roadrunner: Fast, Low-Dependency Machine Learning Algorithms

roadrunner is a collection of fast, low-dependency implementations of classical machine learning algorithms with thin, base-R-style interfaces. The two algorithms shipped today are:

Details

• ares() – Multivariate Adaptive Regression Splines (MARS) with GCV / cross-validated pruning, hyperparameter autotune, bagging, weights, binomial / poisson / gamma families, and prediction intervals. Built on Friedman's (1991) fast least-squares forward selection with a parallelised knot search.

• krls() – Kernel Regularized Least Squares (Hainmueller and Hazlett 2014) with closed-form leave-one-out lambda selection, per-observation marginal effects, and binary first-difference handling.

Both engines are C++ on top of Rcpp, RcppArmadillo, and RcppParallel. Fits are deterministic across thread counts at a fixed seed.

References

Friedman, J. H. (1991). Multivariate Adaptive Regression Splines. Annals of Statistics, 19(1):1-67.

Hainmueller, J. and Hazlett, C. (2014). Kernel Regularized Least Squares: Reducing Misspecification Bias with a Flexible and Interpretable Machine Learning Approach. Political Analysis, 22(2):143-168.

See also

Useful links:

• https://cetialphafive.github.io/roadrunner/

• https://github.com/CetiAlphaFive/roadrunner

• Report bugs at https://github.com/CetiAlphaFive/roadrunner/issues

Author

Maintainer: Jack T. Rametta jtrametta@gmail.com (ORCID)

Authors:

• Jack T. Rametta jtrametta@gmail.com (ORCID)