mmvbvs {MMVBVS}

Main function for variable selection

Description

Main function for variable selection

Usage

mmvbvs(X, Y, initial_chain, Phi, marcor, niter = 1000L, bgiter = 500L,
  hiter = 50L, burnin = 100000L, Vbeta = 1L, smallchange = 1e-04,
  verbose = TRUE)

Arguments

X

covariate with length N, sample size

Y

multivariate normal response variable N by P

initial_chain

list of starting points for beta, gamma, sigma, and sigmabeta. beta is length P for the coefficients, gamma is length P inclusion vector where each element is 0 or 1. sigma should be P x P covariance matrix, and sigmabeta should be the expected variance of the betas.

Phi

prior for the covariance matrix. We suggest identity matrix if there is no appropriate prior information

marcor

length P vector of correlation between X and each variable of Y

niter

total number of iteration for MCMC

bgiter

number of MH iterations within one iteration of MCMC to fit Beta and Gamma

hiter

number of first iterations to ignore

burnin

number of MH iterations for h, proportion of variance explained

Vbeta

variance of beta

smallchange

perturbation size for MH algorithm

verbose

if set TRUE, print gamma for each iteration

Value

list of posterior beta, gamma, and covariance matrix sigma

Examples

beta = c(rep(0.5, 3), rep(0,3))
n = 200; T = length(beta); nu = T+5
Sigma = matrix(0.8, T, T); diag(Sigma) = 1
X = as.numeric(scale(rnorm(n)))
 error = MASS::mvrnorm(n, rep(0,T), Sigma)
 gamma = c(rep(1,3), rep(0,3))
 Y = X %*% t(beta) + error; Y = scale(Y)
 Phi = matrix(0.5, T, T); diag(Phi) = 1
 initial_chain = list(beta = rep(0,T),
                        gamma = rep(0,T),
                        Sigma = Phi,
                        sigmabeta = 1)
     result = mmvbvs(X = X,
                     Y = Y,
                     initial_chain = initial_chain,
                     Phi = Phi,
                     marcor = colMeans(X*Y, na.rm=TRUE),
                     niter=10,
                     verbose = FALSE)