README.md

In nimble-dev/rcjit: Just-in-Time Compiler for R Code

Experimental jit compiler for R built on NIMBLE

devtools::install_github('nimble-dev/rcjit')
library(rcjit)

logistic_map <- function(start, scale, steps) {
    result <- start
    for (i in 1:steps) {
        result <- scale * result * (1.0 - result)
    }
    return(result)
}

jit_logistic_map <- jit(logistic_map)  # <---------- jit( )

print(logistic_map(0.1, 3.6, 1000000))
# 0.791767

print(jit_logistic_map(0.1, 3.6, 1000000))
# 0.791767

library(microbenchmark)
microbenchmark(jit_logistic_map(0.1, 3.6, 1000000),
                   logistic_map(0.1, 3.6, 1000000))
# Unit: milliseconds
#                              expr       min        lq      mean    median       uq       max neval
# jit_logistic_map(0.1, 3.6, 1e+06)  2.472324  2.473969  2.490519  2.476032  2.49078  2.890946   100
#     logistic_map(0.1, 3.6, 1e+06) 40.078393 40.120312 40.919318 40.298471 41.74336 47.068705   100

Rcjit is a thin wrapper library around NIMBLE to make NIMBLE's compiler infrastructure easy to use.

What can be jitted?

The rcjit jit() function handles only a subset of NIMBLE code, but it does try to handle simple R functions that "look like C++". For complete detail, check out the NIMBLE User Manual. Some features of NIMBLE are not yet avialable in rcjit, for example recursion or allowing one jitted function to use another; if you need these features immediately, try using NIMBLE directly.

This is an experimental interface and will be changing rapidly, but we'd love your feedback and feature requests.

What is NIMBLE?

NIMBLE is an R package for programming with BUGS models. Some probabilistic programming languages perform black-box inference on user-defined models. NIMBLE provides more flexibility and lets users additionally define algorithms against models: inference algorithms, experimental design algorithms, basically anything that uses a graphical model as basic concept. To make NIMBLE algorithms fast, the authors built a compiler from parts of R to C++. You can read details in :

"Programming with models: writing statistical algorithms for general model structures with NIMBLE" Perry de Valpine, Daniel Turek, Christopher J. Paciorek, Clifford Anderson-Bergman, Duncan Temple Lang, Rastislav Bodik https://arxiv.org/pdf/1505.05093.pdf

nimble-dev/rcjit documentation built on May 31, 2019, 8:29 a.m.