Progressively

Add functions useful for showing progress bar in R packages purrr and furrr, using package progressr.

Works well in jupyter/jupyterlab where handlers(global = TRUE) in progressr gives error, and .progress=TRUE in purrr cannot remove previous bar.

Usage

# install.packages(c('purrr', 'furrr', 'progressr'))
# mamba install r-purrr r-furrr r-progressr
source('progressively.R')
plan(multicore, workers = availableCores()%/%2)
fvmap(1:200, ~Sys.sleep(.x/100))
# |============================================                          | 126/200         

Functions

• base purrr and furrr function (abbreved):
  • map, map2, imap and pmap
  • fmap, fmap2, fimap and fpmap
• prefix with "v": add progress bar
  • vmap, vmap2, vimap and vpmap
  • fvmap, fvmap2, fvimap and fvpmap
• suffix with "s": simplified result (actually same as "_vec" suffix in purrr)
  • maps, map2s, imaps and pmaps
  • fmaps, fmap2s, fimaps and fpmaps
• prefix with "v" and suffix with "s":
  • vmaps, vmap2s, vimaps and vpmaps
  • fvmaps, fvmap2s, fvimaps and fvpmaps

Explanation

• default handler registered for progressr is handlers(myhandler_txtprogressbar(style = 4L, file = "", intrusiveness = 1, clear = T)), whose parameters can be changed.
  • style 4 is a new added one that shows the exact iteration times instead of percent.
• parameter rp in "fv" series: since updating progress bar has performance loss especially when length of .x is large, set rp to a higher value allows only updating with a 1/rp possibility.
  • e.g. future_map(1:100000, ~Sys.sleep(0.001)) with 10 cores could spend 10s, but with fvmap it spends several minutes. However, setting rp=100, i.e. fvmap(1:100000, ~Sys.sleep(0.001), rp=100), only spend 14s, with (almost accurate) progress bar.
  • (similar to parameter intrusiveness in progressr, but faster.)
• change the chunk_size to 1 if the length of input list is less than 256 in furrr functions, which is beneficial for very varying sizes of input list.