+
=
+
=
The futurize package allows you to easily turn sequential code
into parallel code by piping the sequential code to the futurize()
function. Easy!
library(futurize)
plan(multisession)
library(sandwich)
fm <- lm(dist ~ speed, data = cars)
v <- vcovBS(fm) |> futurize()
The sandwich package provides model-agnostic robust covariance matrix estimators.
Example adopted from help("vcovBS", package = "sandwich"):
library(futurize)
plan(multisession)
library(sandwich)
## fit a simple linear model
fm <- lm(dist ~ speed, data = cars)
## bootstrap covariance matrix estimation in parallel
v <- vcovBS(fm, R = 250) |> futurize()
## summary of coefficients with robust standard errors
library(lmtest)
coeftest(fm, vcov = v)
This will parallelize the bootstrap replications, given that we have set up parallel workers, e.g.
plan(multisession)
The built-in multisession backend parallelizes on your local
computer and works on all operating systems. There are other
parallel backends to choose from, including alternatives to
parallelize locally as well as distributed across remote machines,
e.g.
plan(future.mirai::mirai_multisession)
and
plan(future.batchtools::batchtools_slurm)
The following sandwich functions are supported by futurize():
vcovBS() with seed = TRUE as the defaultvcovJK() with seed = TRUE as the defaultFor comparison, here is what it takes to parallelize vcovBS()
using the sandwich package directly, without futurize:
library(sandwich)
library(parallel)
## Fit a simple linear model
fm <- lm(dist ~ speed, data = cars)
## Bootstrap covariance matrix estimation in parallel using cores
v <- vcovBS(fm, R = 250, cores = 4L)
While sandwich has a built-in cores argument, it only supports
local multicore or PSOCK clusters depending on the OS. With
futurize, you can use any future backend, including remote
clusters and HPC environments, just by piping to futurize() and
controlling the backend with plan().