LightNews — News, ranked by scholars

Andrew Heiss

@andrew.heiss.phd

#rstats question! Now that {purrr} has built-in support for parallelization with {mirai} and in_parallel(), should we still be using {furrr} and future_map()? Or is the best practice going forward going to be purrr::in_parallel()?

Parallel processing in purrr 1.1.0

The functional programming toolkit for R gains new capabilities for parallel processing and distributed computing using mirai.

www.tidyverse.org

August 27, 2025 at 2:21 PM

Hadley Wickham

@hadley.nz

purrr 1.2.0 out now — mostly removing long deprecated functions but a few small performance nad parallel processing improvements. Learn more at www.tidyverse.org/blog/2025/11... #rstats

purrr 1.2.0

This release tightens up the package by removing long-deprecated functions, making `map_chr()` and predicate functions more type-safe, and requiring a newer version of carrier to make `in_parallel()`...

www.tidyverse.org

November 4, 2025 at 2:10 PM

Charlie Gao

@shikokuchuo.net

Bleeding edge update for the #tidyverse purrr package with even more seamless #rstats parallel maps.

Introducing our shiniest new adverb: `in_parallel()`. Just wrap your function to take advantage of blazing fast parallel processing via mirai.

pak::pak("tidyverse/purrr")

purrr.tidyverse.org/dev/

Functional Programming Tools

A complete and consistent functional programming toolkit for R.

purrr.tidyverse.org

June 13, 2025 at 3:32 PM

Ambassador Frank Hull

@frankiethull.bsky.social

purrr::in_parallel is all you need

July 11, 2025 at 4:34 PM

Jordi Rosell

@jrosell.bsky.social

From .parallel to in_parallel() 👏

June 13, 2025 at 7:16 PM

Itô, Hiroki

@ito4303.bsky.social

Rのpurrrで、miraiパッケージ利用のin_parallelを使ってmapを並列化するというのをためしてみました（←事情を知らない人には訳がわからん文だな）。

ito4303.sakura.ne.jp/posts/2026-0...

Rのpurrr::mapをmiraiで並列化 – 伊東生態統計研究室

ito4303.sakura.ne.jp

January 29, 2026 at 11:49 PM

Andrew Heiss

@andrew.heiss.phd

lol I'm going for it :)

$#rstats code to seed hack my last name: library(tidyverse) mirai::daemons(9) possible_seeds <- data.frame(seed = -1e8:1e8) |> mutate( my_name = map_lgl( seed, in_parallel(\(s) { withr::with_seed(s, { paste0(LETTERS[sample(26, 5, replace = TRUE)], collapse = "") == "HEISS" }) }) ) )$

April 16, 2026 at 12:04 AM

Andrew Heiss

@andrew.heiss.phd

Oh awesome, this is good to know. I was just converting a future_map thing into in_parallel and ran into the need-to-add-namespace-prefix-or-load-libraries-in-the-function thing and it was a little surprising/annoying

But it's neat to have it built-in to purrr and I'll probably still convert!

$fit_logitr_purrr <- function(.split, ...) { # Have to load libraries or prefix everything library(dplyr) library(tidyr) library(logitr) .df <- rsample::analysis(.split) |> # Assign new unique respondent IDs (since some will be repeated through the # bootstrapping process) mutate(resp_id = row_number()) |> # Unnest the respondent-specific data unnest(data) |> group_by(resp_id, question) |> mutate(choice_id = cur_group_id()) |> ungroup() # Fit logitr model model_logitr <- logitr( data = .df, outcome = "choice", obsID = "choice_id", pars = c("price", "packaging", "flavor") ) } mirai::daemons(8) boot_results_purrr <- bootstrapped_stickers_logitr |> slice(1:10) |> mutate( model = map( splits, in_parallel(\(x) fit_logitr_purrr(x), fit_logitr_purrr = fit_logitr_purrr), .progress = TRUE ) ) mirai::daemons(0)$ $fit_logitr_furrr <- function(.split, ...) { .df <- rsample::analysis(.split) |> # Assign new unique respondent IDs (since some will be repeated through the # bootstrapping process) mutate(resp_id = row_number()) |> # Unnest the respondent-specific data unnest(data) |> group_by(resp_id, question) |> mutate(choice_id = cur_group_id()) |> ungroup() # Fit logitr model model_logitr <- logitr( data = .df, outcome = "choice", obsID = "choice_id", pars = c("price", "packaging", "flavor") ) } library(furrr) plan(multisession, workers = 8) boot_results_furrr <- bootstrapped_stickers_logitr |> slice(1:10) |> mutate( model = future_map( splits, \(x) fit_logitr_furrr(x), .progress = TRUE, .options = furrr_options(seed = TRUE) ) )$

August 27, 2025 at 3:09 PM

tj mahr 🤘

@tjmahr.com

i could have written 8 in_parallel() calls with a risk of a copy-paste or naming inconsistency and i could have just 2 calls

$data_straps1 <- d |> rsample::bootstraps(times = 2000, apparent = TRUE) |> mutate( # Extract the analysis-ready dataframe for each resample data = map(splits, \(x) as.data.frame(rsample::analysis(x))) ) # Avoid copy-pasting errors or inconsistencies by writing the 2 generic function # calls and iterating each pair of variables through them pairs <- tribble( ~x, ~y, "VSS4", "GMFCS4", "VSS14", "GMFCS14", "VSS4", "VSS14", "GMFCS4", "GMFCS14" ) # Use 10 parallel workers to compute correlations on the resamples mirai::daemons(10) # in_parallel(f) requires f to be a full function definition with no nonlocal # variables. (everything the function needs should be in the function or the # data handed to it.) this allows the function to be carried out in separate R # sessions for parallelization. for (i in seq_len(nrow(pairs))) { x <- pairs$x[i] y <- pairs$y[i] polycor_name <- paste0("polycor_", x, "_", y) spearman_name <- paste0("spearman_", x, "_", y) data_straps1[[polycor_name]] <- data_straps1$data |> map_dbl( # `!! x` will replace `!! x` with the value of `x` at the last second so # it will hardcode column names for those separate R sessions. in_parallel(function(d) polycor::polychor(d[[!! x]], d[[!! y]], ML = TRUE)), .progress = polycor_name ) data_straps1[[spearman_name]] <- data_straps1$data |> map_dbl( in_parallel(function(d) cor(d[[!! x]], d[[!! y]], method = "spearman")), .progress = spearman_name ) }$

March 4, 2026 at 8:33 PM