dynamic mixtures [at NBBC15]

Posted in R, Statistics with tags , , , , , , , , , , , , on June 18, 2015 by xi'an

A funny coincidence: as I was sitting next to Arnoldo Frigessi at the NBBC15 conference, I came upon a new question on Cross Validated about a dynamic mixture model he had developed in 2002 with Olga Haug and Håvård Rue [whom I also saw last week in Valencià]. The dynamic mixture model they proposed replaces the standard weights in the mixture with cumulative distribution functions, hence the term dynamic. Here is the version used in their paper (x>0)

$(1-w_{\mu,\tau}(x))f_{\beta,\lambda}(x)+w_{\mu,\tau}(x)g_{\epsilon,\sigma}(x)$

where f is a Weibull density, g a generalised Pareto density, and w is the cdf of a Cauchy distribution [all distributions being endowed with standard parameters]. While the above object is not a mixture of a generalised Pareto and of a Weibull distributions (instead, it is a mixture of two non-standard distributions with unknown weights), it is close to the Weibull when x is near zero and ends up with the Pareto tail (when x is large). The question was about simulating from this distribution and, while an answer was in the paper, I replied on Cross Validated with an alternative accept-reject proposal and with a somewhat (if mildly) non-standard MCMC implementation enjoying a much higher acceptance rate and the same fit.

capture mark recapture with no mark and no recapture [aka 23andmyfish]

Posted in Mountains, pictures, Statistics, Travel, University life with tags , , , , , , , , on June 11, 2015 by xi'an

A very exciting talk today at NBBC15 here in Reykjavik was delivered by Mark Bravington yesterday on Close-kin mark recapture by modern magic (!). Although Mark is from Australia, being a Hobart resident does qualify him for the Nordic branch of the conference! The exciting idea is to use genetic markers to link catches in a (fish) population as being related as parent-offspring or as siblings. This sounds like science-fantasy when you first hear of it!, but it is actually working better than standard capture-mark-recapture methods for populations of a certain size (so that the chances to find related animals are not the absolute zero!, as, e.g., krill populations). The talk was focussed on bluefin tuna, whose survival is unlikely under the current fishing pressure… Among the advantages, a much more limited impact of the capture on the animal, since only a small amount of genetic material is needed, no tag loss, tag destruction by hunters, or tag impact of the animal survival, no recapture, a unique identification of each animal, and the potential for a detailed amount of information through the genetic record. Ideally, the entire sample could lead to a reconstruction of its genealogy all the way to the common ancestor, a wee bit like what 23andme proposes for humans, but this remains at the science-fantasy level given what is currently know about the fish species genomes.

Icelandic landscape [#2]

Posted in Mountains, pictures, Travel with tags , , , , , , , on June 10, 2015 by xi'an