Rborist, the R package version of the Arborist decision-tree utility, is now available from CRAN. Version 0-1.0 is the initial release.
The Arborist implements the Random Forest algorithm with an emphasis on performance and extensibility. Particular attention has been paid to the role of data locality, for example, and key elements of the software are parallelized for multicore hardware. Quantile regression is provided natively, and additional internalized workflows are under development. Both numerical and categorical data types are supported, with no preset limit on factor cardinality.
The Arborist is currently an in-memory implementation, with out-of-memory extensions envisioned in the medium term. More immediate plans for the software include specialized GPU implementations, also currently under development.
The Rborist package interface is strongly influenced by the venerable randomForest package. Options and results are similar, although not identical, hopefully providing an accessible look-and-feel to randomForest users. As with the randomForest package, prediction methods allow trained forests to be reused on separately-provided data sets.
Please download the package and give it a try. Constructive criticism is always appreciated.