bceMCMC: Univariate, Uniform step length MCMC function, meant to be...

In galotalp/gpMCMC: Generate MCMC or FBI samples from the posterior distribution of Gaussian Process hyper-parameters

Description

Univariate, Uniform step length MCMC function, meant to be used by gpMCMC function

Usage

bceMCMC(nmcmc, burn, thin, x, y, reg, step, priortheta)

Arguments

nmcmc	number of MCMC samples to be generated before thinning and burning
burn	number of mcmc samples to burn
thin	keep one of every 'thin' samples
x	covariates
y	response
reg	only option currently is "constant"
step	step length for mcmc
priortheta	only option currently is "Exp"

Value

returns a list containing mcmc.ma (samples) and accept (acceptance rates)

Examples

nsamp <- 100
burn <- 200
thin <- 10

n <- 10
x1 <- seq(-5,10,length.out = n)
x2 <- seq(0,15,length.out = n)
x <- expand.grid(x1,x2)
d2 <- c(0.01,0.2,0,0) #here we set the theta parameters to be 0.01 and 0.2.
# These are the modes of the distribution that we will sample from using MCMC
cor.par <- data.frame(matrix(data = d2,nrow = dim(x)[2],ncol = 2))
names(cor.par) <- c("Theta.y","Alpha.y")

R <- cor.matrix(x,cor.par) # obtain covariance matrix
L <- chol(R)
z <- as.matrix(rnorm(n^2))
y <- L%*%z

gp <- bceMCMC(1000,10,10,x,y,reg = "constant",step =0.1, priortheta = "Exp")
mean(gp$mcmc.ma[,2]) #these means should be similar to the theta parameters set above
mean(gp$mcmc.ma[,1])

galotalp/gpMCMC documentation built on May 16, 2019, 5:36 p.m.