Parallelize 'sandwich' functions

Henrik Bengtsson

+ =

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(sandwich)

fm <- lm(dist ~ speed, data = cars)
v <- vcovBS(fm) |> futurize()

Introduction

The sandwich package provides model-agnostic robust covariance matrix estimators.

Example: Clustered bootstrap covariance matrix

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)

Supported Functions

The following sandwich functions are supported by futurize():

• vcovBS() with seed = TRUE as the default
• vcovJK() with seed = TRUE as the default

Without futurize: Manual setup

For 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().