At MCMskv, Alexander Ly (from Amsterdam) pointed out to me some R programming mistakes I made in the introduction to Metropolis-Hastings algorithms I wrote a few months ago for the Wiley on-line encyclopedia! While the outcome (Monte Carlo posterior) of the corrected version is moderately changed this is nonetheless embarrassing! The example (if not the R code) was a mixture of a Poisson and a Geometric distributions borrowed from our testing as mixture paper. Among other things, I used a flat prior on the mixture weights instead of a Beta(1/2,1/2) prior and a simple log-normal random walk on the mean parameter instead of a more elaborate second order expansion discussed in the text. And I also inverted the probabilities of success and failure for the Geometric density. The new version is now available on arXiv, and hopefully soon on the Wiley site, but one (the?) fact worth mentioning here is that the (right) corrections in the R code first led to overflows, because I was using the Beta random walk Be(εp,ε(1-p)) which major drawback I discussed here a few months ago. With the drag that nearly zero or one values of the weight parameter produced infinite values of the density… Adding 1 (or 1/2) to each parameter of the Beta proposal solved the problem. And led to a posterior on the weight still concentrating on the correct corner of the unit interval. In any case, a big thank you to Alexander for testing the R code and spotting out the several mistakes…
Archive for mixtures
“From a Bayesian perspective, perhaps the most natural approach is to treat the numberof components like any other unknown parameter and put a prior on it.”
Another mixture paper on arXiv! Indeed, Jeffrey Miller and Matthew Harrison recently arXived a paper on estimating the number of components in a mixture model, comparing the parametric with the non-parametric Dirichlet prior approaches. Since priors can be chosen towards agreement between those. This is an obviously interesting issue, as they are often opposed in modelling debates. The above graph shows a crystal clear agreement between finite component mixture modelling and Dirichlet process modelling. The same happens for classification. However, Dirichlet process priors do not return an estimate of the number of components, which may be considered a drawback if one considers this is an identifiable quantity in a mixture model… But the paper stresses that the number of estimated clusters under the Dirichlet process modelling tends to be larger than the number of components in the finite case. Hence that the Dirichlet process mixture modelling is not consistent in that respect, producing parasite extra clusters…
In the parametric modelling, the authors assume the same scale is used in all Dirichlet priors, that is, for all values of k, the number of components. Which means an incoherence when marginalising from k to (k-p) components. Mild incoherence, in fact, as the parameters of the different models do not have to share the same priors. And, as shown by Proposition 3.3 in the paper, this does not prevent coherence in the marginal distribution of the latent variables. The authors also draw a comparison between the distribution of the partition in the finite mixture case and the Chinese restaurant process associated with the partition in the infinite case. A further analogy is that the finite case allows for a stick breaking representation. A noteworthy difference between both modellings is about the size of the partitions
in the finite (homogeneous partitions) and infinite (extreme partitions) cases.
An interesting entry into the connections between “regular” mixture modelling and Dirichlet mixture models. Maybe not ultimately surprising given the past studies by Peter Green and Sylvia Richardson of both approaches (1997 in Series B and 2001 in JASA).
When playing with Peter Rossi’s bayesm R package during a visit of Jean-Michel Marin to Paris, last week, we came up with the above Gibbs outcome. The setting is a Gaussian mixture model with three components in dimension 5 and the prior distributions are standard conjugate. In this case, with 500 observations and 5000 Gibbs iterations, the Markov chain (for one component of one mean of the mixture) has two highly distinct regimes: one that revolves around the true value of the parameter, 2.5, and one that explores a much broader area (which is associated with a much smaller value of the component weight). What we found amazing is the Gibbs ability to entertain both regimes, simultaneously.
Phil O’Neill and Theodore Kypraios from the University of Nottingham have arXived last week a paper on “Bayesian model choice via mixture distributions with application to epidemics and population process models”. Since we discussed this paper during my visit there earlier this year, I was definitely looking forward the completed version of their work. Especially because there are some superficial similarities with our most recent work on… Bayesian model choice via mixtures! (To the point that I misunderstood at the beginning their proposal for ours…)
The central idea in the paper is that, by considering the mixture likelihood
where x corresponds to the entire sample, it is straighforward to relate the moments of α with the Bayes factor, namely
which means that estimating the mixture weight α by MCMC is equivalent to estimating the Bayes factor.
What puzzled me at first was that the mixture weight is in fine estimated with a single “datapoint”, made of the entire sample. So the posterior distribution on α is hardly different from the prior, since it solely varies by one unit! But I came to realise that this is a numerical tool and that the estimator of α is not meaningful from a statistical viewpoint (thus differing completely from our perspective). This explains why the Beta prior on α can be freely chosen so that the mixing and stability of the Markov chain is improved: This parameter is solely an algorithmic entity.
There are similarities between this approach and the pseudo-prior encompassing perspective of Carlin and Chib (1995), even though the current version does not require pseudo-priors, using true priors instead. But thinking of weakly informative priors and of the MCMC consequence (see below) leads me to wonder if pseudo-priors would not help in this setting…
Another aspect of the paper that still puzzles me is that the MCMC algorithm mixes at all: indeed, depending on the value of the binary latent variable z, one of the two parameters is updated from the true posterior while the other is updated from the prior. It thus seems unlikely that the value of z would change quickly. Creating a huge imbalance in the prior can counteract this difference, but the same problem occurs once z has moved from 0 to 1 or from 1 to 0. It seems to me that resorting to a common parameter [if possible] and using as a proposal the model-based posteriors for both parameters is the only way out of this conundrum. (We do certainly insist on this common parametrisation in our approach as it is paramount to the use of improper priors.)
“In contrast, we consider the case where there is only one datum.”
The idea in the paper is therefore fully computational and relates to other linkage methods that create bridges between two models. It differs from our new notion of Bayesian testing in that we consider estimating the mixture between the two models in comparison, hence considering instead the mixture
which is another model altogether and does not recover the original Bayes factor (Bayes factor that we altogether dismiss in favour of the posterior median of α and its entire distribution).
Here are the slides of my talk today at the BAYSM’14 conference in Vienna. Mostly an overview of some of my papers on mixtures, with the most recent stuff…
Next week I am giving a talk at BAYSM in Vienna. BAYSM is the Bayesian Young Statisticians meeting so one may wonder why, but with Chris Holmes and Mike West, we got invited as more… erm… senior speakers! So I decided to give a definitely senior talk on a thread pursued throughout my career so far, namely mixtures. Plus it also relates to works of the other senior speakers. Here is the abstract for the talk:
Mixtures of distributions are fascinating objects for statisticians in that they both constitute a straightforward extension of standard distributions and offer a complex benchmark for evaluating statistical procedures, with a likelihood both computable in a linear time and enjoying an exponential number of local models (and sometimes infinite modes). This fruitful playground appeals in particular to Bayesians as it constitutes an easily understood challenge to the use of improper priors and of objective Bayes solutions. This talk will review some ancient and some more recent works of mine on mixtures of distributions, from the 1990 Gibbs sampler to the 2000 label switching and to later studies of Bayes factor approximations, nested sampling performances, improper priors, improved importance samplers, ABC, and a inverse perspective on the Bayesian approach to testing of hypotheses.
I am very grateful to the scientific committee for this invitation, as it will give me the opportunity to meet the new generation, learn from them and in addition discover Vienna where I have never been, despite several visits to Austria. Including its top, the Großglockner. I will also give a seminar in Linz the day before. In the Institut für Angewandte Statistik.
- Dan Simpson replied to my comments of last Tuesday about his PC construction;
- Arnaud Doucet precised some issues about his adaptive subsampling paper;
- Amandine Schreck clarified why I had missed some points in her Bayesian variable selection paper;
- Randal Douc defended the efficiency of using Carlin and Chib (1995) method for mixture simulation.
Thanks to them for taking the time to answer my musings…