Parallelize 'tm' 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(tm)

data("crude")
m <- tm_map(crude, content_transformer(tolower)) |> futurize()

Introduction

This vignette demonstrates how to use this approach to parallelize tm functions such as tm_map().

The tm package provides a variety of text-mining methods. The tm_map() function applies transformations to a corpus of text documents, and TermDocumentMatrix() constructs document-term matrices. When working with large corpora, these operations benefit greatly from parallelization.

Example: Transforming a text corpus

The tm_map() function applies a transformation to each document in a corpus:

library(tm)

## Load the crude oil news corpus holding 20 documents
data("crude")

## Convert all text to lowercase
m <- tm_map(crude, content_transformer(tolower))

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

library(tm)
library(futurize)
plan(multisession)

data("crude")
m <- tm_map(crude, content_transformer(tolower)) |> futurize()

This will distribute the document transformations 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 tm functions are supported by futurize():

• tm_map()
• tm_index()
• TermDocumentMatrix()

Without futurize: Manual PSOCK cluster setup

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

library(tm)
library(parallel)

data("crude")

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

## Configure tm to use the cluster
old_engine <- tm_parLapply_engine()
tm_parLapply_engine(function(X, FUN, ...) parLapply(cl, X, FUN, ...))

## Transform the corpus in parallel
m <- tm_map(crude, content_transformer(tolower))

## Restore the old engine and tear down the cluster
tm_parLapply_engine(old_engine)
stopCluster(cl)

This requires you to manually create a cluster, configure tm’s internal parallel engine, and remember to restore the engine and tear down the cluster when done. 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, 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().