fit.cbass {cbass}

Fit CBASS model using reversible jump MCMC

Description

Fit CBASS model using reversible jump MCMC

Usage

fit.cbass(
  X,
  y,
  max.int = 3,
  max.basis = 10 * ncol(X),
  tau2 = 10,
  nmcmc = 10000,
  nburn = round(nmcmc/2),
  nthin = 10,
  h1 = 4,
  h2 = 20 * (length(unique(y)) - 1)/nrow(X),
  p.int.prior = 1/(1:max.int),
  verbose = FALSE,
  print.interval = round(nmcmc/100),
  init1 = FALSE,
  ordinal = NULL,
  writeout = FALSE,
  writedir = tempdir(),
  mod = NULL,
  restart = FALSE
)

Arguments

X

n by p matrix of inputs on unit invteral

y

n-length factor factor of categories

max.int

maximum number of interactions, default 3

max.basis

maximum number of basis functions for each latent mean function, default ncol(X)*10

tau2

prior variance of basis regression coefficients, default 10

nmcmc

number of MCMC samples, default 1e4

nburn

number of MCMC samples to burn, default nmcmc/2

nthin

number of samples by which to thin, default 10

h1

first parameter for Gamma hyperprior on tau2, default 4

h2

second parameter for Gamma hyperprior on tau2, default 20(d-1)/n

p.int.prior

prior for number of interactions, default 1/(1:max.int)

verbose

should progress be printed out? default false

print.interval

how often should progress be printed out? default every 1%

init1

should model be initialized with single interaction model? default FALSE

ordinal

indicator of ordinal predictors (non-categorical), usually computed automatically

writeout

should samples be written out to text file? default FALSE

writedir

where should text files be written? default tempdir()

mod

initial / previous model fit, default NULL

restart

should initial input model be used for starting chain? default FALSE

Value

A list of CBASS model parameters. LIST THEM.

Examples

set.seed(1)
n <- 100;  d <- 3
X <- matrix(runif(n*2, 0, 1), ncol=2)
mu <- scale(X)
bound <- qnorm(1/d^(1/(d-1)))
mu <- cbind(bound, mu)
z <- mu
z[,-1] <- rnorm(length(mu[,-1]), mu[,-1], 1)
y <- apply(z, 1, which.max)
mod <- fit.cbass(X, y, max.int=1, max.basis=10, nmcmc=1e3, nburn=500, nthin=10)
pred.chain <- pred.cbass(mod, X)
mu.hat <- apply(pred.chain, 2:3, mean)
mean(abs(mu - mu.hat))
plot(c(mu), c(mu.hat))