model.search: Perform Model Selection Using BIC

In mattdneal/gaussianProcess: Gaussian Process Fitting and Prediction

Description

Perform Model Selection Using BIC

Usage

model.search(x, y, base.model.tree, plot.gp = FALSE, reset.params = FALSE,
  max.new.nodes = 3, revisit.kernels = TRUE,
  scoring.function = bayesian.information.criterion,
  return.all.models = FALSE, alternate.scoring.functions = list(), ...)

Arguments

x	A matrix or data frame of predictors
y	A numeric vector of responses
base.model.tree	The model tree at which the search starts
plot.gp	Whether to plot the gaussian processes encountered during the search. If ncol(x) > 1 this parameter is ignored and no plots are created.
reset.params	By default the search starts with the optimum hyperparameter values found in the previous step in the search (with new hyperparameters fitted from random start points). Setting this to TRUE causes all hyperparameters to be randomly set at each step.
max.new.nodes	The number of kernel instances to add to the base model.
revisit.kernels	Whether to revisit previously-assessed kernels when deleting nodes from the current best candidate.
scoring.function	The scoring function for the GPs in the search. Must take a trained gaussianProcess object as its only parameter, and return a scalar score. If scoring.function returns a numeric vector only the first entry will be used.
...	Additional parameters to be passed to gp.obj$fit.hyperparameters (see gaussianProcess)

Value

The trained gaussianProcess object with the best BIC.

Examples

x <- rnorm(50)
y <- sin(1/(x^2 + 0.15))
mt <- create.model.tree.builtin()
gp <- model.search(x, y, mt)

mattdneal/gaussianProcess documentation built on May 21, 2019, 12:58 p.m.