Performance of some variants of group_by() + summarise()

grouped_summarise_benchmark.R

reprex::reprex_info()
#> Created by the reprex package v0.1.1.9000 on 2018-01-22

library(dplyr, warn.conflicts = FALSE)
data("flights", package = "nycflights13")

microbenchmark::microbenchmark(
  summarise_only = flights %>%
    group_by(year, month, day, origin) %>%
    summarise(dep_delay_avg = mean(dep_delay),
              arr_delay_avg = mean(arr_delay),
              total_delay_avg = dep_delay_avg + arr_delay_avg),
  
  summarise_mutate = flights %>%
    group_by(year, month, day, origin) %>%
    summarise(dep_delay_avg = mean(dep_delay),
              arr_delay_avg = mean(arr_delay)) %>%
    mutate(total_delay_avg = dep_delay_avg + arr_delay_avg),
  
  summarise_mutate_ungroup = flights %>%
    group_by(year, month, day, origin) %>%
    summarise(dep_delay_avg = mean(dep_delay),
              arr_delay_avg = mean(arr_delay)) %>%
    ungroup() %>%
    mutate(total_delay_avg = dep_delay_avg + arr_delay_avg)
)
#> Unit: milliseconds
#>                      expr      min       lq     mean   median       uq
#>            summarise_only 68.69892 73.66465 83.79183 77.52717 83.25379
#>          summarise_mutate 53.19234 55.92301 61.14978 58.69358 64.01526
#>  summarise_mutate_ungroup 42.87608 45.61051 57.87704 48.22187 52.51579
#>       max neval
#>  242.5709   100
#>  110.4376   100
#>  640.3410   100