bceMCMC_mvrnorm_2: This is an MCMC function with a particular multivariate...

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

Description

MCMC 4, multivariate normal distributed steps, each step is one p-dimensional vector from a multivariate normal distribution, in this case the multivariate normal distribution is the laplace approximation to the variance code for sampling gaussian correlation parameters in a gaussian process

Usage

bceMCMC_mvrnorm_2(nmcmc, burn, thin, x, y, reg, step, priortheta)

Arguments

nmcmc	Number of mcmc samples to generate, before burning and thinning
burn	number of samples to burn
thin	keep only one of every 'thin' samples
x	predictors
y	response
reg	currently only option is "constant"
step	actually has no purpose in this function as the step length is defined by the fisher approximation to the covariance
priortheta	only currently only option is "Exp", "Higs" and "none" will also be implemented

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)
x <- as.matrix(x)
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_mvrnorm_2(1000,10,10,x,y,reg = "constant",step =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.