Exploring duckplyr

duckplyr_exploration.R

# dataset: https://zenodo.org/records/2594012
df = arrow::read_parquet("personal/Avazu/test.parquet") |> 
  tibble::as_tibble()

dim(df) # 4,218,938 X 24

res = 
  bench::mark(
    # using `duckplyr`
    duckplyr = {
      df |> 
      duckplyr::as_duckplyr_df() |> 
      duckplyr::summarise(msp = median(click),
                          .by = c(site_id, device_model, app_domain)
                          ) |> 
      duckplyr::arrange(site_id, device_model, app_domain)
    }
    ,
    # using `dplyr`
    dplyr = {
      df |> 
      dplyr::summarise(msp = median(click),
                          .by = c(site_id, device_model, app_domain)
                       ) |> 
      dplyr::arrange(site_id, device_model, app_domain)
    }
    ,
    # using `tidytable` (internally uses `data.table`)
    tidytable = {
      df |> 
      tidytable::summarise(msp = median(click),
                          .by = c(site_id, device_model, app_domain)
                          ) |> 
      tidytable::arrange(site_id, device_model, app_domain)
    }
    ,
    iterations = 10,
    check = FALSE
  )

res |> 
  dplyr::select(min, median, iter_per_sec = `itr/sec`) |> 
  dplyr::mutate(backend = c("duckplyr", "dplyr", "tidytable")) |> 
  dplyr::relocate(backend)

# on my laptop:
# # A tibble: 3 × 4
#   backend        min   median iter_per_sec
#   <chr>     <bch:tm> <bch:tm>        <dbl>
# 1 duckplyr    3.99ms   4.76ms      215.   
# 2 dplyr        3.89s    4.52s        0.225
# 3 tidytable 380.64ms  395.7ms        2.40

sessionInfo()