rngtools 1.2

Generate Sequence of Random Streams

Usage

RNGseq(n, seed = NULL, ..., simplify = TRUE, version = 2)

RNGseq_seed(seed = NULL, normal.kind = NULL, verbose = FALSE, version = 2)

Arguments

n
Number of streams to be created
seed
seed specification used to initialise the set of streams using RNGseq_seed.
simplify
a logical that specifies if sequences of length 1 should be unlisted and returned as a single vector.
...
extra arguments passed to RNGseq_seed.
normal.kind
Type of Normal random generator. See RNG.
verbose
logical to toggle verbose messages
version
version of the function to use, to reproduce old behaviours. Version 1 had a bug which made the generated stream sequences share most of their seeds (!), as well as being not equivalent to calling set.seed(seed); RNGseq_seed(NULL). Version 2 fixes this bug.

Value

a list of integer vectors (or a single integer vector if n=1 and unlist=TRUE).

a 7-length numeric vector.

Description

Create a given number of seeds for L'Ecuyer's RNG, that can be used to seed parallel computation, making them fully reproducible.

RNGseq_seed generates the -- next -- random seed used as the first seed in the sequence generated by RNGseq.

Details

This ensures complete reproducibility of the set of run. The streams are created using L'Ecuyer's RNG, implemented in R core since version 2.14.0 under the name "L'Ecuyer-CMRG" (see RNG).

Generating a sequence without specifying a seed uses a single draw of the current RNG. The generation of a sequence using seed (a single or 6-length numeric) a should not affect the current RNG state.

Examples

RNGseq(3)



[[1]]
[1]        407 -978439420 1478102581 -305458830 -856571317  650208528 1074773457

[[2]]
[1]         407  1609044590  1253481101  1620516322  -928645980 -2106955413  1296706785

[[3]]
[1]         407  -955365930  1376808480 -1027324033  1373558928   821426560 -1902732257


RNGseq(3)



[[1]]
[1]         407  1988955604   645650117  -237503294 -1634909861  1690957024  -334989471

[[2]]
[1]         407 -2020884048 -1713953665  2113742629   159255880 -1097925173  2099880136

[[3]]
[1]         407   594549532  1801224098   798612488 -1899955208 -1353578868  1074518195


RNGseq(3, seed=123)



[[1]]
[1]         407   642048078    81368183 -2093158836   506506973  1421492218 -1906381517

[[2]]
[1]         407  1340772676 -1389246211  -999053355  -953732024  1888105061  2010658538

[[3]]
[1]         407 -1318496690  -948316584   683309249  -990823268 -1895972179  1275914972


# or identically
set.seed(123)

identical(RNGseq(3), RNGseq(3, seed=123))



[1] TRUE

RNGseq(3, seed=1:6, verbose=TRUE)


# Original RNG: Mersenne-Twister - Inversion [403, 1, 515190382, 2133433928, 917665867, 1283494313, 1101294840, ...]
# Directly use 6-long seed: 1, 2, 3, 4, 5, 6 ... 
OK
# Seed RNGkind is: L'Ecuyer-CMRG - Inversion [407, 1, 2, 3, 4, 5, 6]
# Restoring RNG: Mersenne-Twister - Inversion [403, 1, 515190382, 2133433928, 917665867, 1283494313, 1101294840, ...]

[[1]]
[1] 407   1   2   3   4   5   6

[[2]]
[1]         407  -447371532   542750874  -935969228  -269326340   701604884 -1748056907

[[3]]
[1]         407   311773008 -1393648596   433058656  -545474683  2059732357   994549473


# select Normal kind
RNGseq(3, seed=123, normal.kind="Ahrens")



[[1]]
[1]         107   642048078    81368183 -2093158836   506506973  1421492218 -1906381517

[[2]]
[1]         107  1340772676 -1389246211  -999053355  -953732024  1888105061  2010658538

[[3]]
[1]         107 -1318496690  -948316584   683309249  -990823268 -1895972179  1275914972


## generate a seed for RNGseq
# random
RNGseq_seed()



[1]        407 -770194214  920761491  494670008 -102057127 -942372026 1535907087

RNGseq_seed()



[1]         407    82000378 -1377755341  -888237352  -352809223  1464214118 -1790002769

RNGseq_seed(NULL)



[1]         407  1285013706  -887512125 -1801713112  -140498423 -1754456522  -350228673

# fixed
RNGseq_seed(1)



[1]         407 -1535484873  1222746892  1963142301   158053050 -1240755981  -292394600

RNGseq_seed(1:6)



[1] 407   1   2   3   4   5   6

# `RNGseq_seed(1)` is identical to
set.seed(1)

s <- RNGseq_seed()

identical(s, RNGseq_seed(1))



[1] TRUE

See also

RNGseq