A framework for parallelizing dependent tasks
Author: Jeremy Coyle
What’s delayed?
delayed is an R package that provides a framework for parallelizing dependent tasks in an efficient manner. It brings to R a subset of the functionality implemented in Python’s Dask library. For details on how best to use delayed, please consult the package documentation and vignette online, or do so from within R.
Installation
For standard use, we recommend installing the package from CRAN via
install.packages("delayed")Install the most recent stable release from GitHub via devtools:
devtools::install_github("tlverse/delayed")Example
This minimal example shows how to use delayed to handle dependent computations via chaining of tasks:
library(delayed)
#> delayed v0.4.0: A Framework for Parallelizing Dependent Tasks
# delay a function that does a bit of math
mapfun <- function(x, y) {(x + y) / (x - y)}
delayed_mapfun <- delayed_fun(mapfun)
set.seed(14765)
library(future)
plan(multicore, workers = 2)
const <- 7
# re-define the delayed object from above
delayed_norm <- delayed(rnorm(n = const))
delayed_pois <- delayed(rpois(n = const, lambda = const))
chained_norm_pois <- delayed_mapfun(delayed_norm, delayed_pois)
# compute it using the future plan (multicore with 2 cores)
chained_norm_pois$compute(nworkers = 2, verbose = TRUE)
#> run:0 ready:2 workers:2
#> updating rnorm(n = const) from ready to running
#> run:1 ready:1 workers:2
#> updating rpois(n = const, lambda = const) from ready to running
#> run:2 ready:0 workers:2
#> updating rnorm(n = const) from running to resolved
#> updating rpois(n = const, lambda = const) from running to resolved
#> updating mapfun(x = delayed_norm, y = delayed_pois) from waiting to ready
#> run:0 ready:1 workers:2
#> updating mapfun(x = delayed_norm, y = delayed_pois) from ready to running
#> run:1 ready:0 workers:2
#> updating mapfun(x = delayed_norm, y = delayed_pois) from running to resolved
#> [1] -0.6688907 -1.2691496 -1.1808899 -1.7605806 -0.5992127 -0.6838026 -1.4086257Remark: In the above, the delayed computation is carried out in parallel using the framework offered by the excellent future package and its associated ecosystem.
License
© 2017-2021 Jeremy R. Coyle
The contents of this repository are distributed under the GPL-3 license. See file LICENSE for details.
