R thing of the day: Speed up base R using the collapse package

1. Begin your script by setting the collapse_mask option. See help('collapse-options') for more options, e.g. "manip".
2. Needs to come after library(dplyr) for collapse_mask to work

I’m a huge fan of the tidyverse, but I also adore data.table (especially when I need/want an extra speedup 🏎️). My go-to is fread to load CSV files, and I often consider trying out vroom more often as it seems to be competitive. For data wrangling, packages like dtplyr and tidyfast are also compelling alternatives as they bring the efficiency of data.table with the familiar tidyverse syntax.

I often mean to give the collapse package 📦 by Sebastian Krantz a go, but I tend to slip into familiar ways before I ever remember to do so. This is unfortunate because it seems that it can seamlessly augment and speed up my current practice, be it alongside dplyr, data.table, or even when making maps using sf.

Grant McDermott recently tweeted how one can simply add a couple lines of code to get the benefits of collapse without really any additional code changes:

TL;DR many R users will get a major speedup simply by adding

options(collapse_mask = "manip")
library(collapse)

to the top of their scripts. (First line could even go into your ~/.Rprofile file for a permanent change.) Everything else remains unchanged. https://t.co/Mt3FboE8uC

— Grant McDermott (@grant_mcdermott) February 15, 2022

He also shared a gist comparing dplyr, data.table and the power of collapse using the same ol' dplyr code. I attempted to run the same benchmark to see if I too got the same performance bump using collapse. I had trouble installing the microbenchmark 📦, so used rbenchmark instead.

## Context: https://twitter.com/grant_mcdermott/status/1493400952878952448

options(collapse_mask = "all")

library(dplyr)
library(data.table)
library(collapse) # Needs to come after library(dplyr) for collapse_mask to work

flights = fread('https://raw.githubusercontent.com/Rdatatable/data.table/master/vignettes/flights14.csv')

vars = c('dep_delay', 'arr_delay', 'air_time', 'distance', 'hour')

## Note we explicitly call dplyr::<function> for the 1st line in this benchmark, 
## since we've masked the regular dplyr operations with their collapse 
## equivalents (i.e. 2nd line). 
library(rbenchmark)
benchmark(
  dplyr      = flights %>% dplyr::group_by(month, day, origin, dest) %>% dplyr::summarise(across(vars, sum)),
  collapse   = flights %>% group_by(month, day, origin, dest) %>% summarise(across(vars, sum)),
  data.table = flights[, lapply(.SD, sum), by=.(month, day, origin, dest), .SDcols=vars],
  replications = 20
  )
#        test replications elapsed relative user.self sys.self user.child  sys.child
#2   collapse           20   0.275    1.062     0.263    0.012          0         0
#3 data.table           20   0.259    1.000     5.502    0.019          0         0
#1      dplyr           20  32.587  125.819    31.736    1.251          0         0

🚀🚀🚀