It’s been a while since I last wrote a blog post because I’ve been firmly keeping my shoulder to the wheel, where “shoulder”=brain and “wheel”=finishing the draft of my main in-progress malaria paper. So the following is simply a collection of random bits & pieces I thought about but didn’t find time to comment on from the last few weeks of astro ph:

– first, Brendon Brewer has a new paper on the arXiv describing an algorithm for inference from trans-dimensional models (here w/ birth & death moves) via diffusive nested sampling. Versions of this problem, though not using the example of detecting galaxies in a crowded field, have been considered in many guises in the statistics literature (e.g. for mixture modelling) so the motivation for adapting nested sampling to this purpose is to (hopefully) better handle posteriors (“handle posteriors”, lol) with phase transitions. The issue of thermodynamic integration failing on phase transitions was perhaps first introduced by Skilling in his 2006 paper with the “row of atoms” example given in Section 16. However, this example feels a bit contrived to me because a cursory inspection of the likelihood function and domain reveals a number of group symmetries that one would be crazy to ignore in any earnest attempt to conduct posterior inference. So, when I one day get a chance I look forward to playing with the examples in Brendon’s paper to get more of a feel for the relative performances of thermodynamic integration vs. nested sampling; and, more interestingly, sequential Monte Carlo vs. nested sampling.

– speaking of particle-based methods I noticed a rare use of PMC in astronomy a recent paper by Durkalec et al. Here the implementation is by way of the cosmoPMC code of Wraith et al. (2009) and Kilbinger et al. (2011), used previously for cosmological model selection.

– from today’s astro ph there’s another example of model fitting by KS-test statistic (section 4.3 of Bruzesse et al. ): a surprisingly common fudge in astronomical work (including early examples from yours truly, and apparently also Daniel Mortlock [priv. comm.])

– meanwhile the march of astrostatistics continues with a new astrostatistics coffee group set up this month at Oxford, and an upcoming workshop on astrostatistics at the RSS organised under the umbrella of the SuStain project; but the million dollar question remains unanswered: will any of this lead to some new tenured positions opening up in astrostatistics in this lifetime?

### Like this:

Like Loading...

*Related*

You might be interested to know that my paper was rejected. The two referees didn’t realise this: “the motivation for adapting nested sampling to this purpose is to (hopefully) better handle posteriors (“handle posteriors”, lol) with phase transitions”, so to them I seemed to be arbitrarily applying a complicated method to a well-solved problem for no reason.

Unlucky! I wonder if the structure I see in certain papers (e.g. Rue et al. 2008’s INLA paper: http://www.statslab.cam.ac.uk/~rjs57/RSS/0708/Rue08.pdf ) is developed from years of experience trying to avoid this problem: they start by discussing the general framework their investigation lies within, cite all the approaches within that framework, list some more specific approaches, state a more specific version of the topic they’ll consider …. and so on until the middle of page 6 when they first explain what exactly it is they’re proposing!