RcppTN is an R package that relies on the fantastic Rcpp R package. The single, clear goal of RcppTN is to facilitate working with truncated Normal distributions. RcppTN has grown out of my own need to draw from arbitrary truncated Normal distributions and calculate the associated moments of these distributions.
Other RNGs for the truncated Normal distribution exist, so why another? Other RNGs didn’t meet my needs. The existing R implementations (even if written in C or C++) did not readily expose themselves at the compiled code level. That meant my compiled code could not use the same RNG as my R code. The existing C++ implementations didn’t have an R–level interface and didn’t obey R’s RNG state. That meant lots of work spent on a retro-fit.
My solution was an implementation designed for use at the R level and the C++ level in R-based computation.
- the R package implements an efficient sampler (Robert 1995) in an efficient language (C++)
- copies are avoided with references and STL constructs
- the RNG at the R level and the C++ level are in sync with each other and the rest of an R session
- the RNG is easily seeded with R’s
- R–level and C++–level interfaces call the same underlying code
- the R and C++ interfaces are simple to use in your own R or C++ code thanks to Rcpp
- the R package has a growing suite of unit tests testing for correctness
- C++ memory operations profiled with valgrind, only clean code is released into the wild
For the entire project, see the github repo.
For an explanation on how to use RcppTN at the R or C++ level, see the “using” vignette.
For a performance benchmark of RcppTN to alternatives, see the “speed” vignette.
The current release of the package is RcppTN 0.1-6