Parallelize 'mgcv' 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(mgcv)

## Adopted from example("bam", package = "mgcv")
dat <- gamSim(1, n = 25000, dist = "normal", scale = 20)
bs <- "cr"
k <- 12

b <- bam(y ~ s(x0, bs = bs) + s(x1, bs = bs) + s(x2, bs = bs, k = k) +
             s(x3, bs = bs), data = dat) |> futurize()

Introduction

This vignette demonstrates how to use this approach to parallelize mgcv functions such as bam().

The mgcv package is one of the “recommended” packages in R. It provides methods for fitting Generalized Additive Models (GAMs). The bam() function can be used to fit GAMs for massive datasets (“Big Additive Models”) with many thousands of observations, making it an excellent candidate for parallelization.

Example: Fitting a Big Additive Model

The bam() function supports parallel processing by setting up a parallel cluster and passing it as argument cluster. This is abstracted away by futurize:

library(mgcv)

## Adopted from example("bam", package = "mgcv")
dat <- gamSim(1, n = 25000, dist = "normal", scale = 20)
bs <- "cr"
k <- 12

b <- bam(y ~ s(x0, bs = bs) + s(x1, bs = bs) + s(x2, bs = bs, k = k) +
             s(x3, bs = bs), data = dat) |> futurize()

This will distribute the calculations 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)

Supported Functions

The following mgcv functions are supported by futurize():

• bam()
• predict() for ‘bam’

Without futurize: Manual PSOCK cluster setup

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

library(mgcv)
library(parallel)

## Adopted from example("bam", package = "mgcv")
dat <- gamSim(1, n = 25000, dist = "normal", scale = 20)
bs <- "cr"
k <- 12

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

## Fit the model in parallel
b <- bam(y ~ s(x0, bs = bs) + s(x1, bs = bs) + s(x2, bs = bs, k = k) +
             s(x3, bs = bs), data = dat, cluster = 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().