Parallelize 'parameters' functions

+ =

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(parameters)

model <- lm(mpg ~ wt, data = mtcars)
fit <- bootstrap_model(model, iterations = 1000) |> futurize()

Introduction

This vignette demonstrates how to use this approach to parallelize parameters functions, such as bootstrap_model() and bootstrap_parameters().

The parameters package (part of the easystats ecosystem) provides utilities for processing and summarizing statistical models. The bootstrap_model() function generates a distribution of model estimates by refitting the model multiple times using bootstrapped samples. This process can be computationally demanding, especially for complex models or a large number of iterations. Since each bootstrap iteration is independent, it is a perfect candidate for parallelization.

Example: Bootstrapping a linear model

Consider a linear model where we want to obtain bootstrapped estimates of the coefficients:

library(parameters)

model <- lm(mpg ~ wt + cyl, data = mtcars)

## Generate 1000 bootstrap replicates (sequentially)
boot_dist <- bootstrap_model(model, iterations = 1000)

To parallelize this using futurize, simply pipe the call to futurize():

library(futurize)
library(parameters)

model <- lm(mpg ~ wt + cyl, data = mtcars)

## Generate 1000 bootstrap replicates (in parallel)
boot_dist <- bootstrap_model(model, iterations = 1000) |> futurize()

This will distribute the bootstrap iterations across the available parallel workers, given that we have set up a parallel backend, e.g.

plan(multisession)

Example: Bootstrapped parameters summary

The bootstrap_parameters() function is a higher-level wrapper that calls bootstrap_model() and then summarizes the results. It can also be parallelized in the same way:

library(futurize)
plan(multisession)
library(parameters)

model <- lm(mpg ~ wt + cyl, data = mtcars)
boot_params <- bootstrap_parameters(model, iterations = 1000) |> futurize()

Supported Functions

The following parameters functions are supported by futurize():

bootstrap_model() with seed = TRUE as the default
bootstrap_parameters() with seed = TRUE as the default

Without futurize: Manual PSOCK cluster setup

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

library(parameters)
library(parallel)

model <- lm(mpg ~ wt + cyl, data = mtcars)

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

## Run bootstrapping in parallel
boot_dist <- bootstrap_model(model, iterations = 1000, 
                             parallel = "snow", n_cpus = ncpus, 
                             cluster = cl)

## Tear down the cluster
stopCluster(cl)