## new estimators of evidence

Posted in Books, Statistics with tags , , , , , , , , , , , , on June 19, 2018 by xi'an

In an incredible accumulation of coincidences, I came across yet another paper about evidence and the harmonic mean challenge, by Yu-Bo Wang, Ming-Hui Chen [same as in Chen, Shao, Ibrahim], Lynn Kuo, and Paul O. Lewis this time, published in Bayesian Analysis. (Disclaimer: I was not involved in the reviews of any of these papers!)  Authors who arelocated in Storrs, Connecticut, in geographic and thematic connection with the original Gelfand and Dey (1994) paper! (Private joke about the Old Man of Storr in above picture!)

“The working parameter space is essentially the constrained support considered by Robert and Wraith (2009) and Marin and Robert (2010).”

The central idea is to use a more general function than our HPD restricted prior but still with a known integral. Not in the sense of control variates, though. The function of choice is a weighted sum of indicators of terms of a finite partition, which implies a compact parameter set Ω. Or a form of HPD region, although it is unclear when the volume can be derived. While the consistency of the estimator of the inverse normalising constant [based on an MCMC sample] is unsurprising, the more advanced part of the paper is about finding the optimal sequence of weights, as in control variates. But it is also unsurprising in that the weights are proportional to the inverses of the inverse posteriors over the sets in the partition. Since these are hard to derive in practice, the authors come up with a fairly interesting alternative, which is to take the value of the posterior at an arbitrary point of the relevant set.

The paper also contains an extension replacing the weights with functions that are integrable and with known integrals. Which is hard for most choices, even though it contains the regular harmonic mean estimator as a special case. And should also suffer from the curse of dimension when the constraint to keep the target almost constant is implemented (as in Figure 1).

The method, when properly calibrated, does much better than harmonic mean (not a surprise) and than Petris and Tardella (2007) alternative, but no other technique, on toy problems like Normal, Normal mixture, and probit regression with three covariates (no Pima Indians this time!). As an aside I find it hard to understand how the regular harmonic mean estimator takes longer than this more advanced version, which should require more calibration. But I find it hard to see a general application of the principle, because the partition needs to be chosen in terms of the target. Embedded balls cannot work for every possible problem, even with ex-post standardisation.

## another version of the corrected harmonic mean estimator

Posted in Books, pictures, Statistics, University life with tags , , , , , on June 11, 2018 by xi'an

A few days ago I came across a short paper in the Central European Journal of Economic Modelling and Econometrics by Pajor and Osiewalski that proposes a correction to the infamous harmonic mean estimator that is essentially the one Darren and I made in 2009, namely to restrict the evaluations of the likelihood function to a subset A of the simulations from the posterior. Paper that relates to an earlier 2009 paper by Peter Lenk, which investigates the same object with this same proposal and that we had missed for all that time. The difference is that, while we examine an arbitrary HPD region at level 50% or 80% as the subset A, Lenk proposes to derive a minimum likelihood value from the MCMC run and to use the associated HPD region, which means using all simulations, hence producing the same object as the original harmonic mean estimator, except that it is corrected by a multiplicative factor P(A). Or rather an approximation. This correction thus maintains the infinite variance of the original, a point apparently missed in the paper.

## Metropolis-Hastings importance sampling

Posted in Books, Statistics, University life with tags , , , , , , , , , on June 6, 2018 by xi'an

[Warning: As I first got the paper from the authors and sent them my comments, this paper read contains their reply as well.]

In a sort of crazy coincidence, Daniel Rudolf and Björn Sprungk arXived a paper on a Metropolis-Hastings importance sampling estimator that offers similarities with  the one by Ingmar Schuster and Ilja Klebanov posted on arXiv the same day. The major difference in the construction of the importance sampler is that Rudolf and Sprungk use the conditional distribution of the proposal in the denominator of their importance weight, while Schuster and Klebanov go for the marginal (or a Rao-Blackwell representation of the marginal), mostly in an independent Metropolis-Hastings setting (for convergence) and for a discretised Langevin version in the applications. The former use a very functional L² approach to convergence (which reminded me of the early Schervish and Carlin, 1990, paper on the convergence of MCMC algorithms), not all of it necessary in my opinion. As for instance the extension of convergence properties to the augmented chain, namely (current, proposed), is rather straightforward since the proposed chain is a random transform of the current chain. An interesting remark at the end of the proof of the CLT is that the asymptotic variance of the importance sampling estimator is the same as with iid realisations from the target. This is a point we also noticed when constructing population Monte Carlo techniques (more than ten years ago), namely that dependence on the past in sequential Monte Carlo does not impact the validation and the moments of the resulting estimators, simply because “everything cancels” in importance ratios. The mean square error bound on the Monte Carlo error (Theorem 20) is not very surprising as the term ρ(y)²/P(x,y) appears naturally in the variance of importance samplers.

The first illustration where the importance sampler does worse than the initial MCMC estimator for a wide range of acceptance probabilities (Figures 2 and 3, which is which?) and I do not understand the opposite conclusion from the authors.

Indeed the formulation in our paper is unfortunate. The point we want to stress is that we observed in the numerical experiments certain ranges of step-sizes for which MH importance sampling shows a better performance than the classical MH algorithm with optimal scaling. Meaning that the MH importance sampling with optimal step-size can outperform MH sampling, without using additional computational resources. Surprisingly, the optimal step-size for the MH importance sampling estimator seems to remain constant for an increasing dimension in contrast to the well-known optimal scaling of the MH algorithm (given by a constant optimal acceptance rate).

The second uses the Pima Indian diabetes benchmark, amusingly (?) referring to Chopin and Ridgway (2017) who warn against the recourse to this dataset and to this model! The loss in mean square error due to the importance sampling may again be massive (Figure 5) and setting for an optimisation of the scaling factor in Metropolis-Hastings algorithms sounds unrealistic.

Indeed, Chopin and Ridgway suggest more complex problems with a larger number of covariates as benchmarks. However, the well-studied PIMA data set is a sufficient example in order to illustrate the possible benefits but also the limitations of the MH importance sampling approach. The latter are clearly (a) the required knowledge about the optimal step-size—otherwise the performance can indeed be dramatically worse than for the MH algorithm—and (b) the restriction to a small or at most moderate number of covariates. As you are indicating, optimizing the scaling factor is a challenging task. However, the hope is to derive some simple rule of thumb for the MH importance sampler similar to the well-known acceptance rate tuning for the standard MCMC estimator.

## controlled sequential Monte Carlo [BiPS seminar]

Posted in Statistics with tags , , , , , , , on June 5, 2018 by xi'an

The last BiPS seminar of the semester will be given by Jeremy Heng (Harvard) on Monday 11 June at 2pm, in room 3001, ENSAE, Paris-Saclay about his Controlled sequential Monte Carlo paper:

Sequential Monte Carlo methods, also known as particle methods, are a popular set of techniques to approximate high-dimensional probability distributions and their normalizing constants. They have found numerous applications in statistics and related fields as they can be applied to perform state estimation for non-linear non-Gaussian state space models and Bayesian inference for complex static models. Like many Monte Carlo sampling schemes, they rely on proposal distributions which have a crucial impact on their performance. We introduce here a class of controlled sequential Monte Carlo algorithms, where the proposal distributions are determined by approximating the solution to an associated optimal control problem using an iterative scheme. We provide theoretical analysis of our proposed methodology and demonstrate significant gains over state-of-the-art methods at a fixed computational complexity on a variety of applications.

## best unbiased estimator of θ² for a Poisson model

Posted in Books, Kids, pictures, Statistics, Travel, University life with tags , , , , , , , , , , , , on May 23, 2018 by xi'an

A mostly traditional question on X validated about the “best” [minimum variance] unbiased estimator of θ² from a Poisson P(θ) sample leads to the Rao-Blackwell solution

$\mathbb{E}[X_1X_2|\underbrace{\sum_{i=1}^n X_i}_S=s] = -\frac{s}{n^2}+\frac{s^2}{n^2}=\frac{s(s-1)}{n^2}$

and a similar estimator could be constructed for θ³, θ⁴, … With the interesting limitation that this procedure stops at the power equal to the number of observations (minus one?). But,  since the expectation of a power of the sufficient statistics S [with distribution P(nθ)] is a polynomial in θ, there is de facto no limitation. More interestingly, there is no unbiased estimator of negative powers of θ in this context, while this neat comparison on Wikipedia (borrowed from the great book of counter-examples by Romano and Siegel, 1986, selling for a mere \$180 on amazon!) shows why looking for an unbiased estimator of exp(-2θ) is particularly foolish: the only solution is (-1) to the power S [for a single observation]. (There is however a first way to circumvent the difficulty if having access to an arbitrary number of generations from the Poisson, since the Forsythe – von Neuman algorithm allows for an unbiased estimation of exp(-F(x)). And, as a second way, as remarked by Juho Kokkala below, a sample of at least two Poisson observations leads to a more coherent best unbiased estimator.)

## independent random sampling methods [book review]

Posted in Books, Statistics, University life with tags , , , , , , , , , , , , , on May 16, 2018 by xi'an

Last week, I had the pleasant surprise to receive a copy of this book in the mail. Book that I was not aware had been written or published (meaning that I was not involved in its review!). The three authors, Luca Martino, David Luengo, and Joaquín Míguez, of Independent Random Sampling Methods are from Madrid universities and I have read (and posted on) several of their papers on (population) Monte Carlo simulation in the recent years. Including Luca’s survey of multiple try MCMC which was helpful in writing our WIREs own survey.

The book is a pedagogical coverage of most algorithms used to simulate independent samples from a given distribution, which of course recoups some of the techniques exposed with more details by [another] Luc, namely Luc Devroye’s Non-uniform random variate generation bible, often mentioned here (and studied in uttermost details by a dedicated reading group in Warwick). It includes a whole chapter on accept-reject methods, with in particular a section on Payne-Dagpunar’s band rejection I had not seen previously. And another entire chapter on ratio-of-uniforms techniques. On which the three authors had proposed generalisations [covered by the book], years before I attempted to go the same way, having completely forgotten reading their paper at the time… Or the much earlier 1991 paper by Jon Wakefield, Alan Gelfand and Adrian Smith!

The book also covers the “vertical density representation”, due to Troutt (1991), which consists in considering the distribution of the density p(.) of the random variable X as a random variable, p(X). I remember pondering about this alternative to the cdf transform and giving up on it as the outcome has a distribution depending on p, even when the density is monotonous. Even though I am not certain from reading the section that this is particularly appealing…

Given its title, the book contains very little about MCMC. Except for a last and final chapter that covers adaptive independent Metropolis-Hastings algorithms, in connection with some of the authors’ recent work. Like multiple try Metropolis. Relating to the (unidimensional) ARMS “ancestor” of adaptive MCMC methods. (As noted in a recent blog on Holden et al., 2009 , I have trouble understanding how recycling only rejected proposed values to build a better proposal distribution is enough to guarantee convergence of an adaptive algorithm, but the book does not delve much into this convergence.)

All in all and with the bias induced by me working in the very area, I find the book quite a nice entry on the topic, which can be used in a Monte Carlo course at both undergraduate and graduate levels if one want to avoid going into Markov chains. It is certainly less likely to scare students away than the comprehensive Non-uniform random variate generation and on the opposite may induce some of them to pursue a research career in this domain.

## interdependent Gibbs samplers

Posted in Books, Statistics, University life with tags , , , , , , on April 27, 2018 by xi'an

Mark Kozdoba and Shie Mannor just arXived a paper on an approach to accelerate a Gibbs sampler. With title “interdependent Gibbs samplers“. In fact, it presents rather strong similarities with our SAME algorithm. More of the same, as Adam Johanssen (Warwick) entitled one of his papers! The paper indeed suggests multiplying replicas of latent variables (e.g., an hidden path for an HMM) in an artificial model. And as in our 2002 paper, with Arnaud Doucet and Simon Godsill, the focus here is on maximum likelihood estimation (of the genuine parameters, not of the latent variables). Along with argument that the resulting pseudo-posterior is akin to a posterior with a powered likelihood. And a link with the EM algorithm. And an HMM application.

“The generative model consist of simply sampling the parameters ,  and then sampling m independent copies of the paths”

If anything this proposal is less appealing than SAME because it aims directly at the powered likelihood, rather than utilising an annealed sequence of powers that allows for a primary exploration of the whole parameter space before entering the trapping vicinity of a mode. Which makes me fail to catch the argument from the authors that this improves Gibbs sampling, as a more acute mode has on the opposite the dangerous feature of preventing visits to other modes. Hence the relevance to resort to some form of annealing.

As already mused upon in earlier posts, I find it most amazing that this technique has been re-discovered so many times, both in statistics and in adjacent fields. The idea of powering the likelihood with independent copies of the latent variables is obviously natural (since a version pops up every other year, always under a different name), but earlier versions should eventually saturate the market!