Which is why I think the method should be considered in a parallel environment where every proposal can be recycled towards estimation, in the spirit of our parallel Metropolis-Hastings algorithm paper. (Thanks!)

]]>My “Ensemble MCMC” method (see http://arxiv.org/abs/1101.0387, and my blog entry at http://radfordneal.wordpress.com/2011/01/01/ensemble-mcmc/) can be seen as better way of doing something similar. One point I make in that paper is that one can expect an advantage only when there is some computational shortcut for evaluating the density at K points with less than K times the cost of evaluating the density at one point.

]]>