RLT

Reinforcement Learning Trees is an R package for random forest models with regression, classification, and survival analysis support. The package uses a C++/Rcpp backend with OpenMP for parallel computation.

Website | GitHub | Issues

Highlights

• Regression, classification, and survival forests
• Reinforcement learning splits with embedded model guidance
• Linear combination splits for multivariate split directions
• Variable importance and random forest kernel utilities
• Survival prediction, confidence bands, and tree inspection tools
• Reproducible parallel fitting through recorded random seeds

Installation

Install the released version from CRAN when available:

install.packages("RLT")

Install the development version from GitHub:

# install.packages("remotes")
remotes::install_github("teazrq/RLT")

On Windows, source installation requires Rtools. On macOS and Linux, source installation requires a working C++ toolchain and OpenMP support.

Quick Example

library(RLT)

set.seed(1)
n <- 200
p <- 6
x <- matrix(rnorm(n * p), n, p)
y <- x[, 1] - x[, 2] + rnorm(n)

fit <- RLT(
  x, y,
  model = "regression",
  ntrees = 100,
  importance = TRUE,
  verbose = FALSE
)

pred <- predict(fit, x[1:5, ])
pred$Prediction

importance(fit)

Documentation

The pkgdown site is built from the package vignettes and website-only articles:

• Get Started
• Regression Tutorial
• Classification Tutorial
• Survival Tutorial
• Reference

References

• Zhu, R., Zeng, D., and Kosorok, M. R. (2015). Reinforcement Learning Trees. Journal of the American Statistical Association, 110(512), 1770-1784. https://doi.org/10.1080/01621459.2015.1036994
• Xu, T., Zhu, R., and Shao, X. (2024). On variance estimation of random forests with infinite-order U-statistics. Electronic Journal of Statistics, 18(1), 2135-2207.
• Formentini, S., Liang, K., and Zhu, R. (2022). Survival Function Confidence Band Estimation using Random Forests. https://arxiv.org/abs/2204.12038