## Leave the Pima Indians alone!

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

“…our findings shall lead to us be critical of certain current practices. Specifically, most papers seem content with comparing some new algorithm with Gibbs sampling, on a few small datasets, such as the well-known Pima Indians diabetes dataset (8 covariates). But we shall see that, for such datasets, approaches that are even more basic than Gibbs sampling are actually hard to beat. In other words, datasets considered in the literature may be too toy-like to be used as a relevant benchmark. On the other hand, if ones considers larger datasets (with say 100 covariates), then not so many approaches seem to remain competitive” (p.1)

Nicolas Chopin and James Ridgway (CREST, Paris) completed and arXived a paper they had “threatened” to publish for a while now, namely why using the Pima Indian R logistic or probit regression benchmark for checking a computational algorithm is not such a great idea! Given that I am definitely guilty of such a sin (in papers not reported in the survey), I was quite eager to read the reasons why! Beyond the debate on the worth of such a benchmark, the paper considers a wider perspective as to how Bayesian computation algorithms should be compared, including the murky waters of CPU time versus designer or programmer time. Which plays against most MCMC sampler.

As a first entry, Nicolas and James point out that the MAP can be derived by standard a Newton-Raphson algorithm when the prior is Gaussian, and even when the prior is Cauchy as it seems most datasets allow for Newton-Raphson convergence. As well as the Hessian. We actually took advantage of this property in our comparison of evidence approximations published in the Festschrift for Jim Berger. Where we also noticed the awesome performances of an importance sampler based on the Gaussian or Laplace approximation. The authors call this proposal their gold standard. Because they also find it hard to beat. They also pursue this approximation to its logical (?) end by proposing an evidence approximation based on the above and Chib’s formula. Two close approximations are provided by INLA for posterior marginals and by a Laplace-EM for a Cauchy prior. Unsurprisingly, the expectation-propagation (EP) approach is also implemented. What EP lacks in theoretical backup, it seems to recover in sheer precision (in the examples analysed in the paper). And unsurprisingly as well the paper includes a randomised quasi-Monte Carlo version of the Gaussian importance sampler. (The authors report that “the improvement brought by RQMC varies strongly across datasets” without elaborating for the reasons behind this variability. They also do not report the CPU time of the IS-QMC, maybe identical to the one for the regular importance sampling.) Maybe more surprising is the absence of a nested sampling version.

In the Markov chain Monte Carlo solutions, Nicolas and James compare Gibbs, Metropolis-Hastings, Hamiltonian Monte Carlo, and NUTS. Plus a tempering SMC, All of which are outperformed by importance sampling for small enough datasets. But get back to competing grounds for large enough ones, since importance sampling then fails.

“…let’s all refrain from now on from using datasets and models that are too simple to serve as a reasonable benchmark.” (p.25)

This is a very nice survey on the theme of binary data (more than on the comparison of algorithms in that the authors do not really take into account design and complexity, but resort to MSEs versus CPus). I however do not agree with their overall message to leave the Pima Indians alone. Or at least not for the reason provided therein, namely that faster and more accurate approximations methods are available and cannot be beaten. Benchmarks always have the limitation of “what you get is what you see”, i.e., the output associated with a single dataset that only has that many idiosyncrasies. Plus, the closeness to a perfect normal posterior makes the logistic posterior too regular to pause a real challenge (even though MCMC algorithms are as usual slower than iid sampling). But having faster and more precise resolutions should on the opposite be  cause for cheers, as this provides a reference value, a golden standard, to check against. In a sense, for every Monte Carlo method, there is a much better answer, namely the exact value of the integral or of the optimum! And one is hardly aiming at a more precise inference for the benchmark itself: those Pima Indians [whose actual name is Akimel O’odham] with diabetes involved in the original study are definitely beyond help from statisticians and the model is unlikely to carry out to current populations. When the goal is to compare methods, as in our 2009 paper for Jim Berger’s 60th birthday, what matters is relative speed and relative ease of implementation (besides the obvious convergence to the proper target). In that sense bigger and larger is not always relevant. Unless one tackles really big or really large datasets, for which there is neither benchmark method nor reference value.

## corrected MCMC samplers for multivariate probit models

Posted in Books, pictures, R, Statistics, University life with tags , , , , , , , , on May 6, 2015 by xi'an

“Moreover, IvD point out an error in Nobile’s derivation which can alter its stationary distribution. Ironically, as we shall see, the algorithms of IvD also contain an error.”

Xiyun Jiao and David A. van Dyk arXived a paper correcting an MCMC sampler and R package MNP for the multivariate probit model, proposed by Imai and van Dyk in 2005. [Hence the abbreviation IvD in the above quote.] Earlier versions of the Gibbs sampler for the multivariate probit model by Rob McCulloch and Peter Rossi in 1994, with a Metropolis update added by Agostino Nobile, and finally an improved version developed by Imai and van Dyk in 2005. As noted in the above quote, Jiao and van Dyk have discovered two mistakes in this latest version, jeopardizing the validity of the output.

The multivariate probit model considered here is a multinomial model where the occurrence of the k-th category is represented as the k-th component of a (multivariate) normal (correlated) vector being the largest of all components. The latent normal model being non-identifiable since invariant by either translation or scale, identifying constraints are used in the literature. This means using a covariance matrix of the form Σ/trace(Σ), where Σ is an inverse Wishart random matrix. In their 2005 implementation, relying on marginal data augmentation—which essentially means simulating the non-identifiable part repeatedly at various steps of the data augmentation algorithm—, Imai and van Dyk missed a translation term and a constraint on the simulated matrices that lead to simulations outside the rightful support, as illustrated from the above graph [snapshot from the arXived paper].

Since the IvD method is used in many subsequent papers, it is quite important that these mistakes are signalled and corrected. [Another snapshot above shows how much both algorithm differ!] Without much thinking about this, I [thus idly] wonder why an identifying prior is not taking the place of a hard identifying constraint, as it should solve the issue more nicely. In that it would create less constraints and more entropy (!) in exploring the augmented space, while theoretically providing a convergent approximation of the identifiable parts. I may (must!) however miss an obvious constraint preventing this implementation.

## probit posterior mean

Posted in Statistics, University life with tags , , , on March 9, 2012 by xi'an

In a recent arXiv report, Yuzo Maruyma shows that the posterior expectation of a probit parameter has an almost closed form (under a flat prior), namely

$\mathbb{E}[\beta|X,y] = (X^TX)^{-1} X^T\{2\text{diag}(y)-I_n\}\omega(X,y)$

where ω involves the integration of two quadratic forms over the n-dimensional unit sphere. Although this does not help directly with the MCMC derivation of the full posterior, this is an interesting lemma which shows a closed proximity with the standard least square estimate in linear regression.

## Vanilla Rao-Blackwellisation [re]revised

Posted in R, Statistics with tags , , , , , , , on June 1, 2010 by xi'an

Although the revision is quite minor, it took us two months to complete from the time I received the news in the Atlanta airport lounge… The vanilla Rao-Blackwellisation paper with Randal Douc has thus been resubmitted to the Annals of Statistics. And rearXived. The only significant change is the inclusion of two tables detailing computing time, like the one below

$\left| \begin{matrix} \tau &\text{median} &\text{mean }&q_{.8} &q_{.9} &\text{time}\\ 0.25 &0.0 &8.85 &4.9 &13 &4.2\\ 0.50 &0.0 &6.76 &4 &11 &2.25\\ 1.00 &0.25 &6.15 &4 &10 &2.5\\ 2.00 &0.20 &5.90 &3.5 &8.5 &4.5\\\end{matrix} \right|$

which provides different evaluations of the additional computing effort due to the use of the Rao–Blackwellisation: median and mean numbers of additional iterations, $80\%$ and $90\%$ quantiles for the additional iterations, and ratio of the average R computing times obtained over $10^5$ simulations. (Turning the above table into a formula acceptable by WordPress took me for ever, as any additional white space between the terms of the matrix is mis-interpreted!) Now, the mean time column does not look very supportive of the Rao-Blackwellisation technique, but this is due to the presence of a few outlying runs that required many iterations before hitting an acceptance probability of one. Excessive computing time can be curbed by using a pre-set number of iterations, as described in the paper…

## Vanilla Rao-Blackwellisation for revision

Posted in R, Statistics with tags , , , , , on March 18, 2010 by xi'an

The vanilla Rao-Blackwellisation paper with Randal Douc that had been resubmitted to the Annals of Statistics is now back for a revision, with quite encouraging comments:

The paper has been reviewed by two referees both of whom comment on the clear exposition and the novelty of the results. Both referees point to the empirical results as being suggestive of a more incremental improvement in practice rather than a major advance. However the approach the authors adopt is novel and I believe may motivate further developments in this area.

I cannot but agree on those comments! Since we are reducing the variance of the weights, the overall effect may be difficult to spot in practical applications. In the current version of the paper, we manage 20% reduction in the variance of those weights, but obviously this does not transfer to the same reduction of the variance of the overall estimator! Our vanilla Rao-Blackwellisation does not speed up the Markov chain.

## Two more handbook chapters

Posted in Books, Statistics with tags , , , , , , , , , on February 16, 2010 by xi'an

As mentioned in my earlier post, I had to write a revised edition to my chapter Bayesian Computational Methods in the Handbook of Computational Statistics (second edition), edited by J. Gentle, W. Härdle and Y. Mori. And in parallel I was asked for a second chapter in a handbook on risk analysis, Bayesian methods and expert elicitation, edited by Klaus Böckner. So, on Friday, I went over the first edition of this chapter of the Handbook of Computational Statistics and added the most recent developments I deemed important to mention, like ABC, as well as recent SMC and PMC algorithms, increasing the length by about ten pages. Simultaneously, Jean-Michel Marin completed my draft for the other handbook and I submitted both chapters, as well as arXived one and then the other.

It is somehow interesting (on a lazy blizzardly Sunday afternoon with nothing better to do!) to look for the differences between those chapters aiming at the same description of important computational techniques for Bayesian statistics (and based on the same skeleton). The first chapter is broader and, with its  60 pages, it functions as a (very) short book on the topic. Given that the first version was written in 2003, the focus is more on latent variables with mixture models being repeatedly used as examples. Reversible jump also stands preeminently. In my opinion, it reads well and could be used as a primary entry for a short formal course on computational methods. (Even though Introducing Monte Carlo Methods with R is presumably more appropriate for a short course.)

The second chapter started from the skeleton of the earlier version of the first chapter with the probit model as the benchmark example. I worked on a first draft during the last vacations and then Jean-Michel took over to produce this current version, where reversible jump has been removed and ABC introduced with greater details. In particular, we used a very special version of ABC with the probit model, resorting to the distance between the expectations of the binary observables, namely

$\sum_{j=1}^n (\Phi(x_j^\text{T} \beta) - \Phi(x_j^\text{T} \hat\beta) )^2,$

where $\hat\beta$ is the MLE of $\beta$ based on the observations, instead of the difference between the simulated and the observed binary observables

$\sum_{j=1}^n (y_j-y_j^0)^2$

which incorporates a useless randomness. With this choice, and when using for $\epsilon$ a .01 quantile, the difference with the true posterior on $\beta$ is very small, as shown by the figure (obtained for the Pima Indian dataset in R). Obviously, this stabilising trick only works in specific situations where a predictive of sorts can be computed.

## Vanilla Rao-Blackwellisation updated

Posted in Statistics with tags , , , , , on October 26, 2009 by xi'an

The vanilla Rao-Blackwellisation paper with Randal Douc has been updated, arXived, and resubmitted to the Annals, to include a more detailed study of the variance improvement brought by the Rao-Blackwellisation. In particular, we considered a probit modelling of the Pima Indian diabetes study, which is a standard benchmark that I also used in other papers and in the book Introducing Monte Carlo Methods with R with George Casella,. We also included the reference to the paper by Malefaki and Iliopoulos about Lemma 1, discussed in a previous post. The assessment of the variance improvement is examined in terms of the empirical variances of the individual terms

$h(z_j) \sum_{t=0}^\infty \prod_{m=1}^{t-1}\{1- \alpha(z_j,y_{jm})\}=h(z_j) \xi_j$

and

$h(z_j) \sum_{t=0}^\infty \prod_{m=1}^{t-1}\mathbb{I}\{u_{jm}\ge \alpha(z_j,y_{jm})\}=h(z_j) \mathfrak{n}_j$

involved in the two versions of the estimators, as explained in the original post (with a corrected typo due to the missing product in both terms!) and the variance ratio may be as small as two for some examples, including the Pima Indian diabetes. The improvement in the variances of $\xi_j$ against $\mathfrak{n}_j$ is most clearly seen in the boxplot below, in a exponential toy example where the acceptance rate is 1/13:

A phenomenon about boxplots that I cannot truly explain is that while the variance of the Rao-Blackwellised estimate of $\mathbb{E}[X]$ is 80% of the variance of the original estimator, the improvement is not that visible on the boxplot below of three samples of estimates (the middle one is our Rao-Blackwellised candidate, whose name vanished for being too large):

Now a 20% decrease in the variance means a mere 10% decrease in the standard deviation so this could agree with the difference in the upper bars of the boxplot… Since this is an example where the acceptance probability can be computed as well, I also added the optimal importance sampling estimator on the above boxplot and this confirms the impression that, while the variance decrease may be significant, this is not so clearly visible on the above boxplot.

We also took advantage of the fact that the quantity

$\alpha(z_j,y_{j0}) \sum_{t=0}^\infty \prod_{m=1}^{t-1}\{1- \alpha(z_j,y_{jm})\}$

where $y_{j0}$ is an additional simulation from the proposal $q(z_j,y)$ is a universal control variate for Metropolis-Hastings algorithms since its expectation is one. This means that in practice we can compute the regression coefficient $\hat\beta$ of $h(z_j)\xi_j$ on $\alpha(z_j,y_{j0})\xi_j$ and replace the average of the $h(z_j)\xi_j$‘s with the controled version

$\frac{1}{N} \sum_j h(z_j)\xi_j - \hat\beta(\alpha(z_j,y_{j0})\xi_j-1)$

with, again, improvement in the empirical variances that can reach a factor of 2.