BARTModel: Bayesian Additive Regression Trees Model

Description

Flexible nonparametric modeling of covariates for continuous, binary, categorical and time-to-event outcomes.

Usage

BARTModel(
  K = integer(),
  sparse = FALSE,
  theta = 0,
  omega = 1,
  a = 0.5,
  b = 1,
  rho = numeric(),
  augment = FALSE,
  xinfo = matrix(NA, 0, 0),
  usequants = FALSE,
  sigest = NA,
  sigdf = 3,
  sigquant = 0.9,
  lambda = NA,
  k = 2,
  power = 2,
  base = 0.95,
  tau.num = numeric(),
  offset = numeric(),
  ntree = integer(),
  numcut = 100,
  ndpost = 1000,
  nskip = integer(),
  keepevery = integer(),
  printevery = 1000
)

Value

MLModel class object.

Arguments

K: if provided, then coarsen the times of survival responses per the quantiles \(1/K, 2/K, ..., K/K\) to reduce computational burdern.
sparse: logical indicating whether to perform variable selection based on a sparse Dirichlet prior rather than simply uniform; see Linero 2016.
theta, omega: \(theta\) and \(omega\) parameters; zero means random.
a, b: sparse parameters for \(Beta(a, b)\) prior: \(0.5 <= a <= 1\) where lower values induce more sparsity and typically \(b = 1\).
rho: sparse parameter: typically \(rho = p\) where \(p\) is the number of covariates under consideration.
augment: whether data augmentation is to be performed in sparse variable selection.
xinfo: optional matrix whose rows are the covariates and columns their cutpoints.
usequants: whether covariate cutpoints are defined by uniform quantiles or generated uniformly.
sigest: normal error variance prior for numeric response variables.
sigdf: degrees of freedom for error variance prior.
sigquant: quantile at which a rough estimate of the error standard deviation is placed.
lambda: scale of the prior error variance.
k: number of standard deviations \(f(x)\) is away from +/-3 for categorical response variables.
power, base: power and base parameters for tree prior.
tau.num: numerator in the \(tau\) definition, i.e., \(tau = tau.num / (k * sqrt(ntree))\).
offset: override for the default \(offset\) of \(F^-1(mean(y))\) in the multivariate response probability \(P(y[j] = 1 | x) = F(f(x)[j] + offset[j])\).
ntree: number of trees in the sum.
numcut: number of possible covariate cutoff values.
ndpost: number of posterior draws returned.
nskip: number of MCMC iterations to be treated as burn in.
keepevery: interval at which to keep posterior draws.
printevery: interval at which to print MCMC progress.

Details

Response types:: factor, numeric, Surv

Default values and further model details can be found in the source links below.

Examples

Run this code

# \donttest{
## Requires prior installation of suggested package BART to run

fit(sale_amount ~ ., data = ICHomes, model = BARTModel)
# }