For large datasets, using fn_once
or fn_chain
may take many minutes to run. These types of functions use the furrr package
for the option to use parallel processing and speed things up. To initialize parallel processing, use
plan(multisession)
or plan(multicore)
(depending on your operating system) prior to your piped code with the
fn_once
or fn_chain
functions. Note, parallel processing is best used when your code block takes more than a minute to run,
shorter run times will not benefit from parallel processing.