Archive for Bayesian model choice

projection predictive input variable selection

Posted in Books, Statistics, University life with tags , , , , , , on November 2, 2015 by xi'an

aikiJuho Piironen and Aki Vehtari just arXived a paper on variable selection that relates to two projection papers we wrote in the 1990’s with Costas Goutis (who died near Seattle in a diving accident on July 1996) and Jérôme Dupuis… Except that they move to the functional space of Gaussian processes. The covariance function in a Gaussian process is indeed based on a distance between observations, which are themselves defined as a vector of inputs. Some of which matter and some of which do not matter in the kernel value. When rescaling the distance with “length-scales” for all variables, one could think that non-significant variates have very small scales and hence bypass the need for variable selection but this is not the case as those coefficients react poorly to non-linearities in the variates… The paper thus builds a projective structure from a reference model involving all input variables.

“…adding some irrelevant inputs is not disastrous if the model contains a sparsifying prior structure, and therefore, one can expect to lose less by using all the inputs than by trying to differentiate between the relevant and irrelevant ones and ignoring the uncertainty related to the left-out inputs.”

While I of course appreciate this avatar to our original idea (with some borrowing from McCulloch and Rossi, 1992), the paper reminds me of some of the discussions and doubts we had about the role of the reference or super model that “anchors” the projections, as there is no reason for that reference model to be a better one. It could be that an iterative process where the selected submodel becomes the reference for the next iteration could enjoy better performances. When I first presented this work in Cagliari, in the late 1990s, one comment was that the method had no theoretical guarantee like consistency. Which is correct if the minimum distance is not evolving (how quickly?!) with the sample size n. I also remember the difficulty Jérôme and I had in figuring out a manageable forward-backward exploration of the (huge) set of acceptable subsets of variables. Random walk exploration and RJMCMC are unlikely to solve this problem.

model selection and multiple testing

Posted in Books, pictures, Statistics, Travel, University life with tags , , , , , , , , , , , , , on October 23, 2015 by xi'an

Ritabrata Dutta, Malgorzata Bogdan and Jayanta Ghosh recently arXived a survey paper on model selection and multiple testing. Which provides a good opportunity to reflect upon traditional Bayesian approaches to model choice. And potential alternatives. On my way back from Madrid, where I got a bit distracted when flying over the South-West French coast, from Biarritz to Bordeaux. Spotting the lake of Hourtain, where I spent my military training month, 29 years ago!

“On the basis of comparison of AIC and BIC, we suggest tentatively that model selection rules should be used for the purpose for which they were introduced. If they are used for other problems, a fresh justification is desirable. In one case, justification may take the form of a consistency theorem, in the other some sort of oracle inequality. Both may be hard to prove. Then one should have substantial numerical assessment over many different examples.”

The authors quickly replace the Bayes factor with BIC, because it is typically consistent. In the comparison between AIC and BIC they mention the connundrum of defining a prior on a nested model from the prior on the nesting model, a problem that has not been properly solved in my opinion. The above quote with its call to a large simulation study reminded me of the paper by Arnold & Loeppky about running such studies through ecdfs. That I did not see as solving the issue. The authors also discuss DIC and Lasso, without making much of a connection between those, or with the above. And then reach the parametric empirical Bayes approach to model selection exemplified by Ed George’s and Don Foster’s 2000 paper. Which achieves asymptotic optimality for posterior prediction loss (p.9). And which unifies a wide range of model selection approaches.

A second part of the survey considers the large p setting, where BIC is not a good approximation to the Bayes factor (when testing whether or not all mean entries are zero). And recalls that there are priors ensuring consistency for the Bayes factor in this very [restrictive] case. Then, in Section 4, the authors move to what they call “cross-validatory Bayes factors”, also known as partial Bayes factors and pseudo-Bayes factors, where the data is split to (a) make the improper prior proper and (b) run the comparison or test on the remaining data. They also show the surprising result that, provided the fraction of the data used to proper-ise the prior does not converge to one, the X validated Bayes factor remains consistent [for the special case above]. The last part of the paper concentrates on multiple testing but is more tentative and conjecturing about convergence results, centring on the differences between full Bayes and empirical Bayes. Then the plane landed in Paris and I stopped my reading, not feeling differently about the topic than when the plane started from Madrid.

seminar im München, am Max-Planck-Institut für Astrophysik

Posted in Statistics, Travel, University life with tags , , , , , , , , , , , , on October 15, 2015 by xi'an

On Friday, I give a talk in München on ABC model choice. At the Max-Planck Institute for Astrophysics. As coincidence go, I happen to talk the week after John Skilling gave a seminar there. On Bayesian tomography, not on nested sampling. And the conference organisers put the cover of the book Think Bayes: Bayesian Statistics Made Simple, written by Allen Downey, a book I reviewed yesterday night for CHANCE (soon to appear on the ‘Og!) [not that I understand the connection with the Max-Planck Institute or with my talk!, warum nicht?!] The slides are the same as in Oxford for SPA 2015:

ABC model choice via random forests [and no fire]

Posted in Books, pictures, R, Statistics, University life with tags , , , , , , , , , on September 4, 2015 by xi'an

While my arXiv newspage today had a puzzling entry about modelling UFOs sightings in France, it also broadcast our revision of Reliable ABC model choice via random forests, version that we resubmitted today to Bioinformatics after a quite thorough upgrade, the most dramatic one being the realisation we could also approximate the posterior probability of the selected model via another random forest. (With no connection with the recent post on forest fires!) As discussed a little while ago on the ‘Og. And also in conjunction with our creating the abcrf R package for running ABC model choice out of a reference table. While it has been an excruciatingly slow process (the initial version of the arXived document dates from June 2014, the PNAS submission was rejected for not being enough Bayesian, and the latest revision took the whole summer), the slow maturation of our thoughts on the model choice issues led us to modify the role of random forests in the ABC approach to model choice, in that we reverted our earlier assessment that they could only be trusted for selecting the most likely model, by realising this summer the corresponding posterior could be expressed as a posterior loss and estimated by a secondary forest. As first considered in Stoehr et al. (2014). (In retrospect, this brings an answer to one of the earlier referee’s comments.) Next goal is to incorporate those changes in DIYABC (and wait for the next version of the software to appear). Another best-selling innovation due to Arnaud: we added a practical implementation section in the format of FAQ for issues related with the calibration of the algorithms.

inflation, evidence and falsifiability

Posted in Books, pictures, Statistics, University life with tags , , , , , , , , , , , , on July 27, 2015 by xi'an

[Ewan Cameron pointed this paper to me and blogged about his impressions a few weeks ago. And then Peter Coles wrote a (properly) critical blog entry yesterday. Here are my quick impressions, as an add-on.]

“As the cosmological data continues to improve with its inevitable twists, it has become evident that whatever the observations turn out to be they will be lauded as \proof of inflation”.” G. Gubitosi et al.

In an arXive with the above title, Gubitosi et al. embark upon a generic and critical [and astrostatistical] evaluation of Bayesian evidence and the Bayesian paradigm. Perfect topic and material for another blog post!

“Part of the problem stems from the widespread use of the concept of Bayesian evidence and the Bayes factor (…) The limitations of the existing formalism emerge, however, as soon as we insist on falsifiability as a pre-requisite for a scientific theory (….) the concept is more suited to playing the lottery than to enforcing falsifiability: winning is more important than being predictive.” G. Gubitosi et al.

It is somehow quite hard not to quote most of the paper, because prose such as the above abounds. Now, compared with standards, the authors introduce an higher level than models, called paradigms, as collections of models. (I wonder what is the next level, monads? universes? paradises?) Each paradigm is associated with a marginal likelihood, obtained by integrating over models and model parameters. Which is also the evidence of or for the paradigm. And then, assuming a prior on the paradigms, one can compute the posterior over the paradigms… What is the novelty, then, that “forces” falsifiability upon Bayesian testing (or the reverse)?!

“However, science is not about playing the lottery and winning, but falsifiability instead, that is, about winning given that you have bore the full brunt of potential loss, by taking full chances of not winning a priori. This is not well incorporated into the Bayesian evidence because the framework is designed for other ends, those of model selection rather than paradigm evaluation.” G. Gubitosi et al.

The paper starts by a criticism of the Bayes factor in the point null test of a Gaussian mean, as overly penalising the null against the alternative being only a power law. Not much new there, it is well known that the Bayes factor does not converge at the same speed under the null and under the alternative… The first proposal of those authors is to consider the distribution of the marginal likelihood of the null model under the [or a] prior predictive encompassing both hypotheses or only the alternative [there is a lack of precision at this stage of the paper], in order to calibrate the observed value against the expected. What is the connection with falsifiability? The notion that, under the prior predictive, most of the mass is on very low values of the evidence, leading to concluding against the null. If replacing the null with the alternative marginal likelihood, its mass then becomes concentrated on the largest values of the evidence, which is translated as an unfalsifiable theory. In simpler terms, it means you can never prove a mean θ is different from zero. Not a tremendously item of news, all things considered…

“…we can measure the predictivity of a model (or paradigm) by examining the distribution of the Bayesian evidence assuming uniformly distributed data.” G. Gubitosi et al.

The alternative is to define a tail probability for the evidence, i.e. the probability to be below an arbitrarily set bound. What remains unclear to me in this notion is the definition of a prior on the data, as it seems to be model dependent, hence prohibits comparison between models since this would involve incompatible priors. The paper goes further into that direction by penalising models according to their predictability, P, as exp{-(1-P²)/P²}. And paradigms as well.

“(…) theoretical matters may end up being far more relevant than any probabilistic issues, of whatever nature. The fact that inflation is not an unavoidable part of any quantum gravity framework may prove to be its greatest undoing.” G. Gubitosi et al.

Establishing a principled way to weight models would certainly be a major step in the validation of posterior probabilities as a quantitative tool for Bayesian inference, as hinted at in my 1993 paper on the Lindley-Jeffreys paradox, but I do not see such a principle emerging from the paper. Not only because of the arbitrariness in constructing both the predictivity and the associated prior weight, but also because of the impossibility to define a joint predictive, that is a predictive across models, without including the weights of those models. This makes the prior probabilities appearing on “both sides” of the defining equation… (And I will not mention the issues of constructing a prior distribution of a Bayes factor that are related to Aitkin‘s integrated likelihood. And won’t obviously try to enter the cosmological debate about inflation.)

Bureau international des poids et mesures

Posted in Books, Statistics, University life with tags , , , , , , , , , , on June 15, 2015 by xi'an

Today, I am taking part in a meeting in Paris, for an exotic change!, at the Bureau international des poids et mesures (BIPM), which looks after a universal reference for measurements. For instance, here is its definition of the kilogram:

The unit of mass, the kilogram, is the mass of the international prototype of the kilogram kept in air under three bell jars at the BIPM. It is a cylinder made of an alloy for which the mass fraction of platinum is 90 % and the mass fraction of iridium is 10 %.

And the BIPM is thus interested in the uncertainty associated with such measurements. Hence the workshop on measurement uncertainties. Tony O’Hagan will also be giving a talk in a session that opposes frequentist and Bayesian approaches, even though I decided to introduce ABC as it seems to me to be a natural notion for measurement problems (as far as I can tell from my prior on measurement problems).

speed seminar-ing

Posted in Books, pictures, Statistics, Travel, University life, Wines with tags , , , , , , , , , , on May 20, 2015 by xi'an

harbour in the morning, Carnon, June 15, 2012Yesterday, I  made a quick afternoon trip to Montpellier as replacement of a seminar speaker who had cancelled at the last minute. Most obviously, I gave a talk about our “testing as mixture” proposal. And as previously, the talk generated a fair amount of discussion and feedback from the audience. Providing me with additional aspects to include in a revision of the paper. Whether or not the current submission is rejected, new points made and received during those seminars will have to get in a revised version as they definitely add to the appeal to the perspective. In that seminar, most of the discussion concentrated on the connection with decisions based on such a tool as the posterior distribution of the mixture weight(s). My argument for sticking with the posterior rather than providing a hard decision rule was that the message is indeed in arguing hard rules that end up mimicking the p- or b-values. And the catastrophic consequences of fishing for significance and the like. Producing instead a validation by simulating under each model pseudo-samples shows what to expect for each model under comparison. The argument did not really convince Jean-Michel Marin, I am afraid! Another point he raised was that we could instead use a distribution on α with support {0,1}, to avoid the encompassing model he felt was too far from the original models. However, this leads back to the Bayes factor as the weights in 0 and 1 are the marginal likelihoods, nothing more. However, this perspective on the classical approach has at least the appeal of completely validating the use of improper priors on common (nuisance or not) parameters. Pierre Pudlo also wondered why we could not conduct an analysis on the mixture of the likelihoods. Instead of the likelihood of the mixture. My first answer was that there was not enough information in the data for estimating the weight(s). A few more seconds of reflection led me to the further argument that the posterior on α with support (0,1) would then be a mixture of Be(2,1) and Be(1,2) with weights the marginal likelihoods, again (under a uniform prior on α). So indeed not much to gain. A last point we discussed was the case of the evolution trees we analyse with population geneticists from the neighbourhood (and with ABC). Jean-Michel’s argument was that the scenari under comparison were not compatible with a mixture, the models being exclusive. My reply involved an admixture model that contained all scenarios as special cases. After a longer pondering, I think his objection was more about the non iid nature of the data. But the admixture construction remains valid. And makes a very strong case in favour of our approach, I believe.

masbruguiere2After the seminar, Christian Lavergne and Jean-Michel had organised a doubly exceptional wine-and-cheese party: first because it is not usually the case there is such a post-seminar party and second because they had chosen a terrific series of wines from the Mas Bruguière (Pic Saint-Loup) vineyards. Ending up with a great 2007 L’Arbouse. Perfect ending for an exciting day. (I am not even mentioning a special Livarot from close to my home-town!)


Get every new post delivered to your Inbox.

Join 944 other followers