I noticed that group_bootstraps is fairly slow compared to bootstraps.

As a workaround I am doing nesting and unnesting which is a lot faster.

Are version 1 and version 2 equivilant or am I missing sth?

library(tidyr)
library(rsample)

dat <- tibble(x = rep(1:1000, 2), y = 1:2000) 

f <- function(df, column){
  tibble(
    "estimate" = mean(pull(df, {{column}})),
    "term" = "mean"
  )
}


# Version 1 using rsample::group_bootstraps
start_time <- Sys.time()

dat %>%
  group_bootstraps(group = x, times = 10) %>%
  mutate(mean_stats = purrr::map(splits, ~ f(analysis(.), y))) %>% 
  int_pctl(mean_stats)

print(difftime(Sys.time(), start_time, units = "secs"))

# Version 2 using tidyr nest and unnest
start_time <- Sys.time()

dat %>%
  nest(.by = x) %>%
  bootstraps(times = 10) %>%
  mutate(mean_stats = purrr::map(splits, ~ f(unnest(analysis(.), data), y))) %>% 
  int_pctl(mean_stats)

print(difftime(Sys.time(), start_time, units = "secs"))

• version 1 takes more than 9 seconds on my machine
• version 2 takes less than 0.5 seconds on my machine