Reversible Jump PDMP Samplers for Variable Selection

A new class of Markov chain Monte Carlo (MCMC) algorithms, based on simulating piecewise deterministic processes (PDMPs), has recently shown great promise: they are nonreversible, can mix better than standard MCMC and use subsampling ideas to speed up computation in big data scenarios. However, current PDMP samplers only sample from posterior densities that differentiable almost everywhere, which precludes their for model choice. Motivated by variable selection problems, we show how develop reversible jump jointly explore the discrete space models continuous parameters. Our framework is general: it takes any existing sampler, adds two types trans-dimensional moves allow addition or removal a model. We rates these be calculated so sampler correct invariant distribution. remove when associated parameter zero, this means do not depend likelihood. It is, thus, easy implement version fixed Supplementary materials article available online.