inst/benchmarks/README.md

In bartMachine: Bayesian Additive Regression Trees

Benchmark Suite: bartMachine vs randomForest

This benchmark suite runs repeated K-fold cross-validation and compares out-of-sample performance between bartMachine and randomForest across datasets drawn from standard benchmark libraries (e.g., datasets, MASS, mlbench, ISLR).

Quick start

1. Install the package and optional dataset libraries:

install.packages(c("bartMachine", "randomForest", "mlbench", "ISLR", "pROC"))

1. Run the suite from the package root:

Rscript inst/benchmarks/run_benchmark_suite.R --folds=5 --repeats=3

1. Outputs are written to inst/benchmarks/results:
2. benchmark_folds.csv: per-fold metrics
3. benchmark_summary.csv: mean and standard deviation by dataset/model
4. benchmark_skipped.csv: datasets that were skipped and why
5. benchmark_results.rds: all results + config
6. sessionInfo.txt: session metadata

Notes

• Set Java memory and other parameters before loading bartMachine. You can do this via:
• options(java.parameters = c("-Xmx20g", "--add-modules=jdk.incubator.vector", "-XX:+UseZGC"))
• Use --list to view all dataset definitions and --dry-run to list the filtered selection without running.
• Large or slow datasets can be skipped with --skip-tags=large.
• pROC is optional; if installed, AUC is computed for classification tasks.

Example filters

Rscript inst/benchmarks/run_benchmark_suite.R --packages=datasets,MASS --folds=3
Rscript inst/benchmarks/run_benchmark_suite.R --datasets=Boston,BostonHousing
Rscript inst/benchmarks/run_benchmark_suite.R --skip-tags=large --repeats=2

bartMachine documentation built on Jan. 19, 2026, 9:06 a.m.