miranda

The miranda packages includes some modern pseudo-random number generators (PRNGs). These are up to 20x faster than the built-in PRNGs in R.

PRNG summary

PRNG	BigCrush result	Reference
xoshiro256p	“No Systematic Failures”	http://prng.di.unimi.it/
lehmer64	Passes	Lemire’s blog
splitmix64	Passes	source
romutrio	Passes	romu-random.org
runif() [Mersenne Twister]	Two failures

Installation

You can install from GitHub with:

# install.package('remotes')
remotes::install_github('coolbutuseless/miranda)

`xoshiro256p`

library(miranda)

# Set the seed
set_seed_xoshiro256p(2020)

# Get the internal state of the generator. 32 bytes.
(state <- get_state_xoshiro256p())
 [1] ee 6d 45 f5 0b 1c e9 8e 81 89 5b 6b ee ca 13 07 d9 3f 76 fe 13 4f c2 33 44
[26] 3c a9 de 5e b7 56 f9

# Generate some uniform random numbers from xoshiro256+
runif_xoshiro256p(5)
[1] 0.5322239 0.7788869 0.4601653 0.8475389 0.9540731

# Jump ahead. As if 2^128 numbers had been generated
jump_xoshiro256p()

# Generate some uniform random numbers from xoshiro256+
runif_xoshiro256p(5)
[1] 0.5714143 0.5006207 0.6907487 0.2625934 0.5053289

# reset the state to a prior value
set_state_xoshiro256p(state)

# Generate some uniform random numbers from xoshiro256+
runif_xoshiro256p(5)
[1] 0.5322239 0.7788869 0.4601653 0.8475389 0.9540731

Benchmark: Generating 1 million uniform random numbers

Click to show/hide bench::mark() code

N <- 1e6
res <- bench::mark(
  runif_xoshiro256p(N),
  runif_lehmer64(N),
  runif_splitmix64(N),
  runif_romutrio(N),
  runif(N),
  check = FALSE
)

The PRNGs in this package all roughly run at the same speed
At their best, these PRNGs are about 20x faster than runif()

Acknowledgements

R Core for developing and maintaining such a wonderful language.
CRAN maintainers, for patiently shepherding packages onto CRAN and maintaining the repository

Introducing 'miranda' - a package of fast, modern, uniform pseudo-random number generators