XBART: Accelerated Bayesian Additive Regression Trees

Composition No. III, with Red, Blue, Yellow, and Black, Piet Mondrian (1929)

XBART is a stochastic hill climbing algorithm which provide fast inference of BART posterior. Simulation studies shows that XBART is comparable in computation time and more accurate at function estimation than both random forest and gradient boosting. slides and paper of the algorithm. 中文简介.

The beta version is posted on my website, will be upload to CRAN and pip soon. You are welcome to test the code. Bug reports are valuable and highly appreciated.

1. Installing the pacakge

Install R package from Github

The pacakge relies on Rcpp and RcppArmadillo.
Dowload it from Github.
library(dev_tools) install_github("jingyuhe/xbart")
C++ 11 compiler is required, please install Rtools (Windows) or Xcode and Command Line Tools (Mac).

Install python package from pip

pip install XBART or pip3 install XBART.

2. Sample code

R script and python script, data for python demo.