Matt Moores, Tony Pettitt, and Kerrie Mengersen arXived a paper yesterday comparing different computational approaches to the processing of hidden Potts models and of the intractable normalising constant in the Potts model. This is a very interesting paper, first because it provides a comprehensive survey of the main methods used in handling this annoying normalising constant Z(β), namely pseudo-likelihood, the exchange algorithm, path sampling (a.k.a., thermal integration), and ABC. A massive simulation experiment with individual simulation times up to 400 hours leads to select path sampling (what else?!) as the (XL) method of choice. Thanks to a pre-computation of the expectation of the sufficient statistic E[S(Z)|β]. I just wonder why the same was not done for ABC, as in the recent Statistics and Computing paper we wrote with Matt and Kerrie. As it happens, I was actually discussing yesterday in Columbia of potential if huge improvements in processing Ising and Potts models by approximating first the distribution of S(X) for some or all β before launching ABC or the exchange algorithm. (In fact, this is a more generic desiderata for all ABC methods that simulating directly if approximately the summary statistics would being huge gains in computing time, thus possible in final precision.) Simulating the distribution of the summary and sufficient Potts statistic S(X) reduces to simulating this distribution with a null correlation, as exploited in Cucala and Marin (2013, JCGS, Special ICMS issue). However, there does not seem to be an efficient way to do so, i.e. without reverting to simulating the entire grid X…
Archive for path sampling
The third day [morn] at our ICMS workshop was dedicated to path sampling. And rare events. Much more into [my taste] Monte Carlo territory. The first talk by Rosalind Allen looked at reweighting trajectories that are not in an equilibrium or are missing the Boltzmann [normalizing] constant. Although the derivation against a calibration parameter looked like the primary goal rather than the tool for constant estimation. Again papers in J. Chem. Phys.! And a potential link with ABC raised by Antonietta Mira… Then Jonathan Weare discussed stratification. With a nice trick of expressing the normalising constants of the different terms in the partition as solution(s) of a Markov system
Because the stochastic matrix M is easier (?) to approximate. Valleau’s and Torrie’s umbrella sampling was a constant reference in this morning of talks. Arnaud Guyader’s talk was in the continuation of Toni Lelièvre’s introduction, which helped a lot in my better understanding of the concepts. Rephrasing things in more statistical terms. Like the distinction between equilibrium and paths. Or bias being importance sampling. Frédéric Cérou actually gave a sort of second part to Arnaud’s talk, using importance splitting algorithms. Presenting an algorithm for simulating rare events that sounded like an opposite nested sampling, where the goal is to get down the target, rather than up. Pushing particles away from a current level of the target function with probability ½. Michela Ottobre completed the series with an entry into diffusion limits in the Roberts-Gelman-Gilks spirit when the Markov chain is not yet stationary. In the transient phase thus.
Chris Oates wrotes the following reply to my Icelandic comments on his paper with Theodore Papamarkou, and Mark Girolami, reply that is detailed enough to deserve a post on its own:
Thank you Christian for your discussion of our work on the Og, and also for your helpful thoughts in the early days of this project! It might be interesting to speculate on some aspects of this procedure:
(i) Quadrature error is present in all estimates of evidence that are based on thermodynamic integration. It remains unknown how to exactly compute the optimal (variance minimising) temperature ladder “on-the-fly”; indeed this may be impossible, since the optimum is defined via a boundary value problem rather than an initial value problem. Other proposals for approximating this optimum are compatible with control variates (e.g. Grosse et al, NIPS 2013, Friel and Wyse, 2014). In empirical experiments we have found that the second order quadrature rule proposed by Friel and Wyse 2014 leads to substantially reduced bias, regardless of the specific choice of ladder.
(ii) Our experiments considered first and second degree polynomials as ZV control variates. In fact, intuition specifically motivates the use of second degree polynomials: Let us presume a linear expansion of the log-likelihood in θ. Then the implied score function is constant, not depending on θ. The quadratic ZV control variates are, in effect, obtained by multiplying the score function by θ. Thus control variates can be chosen to perfectly correlate with the log-likelihood, leading to zero-variance estimators. Of course, there is an empirical question of whether higher-order polynomials are useful when this Taylor approximation is inappropriate, but they would require the estimation of many more coefficients and in practice may be less stable.
(iii) We require that the control variates are stored along the chain and that their sample covariance is computed after the MCMC has terminated. For the specific examples in the paper such additional computation is a negligible fraction of the total computational, so that we did not provide specific timings. When non-diffegeometric MCMC is used to obtain samples, or when the score is unavailable in closed-form and must be estimated, the computational cost of the procedure would necessarily increase.
For the wide class of statistical models with tractable likelihoods, employed in almost all areas of statistical application, the CTI we propose should provide state-of-the-art estimation performance with negligible increase in computational costs.
Chris Oates, Theodore Papamarkou, and Mark Girolami (all from the University of Warwick) just arXived a paper on a new form of thermodynamic integration for computing marginal likelihoods. (I had actually discussed this paper with the authors on a few occasions when visiting Warwick.) The other name of thermodynamic integration is path sampling (Gelman and Meng, 1998). In the current paper, the path goes from the prior to the posterior by a sequence of intermediary distributions using a power of the likelihood. While the path sampling technique is quite efficient a method, the authors propose to improve it through the recourse to control variates, in order to decrease the variance. The control variate is taken from Mira et al. (2013), namely a one-dimensional temperature-dependent transform of the score function. (Strictly speaking, this is an asymptotic control variate in that the mean is only asymptotically zero.) This control variate is then incorporated within the expectation inside the path sampling integral. Its arbitrary elements are then calibrated against the variance of the path sampling integral. Except for the temperature ladder where the authors use a standard geometric rate, as the approach does not account for Monte Carlo and quadrature errors. (The degree of the polynomials used in the control variates is also arbitrarily set.) Interestingly, the paper mixes a lot of recent advances, from the zero variance notion of Mira et al. (2013) to the manifold Metropolis-adjusted Langevin algorithm of Girolami and Calderhead (2011), uses as a base method pMCMC (Jasra et al., 2007). The examples processed in the paper are regression (where the controlled version truly has a zero variance!) and logistic regression (with the benchmarked Pima Indian dataset), with a counter-example of a PDE interestingly proposed in the discussion section. I quite agree with the authors that the method is difficult to envision in complex enough models. I also did not see mentions therein of the extra time involved in using this control variate idea.
With Matt Moores and Kerrie Mengersen, from QUT, we wrote this short paper just in time for the MCMSki IV Special Issue of Statistics & Computing. And arXived it, as well. The global idea is to cut down on the cost of running an ABC experiment by removing the simulation of a humongous state-space vector, as in Potts and hidden Potts model, and replacing it by an approximate simulation of the 1-d sufficient (summary) statistics. In that case, we used a division of the 1-d parameter interval to simulate the distribution of the sufficient statistic for each of those parameter values and to compute the expectation and variance of the sufficient statistic. Then the conditional distribution of the sufficient statistic is approximated by a Gaussian with these two parameters. And those Gaussian approximations substitute for the true distributions within an ABC-SMC algorithm à la Del Moral, Doucet and Jasra (2012).
Across 20 125 × 125 pixels simulated images, Matt’s algorithm took an average of 21 minutes per image for between 39 and 70 SMC iterations, while resorting to pseudo-data and deriving the genuine sufficient statistic took an average of 46.5 hours for 44 to 85 SMC iterations. On a realistic Landsat image, with a total of 978,380 pixels, the precomputation of the mapping function took 50 minutes, while the total CPU time on 16 parallel threads was 10 hours 38 minutes. By comparison, it took 97 hours for 10,000 MCMC iterations on this image, with a poor effective sample size of 390 values. Regular SMC-ABC algorithms cannot handle this scale: It takes 89 hours to perform a single SMC iteration! (Note that path sampling also operates in this framework, thanks to the same precomputation: in that case it took 2.5 hours for 10⁵ iterations, with an effective sample size of 10⁴…)
Since my student’s paper on Seaman et al (2012) got promptly rejected by TAS for quoting too extensively from my post, we decided to include me as an extra author and submitted the paper to this special issue as well.
Jeong Eun (Kate) Lee and I completed this paper, “Importance sampling schemes for evidence approximation in mixture models“, now posted on arXiv. (With the customary one-day lag for posting, making me bemoan the days of yore when arXiv would give a definitive arXiv number at the time of submission.) Kate came twice to Paris in the past years to work with me on this evaluation of Chib’s original marginal likelihood estimate (also called the candidate formula by Julian Besag). And on the improvement proposed by Berkhof, van Mechelen, and Gelman (2003), based on averaging over all permutations, idea that we rediscovered in an earlier paper with Jean-Michel Marin. (And that Andrew seemed to have completely forgotten. Despite being the very first one to publish [in English] a paper on a Gibbs sampler for mixtures.) Given that this averaging can get quite costly, we propose a preliminary step to reduce the number of relevant permutations to be considered in the averaging, removing far-away modes that do not contribute to the Rao-Blackwell estimate and called dual importance sampling. We also considered modelling the posterior as a product of k-component mixtures on the components, following a vague idea I had in the back of my mind for many years, but it did not help. In the above boxplot comparison of estimators, the marginal likelihood estimators are
- Chib’s method using T = 5000 samples with a permutation correction by multiplying by k!.
- Chib’s method (1), using T = 5000 samples which are randomly permuted.
- Importance sampling estimate (7), using the maximum likelihood estimate (MLE) of the latents as centre.
- Dual importance sampling using q in (8).
- Dual importance sampling using an approximate in (14).
- Bridge sampling (3). Here, label switching is imposed in hyperparameters.
Yet another full day working on Bayesian Core with Jean-Michel in Carnon… This morning, I ran along the canal for about an hour and at last saw some pink flamingos close enough to take pictures (if only to convince my daughter that there were flamingos in the area!). Then I worked full-time on the spatial statistics chapter, using a small dataset on sedges that we found in Gaetan and Guyon’s Spatial Statistics and Modelling. I am almost done tonight, with both path sampling and ABC R codes documented and working for this dataset. But I’d like to re-run both codes for longer to achieve smoother outcomes.