## a neat (theoretical) Monte Carlo result

Posted in Books, Statistics, University life with tags , , , , on December 19, 2014 by xi'an

Mark Huber just arXived a short paper where he develops a Monte Carlo approach that bounds the probability of large errors

$\mathbb{P}(|\hat\mu_t-\mu|>\epsilon\mu) < 1/\delta$

by computing a lower bound on the sample size r and I wondered at the presence of μ in the bound as it indicates the approach is not translation invariant. One reason is that the standard deviation of the simulated random variables is bounded by cμ. Another reason is that Mark uses as its estimator the median

$\text{med}(S_1R_1,\ldots,S_tR_t)$

where the S’s are partial averages of sufficient length and the R’s are independent uniforms over (1-ε,1+ε): using those uniforms may improve the coverage of given intervals but it also means that the absolute scale of the error is multiplied by the scale of S, namely μ. I first thought that some a posteriori recentering could improve the bound but since this does not impact the variance of the simulated random variables, I doubt it is possible.

## density normalization for MCMC algorithms

Posted in Statistics, University life with tags , , , , , , , , on November 6, 2014 by xi'an

Another paper addressing the estimation of the normalising constant and the wealth of available solutions just came out on arXiv, with the full title of “Target density normalization for Markov chain Monte Carlo algorithms“, written by Allen Caldwell and Chang Liu. (I became aware of it by courtesy of Ewan Cameron, as it appeared in the physics section of arXiv. It is actually a wee bit annoying that papers in the subcategory “Data Analysis, Statistics and Probability” of physics do not get an automated reposting on the statistics lists…)

In this paper, the authors compare three approaches to the problem of finding

$\mathfrak{I} = \int_\Omega f(\lambda)\,\text{d}\lambda$

when the density f is unormalised, i.e., in more formal terms, when f is proportional to a probability density (and available):

1. an “arithmetic mean”, which is an importance sampler based on (a) reducing the integration volume to a neighbourhood ω of the global mode. This neighbourhood is chosen as an hypercube and the importance function turns out to be the uniform over this hypercube. The corresponding estimator is then a rescaled version of the average of f over uniform simulations in ω.
2.  an “harmonic mean”, of all choices!, with again an integration over the neighbourhood ω of the global mode in order to avoid the almost sure infinite variance of harmonic mean estimators.
3. a Laplace approximation, using the target at the mode and the Hessian at the mode as well.

The paper then goes to comparing those three solutions on a few examples, demonstrating how the diameter of the hypercube can be calibrated towards a minimum (estimated) uncertainty. The rather anticlimactic conclusion is that the arithmetic mean is the most reliable solution as harmonic means may fail in larger dimension and more importantly fail to signal its failure, while Laplace approximations only approximate well quasi-Gaussian densities…

What I find most interesting in this paper is the idea of using only one part of the integration space to compute the integral, even though it is not exactly new. Focussing on a specific region ω has pros and cons, the pros being that the reduction to a modal region reduces needs for absolute MCMC convergence and helps in selecting alternative proposals and also prevents from the worst consequences of using a dreaded harmonic mean, the cons being that the region needs be well-identified, which means requirements on the MCMC kernel, and that the estimate is a product of two estimates, the frequency being driven by a Binomial noise.  I also like very much the idea of calibrating the diameter Δof the hypercube ex-post by estimating the uncertainty.

As an aside, the paper mentions most of the alternative solutions I just presented in my Monte Carlo graduate course two days ago (like nested or bridge or Rao-Blackwellised sampling, including our proposal with Darren Wraith), but dismisses them as not “directly applicable in an MCMC setting”, i.e., without modifying this setting. I unsurprisingly dispute this labelling, both because something like the Laplace approximation requires extra-work on the MCMC output (and once done this work can lead to advanced Laplace methods like INLA) and because other methods could be considered as well (for instance, bridge sampling over several hypercubes). As shown in the recent paper by Mathieu Gerber and Nicolas Chopin (soon to be discussed at the RSS!), MCqMC has also become a feasible alternative that would compete well with the methods studied in this paper.

Overall, this is a paper that comes in a long list of papers on constant approximations. I do not find the Markov chain of MCMC aspect particularly compelling or specific, once the effective sample size is accounted for. It would be nice to find generic ways of optimising the visit to the hypercube ω and to estimate efficiently the weight of ω. The comparison is solely run over examples, but they all rely on a proper characterisation of the hypercube and the ability to simulate efficiently f over that hypercube.

## my life as a mixture [BAYSM 2014, Wien]

Posted in Books, Kids, Mountains, pictures, Statistics, Travel, University life with tags , , , , , , , , , , on September 12, 2014 by xi'an

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.

## PMC for combinatoric spaces

Posted in Statistics, University life with tags , , , , , , , on July 28, 2014 by xi'an

I received this interesting [edited] email from Xiannian Fan at CUNY:

I am trying to use PMC to solve Bayesian network structure learning problem (which is in a combinatorial space, not continuous space).

In PMC, the proposal distributions qi,t can be very flexible, even specific to each iteration and each instance. My problem occurs due to the combinatorial space.

For importance sampling, the requirement for proposal distribution, q, is:

support (p) ⊂ support (q)             (*)

For PMC, what is the support of the proposal distribution in iteration t? is it

support (p) ⊂ U support(qi,t)    (**)

or does (*) apply to every qi,t?

For continuous problem, this is not a big issue. We can use random walk of Normal distribution to do local move satisfying (*). But for combination search, local moving only result in finite states choice, just not satisfying (*). For example for a permutation (1,3,2,4), random swap has only choose(4,2)=6 neighbor states.

Fairly interesting question about population Monte Carlo (PMC), a sequential version of importance sampling we worked on with French colleagues in the early 2000’s.  (The name population Monte Carlo comes from Iba, 2000.)  While MCMC samplers do not have to cover the whole support of p at each iteration, it is much harder for importance samplers as their core justification is to provide an unbiased estimator to for all integrals of interest. Thus, when using the PMC estimate,

1/n ∑i,t {p(xi,t)/qi,t(xi,t)}h(qi,t),  xi,t~qi,t(x)

this estimator is only unbiased when the supports of the qi,t “s are all containing the support of p. The only other cases I can think of are

1. associating the qi,t “s with a partition Si,t of the support of p and using instead

i,t {p(xi,t)/qi,t(xi,t)}h(qi,t), xi,t~qi,t(x)

2. resorting to AMIS under the assumption (**) and using instead

1/n ∑i,t {p(xi,t)/∑j,t qj,t(xi,t)}h(qi,t), xi,t~qi,t(x)

but I am open to further suggestions!

## another R new trick [new for me!]

Posted in Books, Kids, R, Statistics, University life with tags , , , , , , , on July 16, 2014 by xi'an

While working with Andrew and a student from Dauphine on importance sampling, we wanted to assess the distribution of the resulting sample via the Kolmogorov-Smirnov measure

$\max_x |\hat{F_n}(x)-F(x)|$

where F is the target.  This distance (times √n) has an asymptotic distribution that does not depend on n, called the Kolmogorov distribution. After searching for a little while, we could not figure where this distribution was available in R. It had to, since ks.test was returning a p-value. Hopefully correct! So I looked into the ks.test function, which happens not to be entirely programmed in C, and found the line

PVAL <- 1 - if (alternative == "two.sided")
.Call(C_pKolmogorov2x, STATISTIC, n)


which means that the Kolmogorov distribution is coded as a C function C_pKolmogorov2x in R. However, I could not call the function myself.

> .Call(C_pKolmogorov2x,.3,4)


Hence, as I did not want to recode this distribution cdf, I posted the question on stackoverflow (long time no see!) and got a reply almost immediately as to use the package kolmim. Followed by the extra comment from the same person that calling the C code only required to add the path to its name, as in

> .Call(stats:::C_pKolmogorov2x,STAT=.3,n=4)
[1] 0.2292


## R/Rmetrics in Paris [alas!]

Posted in Mountains, pictures, R, Statistics, Travel, University life with tags , , , , , , , , , , , , , , , , , , , on June 30, 2014 by xi'an

Today I gave a talk on Bayesian model choice in a fabulous 13th Century former monastery in the Latin Quarter of Paris… It is the Collège des Bernardins, close to Jussieu and Collège de France, unbelievably hidden to the point I was not aware of its existence despite having studied and worked in Jussieu since 1982… I mixed my earlier San Antonio survey on importance sampling approximations to Bayes factors with an entry to our most recent work on ABC with random forests. This was the first talk of the 8th R/Rmetrics workshop taking place in Paris this year. (Rmetrics is aiming at aggregating R packages with econometrics and finance applications.) And I had a full hour and a half to deliver my lecture to the workshop audience. Nice place, nice people, new faces and topics (and even andouille de Vire for lunch!): why should I complain with an alas in the title?!What happened is that the R/Rmetrics meetings have been till this year organised in Meielisalp, Switzerland. Which stands on top of Thuner See and… just next to the most famous peaks of the Bernese Alps! And that I had been invited last year but could not make it… Meaning I lost a genuine opportunity to climb one of my five dream routes, the Mittelegi ridge of the Eiger. As the future R/Rmetrics meetings will not take place there.

A lunch discussion at the workshop led me to experiment the compiler library in R, library that I was unaware of. The impact on the running time is obvious: recycling the fowler function from the last Le Monde puzzle,

> bowler=cmpfun(fowler)
> N=20;n=10;system.time(fowler(pred=N))
user  system elapsed
52.647   0.076  56.332
> N=20;n=10;system.time(bowler(pred=N))
user  system elapsed
51.631   0.004  51.768
> N=20;n=15;system.time(bowler(pred=N))
user  system elapsed
51.924   0.024  52.429
> N=20;n=15;system.time(fowler(pred=N))
user  system elapsed
52.919   0.200  61.960


shows a ten- to twenty-fold gain in system time, if not in elapsed time (re-alas!).

## checking for finite variance of importance samplers

Posted in R, Statistics, Travel, University life with tags , , , , , , , , on June 11, 2014 by xi'an

Over a welcomed curry yesterday night in Edinburgh I read this 2008 paper by Koopman, Shephard and Creal, testing the assumptions behind importance sampling, which purpose is to check on-line for (in)finite variance in an importance sampler, based on the empirical distribution of the importance weights. To this goal, the authors use the upper tail  of the weights and a limit theorem that provides the limiting distribution as a type of Pareto distribution

$\dfrac{1}{\beta}\left(1+\xi z/\beta \right)^{-1-1/\xi}$

over (0,∞). And then implement a series of asymptotic tests like the likelihood ratio, Wald and score tests to assess whether or not the power ξ of the Pareto distribution is below ½. While there is nothing wrong with this approach, which produces a statistically validated diagnosis, I still wonder at the added value from a practical perspective, as raw graphs of the estimation sequence itself should exhibit similar jumps and a similar lack of stabilisation as the ones seen in the various figures of the paper. Alternatively, a few repeated calls to the importance sampler should disclose the poor convergence properties of the sampler, as in the above graph. Where the blue line indicates the true value of the integral.