An Empirical Analysis of Algorithms for Bayesian Sparse Linear Regression

Two conceptually similar algorithms for performing sparse linear regression under a Bayesian framework are presented. A novel implementation of a third algorithm is adapted from the log-linear regression task and applied to the linear regression model. The algorithms differ in terms of the search strategy used to explore the space of possible active variables; one uses a greedy approach, while the others use variants of stochastic local search. The algorithms are evaluated on a variety of simulated data sets, and the relative strengths and weaknesses of each approach are discussed.