The following functions take advantage of the Future framework: NormalizeData, ScaleData, JackStraw, FindMarkers, FindIntegrationAnchors, FindClusters.
In general, R-language, has other options to execute tasks in parallel. Future’s focused on having a ‘simple’ interface for programmers, and support multiple platforms (Linux, Windows, MacOSX).
If you like base R ‘functional’ programming, using
lapply(), then you’d be happy to learn that there is a
corresponding future_lapply() in the
future.apply package. Same goes if you like tidyverse
purrr::map(), there is a corresponding
future_map() in the furrr package! Last, but
not least, if you prefer foreach() from foreach, then
doFuture provides a backend adapter.
These are set with future::plan(). Among others, we
have:
multisession: Resolves futures asynchronously (in parallel) in separate R sessions running in the background on the same machine.
multicore: Resolves futures asynchronously (in parallel) in separate forked R processes running in the background on the same machine. (Doesn’t work on MS Windows.)
Each worker (session/ core) needs access to certain global variables,
these ‘exportation’ is done automatically, and it uses a default size of
500 MiB. If the objects don’t fit, we’d get an error. Use
options(future.globals.maxSize = 2000 * 1024^2) to increase
this limit to 2 GiB. Be advised, increasing this in excess can be
dangerous, specially if you plan to use way too many workers.
Note: you can execute free -mh, top, and
other system monitoring utilities from within the ‘Terminal’ tab inside
the RStudio IDE. If such tab is not already open, next to ‘Console’, you
will need to go into ‘Tools’ Menu.