fastrand
- Version: 0.1-0
- Status:
- License: BSD 2-Clause
- Author: Drew Schmidt
Fast uniform and gaussian (normal) random number generation. The package can be built to support a threaded backend via OpenMP or Intel Thread Building Blocks (TBB), as well as a GPU backend via NVIDIA's CUDA. The package is particularly useful for generating very large vectors for testing numerical algorithms.
The quality of the random numbers is not as good as base R's. But they are generated much more quickly, for a quasi performance-vs-accuracy tradeoff.
Installation
The development version is maintained on GitHub, and can easily be installed by any of the packages that offer installations from GitHub:
### Pick your preference
devtools::install_github("wrathematics/fastrand")
ghit::install_github("wrathematics/fastrand")
remotes::install_github("wrathematics/fastrand")By default, this will select an OpenMP backend. For Linux and Mac, to build with the GPU backend, you might do something like this:
R CMD INSTALL fastrand/ --configure-args="--with-backend=CUDA --with-cuda-home=/path/to/cuda/"For Windows, if you want to use the GPU backend, you will have to edit fastrand/src/Makevars.win appropriately.
Package Use and Benchmarks
n = 1e9
memuse::mu(n*8) # double precision size
## 7.451 GiB
memuse::mu(n*4) # single precision size
## 3.725 GiB
# warm ups
invisible(replicate(10, stats::rnorm(5)))
invisible(replicate(10, fastrand::rnorm(5)))
system.time(stats::rnorm(n))
## user system elapsed
## 53.104 1.072 54.182
invisible(gc())
system.time(fastrand::rnorm(n, type="float"))
## user system elapsed
## 118.104 0.432 32.947 There are other benchmarks located in the inst/benchmarks subtree of the package source.