Skip to contents

Function to include prior information as split probability for the Bayesian additive regression tree model added via engine_bart.

Priors for engine_bart have to be specified as transition probabilities of variables which are internally used to generate splits in the regression tree. Specifying a prior can thus help to 'enforce' a split with a given variable. These can be numeric and coded as values between 0 and 1.

Usage

BARTPrior(variable, hyper = 0.75, ...)

# S4 method for class 'character'
BARTPrior(variable, hyper = 0.75, ...)

Arguments

variable

A character matched against existing predictors or latent effects.

hyper

A numeric object with a number being >0 and equal to 1. Defaults to 0.75.

...

Variables passed on to prior object.

Note

Even if a given variable is included as split in the regression or classification tree, this does not necessarily mean that the prediction changes if the value is non-informative (as the split can occur early on). It does however affect any variable importance estimates calculated from the model.

References

  • Chipman, H., George, E., and McCulloch, R. (2009) BART: Bayesian Additive Regression Trees.

  • Chipman, H., George, E., and McCulloch R. (2006) Bayesian Ensemble Learning. Advances in Neural Information Processing Systems 19, Scholkopf, Platt and Hoffman, Eds., MIT Press, Cambridge, MA, 265-272.