Parallelize 'boot' functions

The 'boot' image + The 'futurize' hexlogo = The 'future' logo

The futurize package allows you to easily turn sequential code into parallel code by piping the sequential code to the futurize() function. Easy!

TL;DR

library(futurize)
plan(multisession)
library(boot)

ratio <- function(pop, w) sum(w * pop$x) / sum(w * pop$u)
b <- boot(bigcity, statistic = ratio, R = 999, stype = "w") |> futurize()

Introduction

This vignette demonstrates how to use this approach to parallelize boot functions such as boot(), censboot(), and tsboot().

The boot package is one of the "recommended" R packages, meaning it is officially endorsed by the R Core Team, well maintained, and installed by default with R. The package generates bootstrap samples and provides statistical methods around them. Given the resampling nature of bootstrapping, the algorithms are excellent candidates for parallelization.

Example: Bootstrap sampling

The core function boot() produces bootstrap samples of a statistic applied to data. For example, consider the bigcity dataset, which contains populations of 49 large U.S. cities in 1920 (u) and 1930 (x):

library(boot)

## Draw 999 bootstrap samples of the population data. For each
## sample, calculate the ratio of mean-1930 over mean-1920 populations
ratio <- function(pop, w) sum(w * pop$x) / sum(w * pop$u)
b <- boot(bigcity, statistic = ratio, R = 999, stype = "w")

Here boot() evaluates sequentially, but we can easily make it evaluate in parallel by piping to futurize():

library(futurize)
library(boot)

ratio <- function(pop, w) sum(w * pop$x) / sum(w * pop$u)
b <- boot(bigcity, statistic = ratio, R = 999, stype = "w") |> futurize()

This will distribute the 999 bootstrap samples across the available parallel workers, 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)

Example: Time series bootstrap

The tsboot() function generates bootstrap samples from time series data. For example, here we fit autoregressive models to bootstrap replicates of the lynx time series:

library(futurize)
plan(multisession)
library(boot)

## Fit AR models to bootstrap replicates of the lynx time series
lynx_fun <- function(tsb) {
    ar_fit <- ar(tsb, order.max = 25)
    c(ar_fit$order, mean(tsb), tsb)
}

lynx_boot <- tsboot(log(lynx), lynx_fun, R = 99, l = 20, sim = "geom") |> futurize()

Supported Functions

The following boot functions are supported by futurize():

Without futurize: Manual PSOCK cluster setup

For comparison, here is what it takes to parallelize boot() using the parallel package directly, without futurize:

library(boot)
library(parallel)

ratio <- function(pop, w) sum(w * pop$x) / sum(w * pop$u)

## Set up a PSOCK cluster
ncpus <- 4L
cl <- makeCluster(ncpus)

## Run bootstrapping in parallel
b <- boot(bigcity, statistic = ratio, R = 999, stype = "w",
          parallel = "snow", ncpus = ncpus, cl = cl)

## Tear down the cluster
stopCluster(cl)

This requires you to manually create and manage the cluster lifecycle. If you forget to call stopCluster(), or if your code errors out before reaching it, you leak background R processes. You also have to decide upfront how many CPUs to use and what cluster type to use. Switching to another parallel backend, e.g. a Slurm cluster, would require a completely different setup. With futurize, all of this is handled for you - just pipe to futurize() and control the backend with plan().