Matlab Code for Bayesian Variable Selection
Marina Vannucci · 2000
This software provides a set of Matlab functions that perform Bayesian variable selection in a multivariate regression setting. There are different sets of functions currently available, implementing different approaches and mod-els for the variable selection problem: bvgs.tar, bvsme.tar, bvssa.tar (written by Marina Vannucci) and bvsgs i.tar, bvsgs g.tar, bvsgs gi.tar (written by Veronique Delouille and Marina Vannucci). Consider the multivariate regression model with p regressors, q responses and n observations, where p can largely exceed n. Bayesian variable selection approaches use a latent vector with p binary entries to identify the different submodels. The marginal posterior distribution of the binary vector is de-rived and, in high dimensions, Markov chain Monte Carlo algorithms are used to sample from this posterior distribution. Also, prediction can be done by computing a weighted average of the predictive distribution for the different models, or at least for a restricted set of them in the case of high dimensions, i.e. of a large p. The weights of the average are determined as the posterior