tar_seed_create: Create a seed for a target.

In targets: Dynamic Function-Oriented 'Make'-Like Declarative Pipelines

Create a seed for a target.

Description

Create a seed for a target.

Usage

tar_seed_create(name, global_seed = NULL)

Arguments

name	Character of length 1, target name.
global_seed	Integer of length 1, the overarching global pipeline seed which governs the seeds of all the targets. Set to NULL to default to tar_option_get("seed"). Set to NA to disable seed setting in targets and make tar_seed_create() return NA_integer_.

Value

Integer of length 1, the target seed.

Seeds

A target's random number generator seed is a deterministic function of its name and the global pipeline seed from tar_option_get("seed"). Consequently,

1. Each target runs with a reproducible seed so that
   different runs of the same pipeline in the same computing
   environment produce identical results.
2. No two targets in the same pipeline share the same seed.
   Even dynamic branches have different names and thus different seeds.

You can retrieve the seed of a completed target with tar_meta(your_target, seed) and run tar_seed_set() on the result to locally recreate the target's initial RNG state. tar_workspace() does this automatically as part of recovering a workspace.

RNG overlap

In theory, there is a risk that the pseudo-random number generator streams of different targets will overlap and produce statistically correlated results. (For a discussion of the motivating problem, see the Section 6: "Random-number generation" in the parallel package vignette: vignette(topic = "parallel", package = "parallel").) However, this risk is extremely small in practice, as shown by L'Ecuyer et al. (2017) \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1016/j.matcom.2016.05.005")} under "A single RNG with a 'random' seed for each stream" (Section 4: under "How to produce parallel streams and substreams"). targets and tarchetypes take the approach discussed in the aforementioned section of the paper using the secretbase package by Charlie Gao (2024) \Sexpr[results=rd]{tools:::Rd_expr_doi("10.5281/zenodo.10553140")}. To generate the 32-bit integer seed argument of set.seed() for each target, secretbase generates a cryptographic hash using the SHAKE256 extendable output function (XOF). secretbase uses algorithms from the ⁠Mbed TLS⁠ C library.

References

• Gao C (2024). secretbase: Cryptographic Hash and Extendable-Output Functions. R package version 0.1.0, \Sexpr[results=rd]{tools:::Rd_expr_doi("10.5281/zenodo.10553140")}.

• Pierre L'Ecuyer, David Munger, Boris Oreshkin, and Richard Simard (2017). Random numbers for parallel computers: Requirements and methods, with emphasis on GPUs. Mathematics and Computers in Simulation, 135, 3-17. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1016/j.matcom.2016.05.005")}.

See Also

Other pseudo-random number generation: tar_seed_get(), tar_seed_set()

targets documentation built on Oct. 3, 2024, 1:11 a.m.