runParallel: runParallel

View source: R/runParallel.r

runParallelR Documentation

runParallel

Description

parallel Package Easy Front-End

Usage

runParallel(
  onecore,
  reps,
  seed = round(runif(1, 0, 10000)),
  cores = max(1, parallel::detectCores() - 1),
  simplify = TRUE,
  along
)

Arguments

onecore

function to run the analysis on one core

reps

total number of repetitions

seed

species the base random number seed. The seed used for core i will be seed + i.

cores

number of cores to use, defaulting to one less than the number available

simplify

set to FALSE to not create an outer list if a onecore result has only one element

along

see Details

Details

Given a function onecore that runs the needed set of simulations on one CPU core, and given a total number of repetitions reps, determines the number of available cores and by default uses one less than that. By default the number of cores is one less than the number available on your machine. reps is divided as evenly as possible over these cores, and batches are run on the cores using the parallel package mclapply function. The current per-core repetition number is continually updated in your system's temporary directory (/tmp for Linux and Mac, TEMP for Windows) in a file name progressX.log where X is the core number. The random number seed is set for each core and is equal to the scalar seed - core number + 1. The default seed is a random number between 0 and 10000 but it's best if the user provides the seed so the simulation is reproducible. The total run time is computed and printed onefile must create a named list of all the results created during that one simulation batch. Elements of this list must be data frames, vectors, matrices, or arrays. Upon completion of all batches, all the results are rbind'd and saved in a single list.

onecore must have an argument reps that will tell the function how many simulations to run for one batch, another argument showprogress which is a function to be called inside onecore to write to the progress file for the current core and repetition, and an argument core which informs onecore which sequential core number (batch number) it is processing. When calling showprogress inside onecore, the arguments, in order, must be the integer value of the repetition to be noted, the number of reps, core, an optional 4th argument other that can contain a single character string to add to the output, and an optional 5th argument pr. You can set pr=FALSE to suppress printing and have showprogress return the file name for holding progress information if you want to customize printing.

If any of the objects appearing as list elements produced by onecore are multi-dimensional arrays, you must specify an integer value for along. This specifies to the abind package abind function the dimension along which to bind the arrays. For example, if the first dimension of the array corresponding to repetitions, you would specify along=1. All arrays present must use the same along unless along is a named vector and the names match elements of the simulation result object. Set simplify=FALSE if you don't want the result simplified if onecore produces only one list element. The default returns the first (and only) list element rather than the list if there is only one element.

See here for examples.

Value

result from combining all the parallel runs, formatting as similar to the result produced from one run as possible

Author(s)

Frank Harrell


Hmisc documentation built on June 22, 2024, 12:19 p.m.