 11:28, 17 June 2008 (UTC)Reply

Ah I see. The AIC value ranks models relative to each other so in principle you can derive results like "out of these five models, number 2 and 4 are the best, while 1,3 and 5 are far worse" and not just "number 2 is the best". But the literature is very insistent on the AIC value only being a relative ranking and no indication whatsoever on how good a given model is in itself. I guess that might make for difficult inputs to a probabilistic-centric system.

mmm... I mean, you can always evaluate the likelihood of any proposed model as a measure of fit to the data and this is an absolute measure in the sense that it never changes, and that you can compare the likelihood of different data under the same model (in phylogenetics, this is basically what you do in a bootstrap approach to assign support to topologies create a null distribution of a phylogenetic measure)

but comparing models is always going to be a comparative process - - i think ?

Comrade jo (talk) 15:08, 30 August 2008 (UTC)Reply

An additional problem (and here I'm doing a wild guess) could be using AIC as a measure of goodness when searching in a fitness landscape. AIC is an estimate of the Kullback–Leibler divergence which is not a true distance metric. This may or may not mess up some search algorithms. EverGreg (talk) 10:30, 19 June 2008 (UTC)Reply

You made a shrewd point about the idiom's use in The Cruel Sea. I feel, though, that describing the novel as 'classic' breaches WP neutrality rules. Would you consider deleting it?

I'd also make a second point about contemporary use. The phrase appears in lyrics from a couple of contemporary bands, which suggests either that it has made a comeback or else that it was old fashioned only in the opinion of Nicholas Montsarrat! Mzilikazi1939 (talk) 22:47, 20 April 2011 (UTC)Reply

That's a very good point - glad to see it's been edited already! Joe Comrade jo (talk) 11:40, 1 June 2011 (UTC)Reply

Hi, this message is to let you know about disambiguation links you've recently created. A link to a disambiguation page is almost always unintended, since a disambiguation page is merely a list of "Did you mean..." article titles. For more information, see the FAQ or drop a line at the DPL WikiProject.

Ben Griffin (British Army soldier) (check to confirm | fix with Dab solver)

was linked to Parachute Regiment, Afghanistan War, General Service Medal, Iraq Medal

Any suggestions for improving this automated tool are welcome. Thanks, DPL bot (talk) 18:38, 14 November 2011 (UTC)Reply

The article Dulwich Hamlet F.C. Ladies has been proposed for deletion because of the following concern:

Club fails to meet any notability guidelines.

While all constructive contributions to Wikipedia are appreciated, pages may be deleted for any of several reasons.

You may prevent the proposed deletion by removing the {{proposed deletion/dated}} notice, but please explain why in your edit summary or on the article's talk page.

Please consider improving the page to address the issues raised. Removing {{proposed deletion/dated}} will stop the proposed deletion process, but other deletion processes exist. In particular, the speedy deletion process can result in deletion without discussion, and articles for deletion allows discussion to reach consensus for deletion.

This bot DID NOT nominate any of your contributions for deletion; please refer to the history of each individual page for details. Thanks, FastilyBot (talk) 10:00, 12 October 2023 (UTC)Reply