Archive for Hilbert

Bayes and unbiased

Posted in Books, Statistics with tags , , , on December 21, 2015 by xi'an

Siamak Noorbaloochi and Glen Meeden have arXived a note on some mathematical connections between Bayes estimators and unbiased estimators.  To start with, I never completely understood the reasons for [desperately] seeking unbiasedness in estimators, given that most transforms of a parameter θ do not allow for unbiased estimators when based on a finite sample from a parametric family with such parameter θ (Lehmann, 1983). (This is not even for a Bayesian reason!) The current paper first seems to use unbiasedness in a generalised sense introduced earlier by the authors in 1983, which is a less intuitive notion since it depends on loss and prior, still without being guaranteed to exist since it involves an infimum over a non-compact space. However, for the squared error loss adopted in this paper, it seems to reduce to the standard notion.

A first mathematical result therein is that the Bayes [i.e., posterior mean] and unbiasedness [i.e., sample mean] operators are adjoint in a Hilbert sense. But this does not seem much more than a consequence of Fubini’s theorem. The authors then proceeds to expose the central (decomposition) result of the paper, namely that every estimable function γ(θ) can be orthogonally decomposed into a function with an unbiased estimator plus a function whose Bayes estimator is zero and that conversely every square-integrable estimator can be orthogonally decomposed into a Bayes estimator (of something) plus an unbiased estimator of zero. This is a neat duality result, whose consequences I however fail to see because the Bayes estimator is estimating something else. And, somewhere, somehow, I have some trouble with the notion of a function α whose Bayes estimator [i.e., posterior mean] is zero for all values of the sample, esp. outside problems with finite observation and finite parameter spaces. For instance, if the sampling distribution is within an exponential family, the above property means that the Laplace transform of this function α is uniformly zero, hence that the function itself is uniformly zero.

discussions on Gerber and Chopin

Posted in Books, Kids, Statistics, University life with tags , , , , , , , , , , , , , , , on May 29, 2015 by xi'an

As a coincidence, I received my copy of JRSS Series B with the Read Paper by Mathieu Gerber and Nicolas Chopin on sequential quasi Monte Carlo just as I was preparing an arXival of a few discussions on the paper! Among the [numerous and diverse] discussions, a few were of particular interest to me [I highlighted members of the University of Warwick and of Université Paris-Dauphine to suggest potential biases!]:

  1. Mike Pitt (Warwick), Murray Pollock et al.  (Warwick) and Finke et al. (Warwick) all suggested combining quasi Monte Carlo with pseudomarginal Metropolis-Hastings, pMCMC (Pitt) and Rao-Bklackwellisation (Finke et al.);
  2. Arnaud Doucet pointed out that John Skilling had used the Hilbert (ordering) curve in a 2004 paper;
  3. Chris Oates, Dan Simpson and Mark Girolami (Warwick) suggested combining quasi Monte Carlo with their functional control variate idea;
  4. Richard Everitt wondered about the dimension barrier of d=6 and about possible slice extensions;
  5. Zhijian He and Art Owen pointed out simple solutions to handle a random number of uniforms (for simulating each step in sequential Monte Carlo), namely to start with quasi Monte Carlo and end up with regular Monte Carlo, in an hybrid manner;
  6. Hans Künsch points out the connection with systematic resampling à la Carpenter, Clifford and Fearnhead (1999) and wonders about separating the impact of quasi Monte Carlo between resampling and propagating [which vaguely links to one of my comments];
  7. Pierre L’Ecuyer points out a possible improvement over the Hilbert curve by a preliminary sorting;
  8. Frederik Lindsten and Sumeet Singh propose using ABC to extend the backward smoother to intractable cases [but still with a fixed number of uniforms to use at each step], as well as Mateu and Ryder (Paris-Dauphine) for a more general class of intractable models;
  9. Omiros Papaspiliopoulos wonders at the possibility of a quasi Markov chain with “low discrepancy paths”;
  10. Daniel Rudolf suggest linking the error rate of sequential quasi Monte Carlo with the bounds of Vapnik and Ĉervonenkis (1977).

 The arXiv document also includes the discussions by Julyan Arbel and Igor Prünster (Turino) on the Bayesian nonparametric side of sqMC and by Robin Ryder (Dauphine) on the potential of sqMC for ABC.

Numerical analysis for statisticians

Posted in Books, R, Statistics, University life with tags , , , , , , , , , on August 26, 2011 by xi'an

“In the end, it really is just a matter of choosing the relevant parts of mathematics and ignoring the rest. Of course, the hard part is deciding what is irrelevant.”

Somehow, I had missed the first edition of this book and thus I started reading it this afternoon with a newcomer’s eyes (obviously, I will not comment on the differences with the first edition, sketched by the author in the Preface). Past the initial surprise of discovering it was a mathematics book rather than an algorithmic book, I became engrossed into my reading and could not let it go! Numerical Analysis for Statisticians, by Kenneth Lange, is a wonderful book. It provides most of the necessary background in calculus and some algebra to conduct rigorous numerical analyses of statistical problems. This includes expansions, eigen-analysis, optimisation, integration, approximation theory, and simulation, in less than 600 pages. It may be due to the fact that I was reading the book in my garden, with the background noise of the wind in tree leaves, but I cannot find any solid fact to grumble about! Not even about  the MCMC chapters! I simply enjoyed Numerical Analysis for Statisticians from beginning till end.

“Many fine textbooks (…) are hardly substitutes for a theoretical treatment emphasizing mathematical motivations and derivations. However, students do need exposure to real computing and thoughtful numerical exercises. Mastery of theory is enhanced by the nitty gritty of coding.” 

From the above, it may sound as if Numerical Analysis for Statisticians does not fulfill its purpose and is too much of a mathematical book. Be assured this is not the case: the contents are firmly grounded in calculus (analysis) but the (numerical) algorithms are only one code away. An illustration (among many) is found in Section 8.4: Finding a Single Eigenvalue, where Kenneth Lange shows how the Raleigh quotient algorithm of the previous section can be exploited to this aim, when supplemented with a good initial guess based on Gerschgorin’s circle theorem. This is brilliantly executed in two pages and the code is just one keyboard away. The EM algorithm is immersed into a larger M[&]M perspective. Problems are numerous and mostly of high standards, meaning one (including me) has to sit and think about them. References are kept to a minimum, they are mostly (highly recommended) books, plus a few papers primarily exploited in the problem sections. (When reading the Preface, I found that “John Kimmel, [his] long suffering editor, exhibited extraordinary patience in encouraging [him] to get on with this project”. The quality of Numerical Analysis for Statisticians is also a testimony to John’s editorial acumen!)

“Every advance in computer architecture and software tempts statisticians to tackle numerically harder problems. To do so intelligently requires a good working knowledge of numerical analysis. This book equips students to craft their own software and to understand the advantages and disadvantages of different numerical methods. Issues of numerical stability, accurate approximation, computational complexity, and mathematical modeling share the limelight in a broad yet rigorous overview of those parts of numerical analysis most relevant to statisticians.”

While I am reacting so enthusiastically to the book (imagine, there is even a full chapter on continued fractions!), it may be that my French math background is biasing my evaluation and that graduate students over the World would find the book too hard. However, I do not think so: the style of Numerical Analysis for Statisticians is very fluid and the rigorous mathematics are mostly at the level of undergraduate calculus. The more advanced topics like wavelets, Fourier transforms and Hilbert spaces are very well-introduced and do not require prerequisites in complex calculus or functional analysis. (Although I take no joy in this, even measure theory does not appear to be a prerequisite!) On the other hand, there is a prerequisite for a good background in statistics. This book will clearly involve a lot of work from the reader, but the respect shown by Kenneth Lange to those readers will sufficiently motivate them to keep them going till assimilation of those essential notions. Numerical Analysis for Statisticians is also recommended for more senior researchers and not only for building one or two courses on the bases of statistical computing. It contains most of the math bases that we need, even if we do not know we need them! Truly an essential book.

Follow

Get every new post delivered to your Inbox.

Join 1,020 other followers