## Bayesian model comparison with intractable constants

Posted in Books, Kids, pictures, Statistics, Travel, University life with tags , , , , , , , , , , , on February 8, 2016 by xi'an Richard Everitt, Adam Johansen (Warwick), Ellen Rowing and Melina Evdemon-Hogan have updated [on arXiv] a survey paper on the computation of Bayes factors in the presence of intractable normalising constants. Apparently destined for Statistics and Computing when considering the style. A great entry, in particular for those attending the CRiSM workshop Estimating Constants in a few months!

A question that came to me from reading the introduction to the paper is why a method like Møller et al.’s (2006) auxiliary variable trick should be considered more “exact” than the pseudo-marginal approach of Andrieu and Roberts (2009) since the later can equally be seen as an auxiliary variable approach. The answer was on the next page (!) as it is indeed a special case of Andrieu and Roberts (2009). Murray et al. (2006) also belongs to this group with a product-type importance sampling estimator, based on a sequence of tempered intermediaries… As noted by the authors, there is a whole spectrum of related methods in this area, some of which qualify as exact-approximate, inexact approximate and noisy versions.

Their main argument is to support importance sampling as the method of choice, including sequential Monte Carlo (SMC) for large dimensional parameters. The auxiliary variable of Møller et al.’s (2006) is then part of the importance scheme. In the first toy example, a Poisson is opposed to a Geometric distribution, as in our ABC model choice papers, for which a multiple auxiliary variable approach dominates both ABC and Simon Wood’s synthetic likelihood for a given computing cost. I did not spot which artificial choice was made for the Z(θ)’s in both models, since the constants are entirely known in those densities. A very interesting section of the paper is when envisioning biased approximations to the intractable density. If only because the importance weights are most often biased due to the renormalisation (possibly by resampling). And because the variance derivations are then intractable as well. However, due to this intractability, the paper can only approach the impact of those approximations via empirical experiments. This leads however to the interrogation on how to evaluate the validity of the approximation in settings where truth and even its magnitude are unknown… Cross-validation and bootstrap type evaluations may prove too costly in realistic problems. Using biased solutions thus mostly remains an open problem in my opinion.

The SMC part in the paper is equally interesting if only because it focuses on the data thinning idea studied by Chopin (2002) and many other papers in the recent years. This made me wonder why an alternative relying on a sequence of approximations to the target with tractable normalising constants could not be considered. A whole sequence of auxiliary variable completions sounds highly demanding in terms of computing budget and also requires a corresponding sequence of calibrations. (Now, ABC fares no better since it requires heavy simulations and repeated calibrations, while further exhibiting a damning missing link with the target density. ) Unfortunately, embarking upon a theoretical exploration of the properties of approximate SMC is quite difficult, as shown by the strong assumptions made in the paper to bound the total variation distance to the true target.

## Unbiased Bayes for Big Data: Path of partial posteriors [a reply from the authors]

Posted in Statistics, University life with tags , , , , , , , , , on February 27, 2015 by xi'an [Here is a reply by Heiko Strathmann to my post of yesterday. Along with the slides of a talk in Oxford mentioned in the discussion.]

Thanks for putting this up, and thanks for the discussion. Christian, as already exchanged via email, here are some answers to the points you make.

First of all, we don’t claim a free lunch — and are honest with the limitations of the method (see negative examples). Rather, we make the point that we can achieve computational savings in certain situations — essentially exploiting redundancy (what Michael called “tall” data in his note on subsampling & HMC) leading to fast convergence of posterior statistics.

Dan is of course correct noticing that if the posterior statistic does not converge nicely (i.e. all data counts), then truncation time is “mammoth”. It is also correct that it might be questionable to aim for an unbiased Bayesian method in the presence of such redundancies. However, these are the two extreme perspectives on the topic. The message that we want to get along is that there is a trade-off in between these extremes. In particular the GP examples illustrate this nicely as we are able to reduce MSE in a regime where posterior statistics have *not* yet stabilised, see e.g. figure 6.

“And the following paragraph is further confusing me as it seems to imply that convergence is not that important thanks to the de-biasing equation.”

To clarify, the paragraph refers to the additional convergence issues induced by alternative Markov transition kernels of mini-batch-based full posterior sampling methods by Welling, Bardenet, Dougal & co. For example, Firefly MC’s mixing time is increased by a factor of 1/q where q*N is the mini-batch size. Mixing of stochastic gradient Langevin gets worse over time. This is not true for our scheme as we can use standard transition kernels. It is still essential for the partial posterior Markov chains to converge (if MCMC is used). However, as this is a well studied problem, we omit the topic in our paper and refer to standard tools for diagnosis. All this is independent of the debiasing device.

Yesterday in Oxford, Pierre Jacob pointed out that if MCMC is used for estimating partial posterior statistics, the overall result is not unbiased. We had a nice discussion how this bias could be addressed via a two-stage debiasing procedure: debiasing the MC estimates as described in the “Unbiased Monte Carlo” paper by Agapiou et al, and then plugging those into the path estimators — though it is (yet) not so clear how (and whether) this would work in our case.
In the current version of the paper, we do not address the bias present due to MCMC. We have a paragraph on this in section 3.2. Rather, we start from a premise that full posterior MCMC samples are a gold standard. Furthermore, the framework we study is not necessarily linked to MCMC – it could be that the posterior expectation is available in closed form, but simply costly in N. In this case, we can still unbiasedly estimate this posterior expectation – see GP regression.

“The choice of the tail rate is thus quite delicate to validate against the variance constraints (2) and (3).”

It is true that the choice is crucial in order to control the variance. However, provided that partial posterior expectations converge at a rate n with n the size of a minibatch, computational complexity can be reduced to N1-α (α<β) without variance exploding. There is a trade-off: the faster the posterior expectations converge, more computation can be saved; β is in general unknown, but can be roughly estimated with the “direct approach” as we describe in appendix.

It is true that for certain classes of models and φ functionals, the direct averaging of expectations for increasing data sizes yields good results (see log-normal example), and we state this. However, the GP regression experiments show that the direct averaging gives a larger MSE as with debiasing applied. This is exactly the trade-off mentioned earlier.

I also wonder what people think about the comparison to stochastic variational inference (GP for Big Data), as this hasn’t appeared in discussions yet. It is the comparison to “non-unbiased” schemes that Christian and Dan asked for.

## Unbiased Bayes for Big Data: Path of partial posteriors

Posted in Statistics, University life with tags , , , , , , , , , on February 26, 2015 by xi'an “Data complexity is sub-linear in N, no bias is introduced, variance is finite.”

Heiko Strathman, Dino Sejdinovic and Mark Girolami have arXived a few weeks ago a paper on the use of a telescoping estimator to achieve an unbiased estimator of a Bayes estimator relying on the entire dataset, while using only a small proportion of the dataset. The idea is that a sequence  converging—to an unbiased estimator—of estimators φt can be turned into an unbiased estimator by a stopping rule T: $\sum_{t=1}^T \dfrac{\varphi_t-\varphi_{t-1}}{\mathbb{P}(T\ge t)}$

is indeed unbiased. In a “Big Data” framework, the components φt are MCMC versions of posterior expectations based on a proportion αt of the data. And the stopping rule cannot exceed αt=1. The authors further propose to replicate this unbiased estimator R times on R parallel processors. They further claim a reduction in the computing cost of $\mathcal{O}(N^{1-\alpha})\qquad\text{if}\qquad\mathbb{P}(T=t)\approx e^{-\alpha t}$

which means that a sub-linear cost can be achieved. However, the gain in computing time means higher variance than for the full MCMC solution:

“It is clear that running an MCMC chain on the full posterior, for any statistic, produces more accurate estimates than the debiasing approach, which by construction has an additional intrinsic source of variance. This means that if it is possible to produce even only a single MCMC sample (…), the resulting posterior expectation can be estimated with less expected error. It is therefore not instructive to compare approaches in that region. “

I first got a “free lunch” impression when reading the paper, namely it sounded like using a random stopping rule was enough to overcome unbiasedness and large size jams. This is not the message of the paper, but I remain both intrigued by the possibilities the unbiasedness offers and bemused by the claims therein, for several reasons: Continue reading

## austerity in MCMC land

Posted in Statistics with tags , , , , , on April 28, 2013 by xi'an

Anoop Korattikara, Yutian Chen and Max Welling recently arXived a paper on the appeal of using only part of the data to speed up MCMC. This is different from the growing literature on unbiased estimators of the likelihood exemplified by Andrieu & Roberts (2009). Here, the approximation to the true target is akin to the approximation in ABC algorithms in that a value of the parameter is accepted if the difference in the likelihoods is larger than a given bound. Expressing this perspective as a test on the mean of the log likelihood leads the authors to use instead a subsample from the whole sample. (The approximation level ε is then a bound on the p-value.) While this idea only applies to iid settings, it is quite interesting and sounds a wee bit like a bootstrapped version of MCMC. Especially since it sounds as if it could provide an auto-evaluation of its error.