Investigate using furrr package (perform purrr functions in parallel)

[jack89roberts]

Has anyone come across/considered furrr before? It implements purrr functions in parallel. May be low hanging fruit to get performance gains wherever purrr::map is used in the package.

Also claims to work on Windows, r.e. #53

Repo: https://github.com/DavisVaughan/furrr

E.g. quick test using the VRPINS graphs (in inst/extdat) gives at least 7 times speedup of netdis_centred_graphlet_counts on my 8 core iMac.

Original (takes ~3.5 seconds):

centred_counts <- purrr::map(query_graphs, netdis_centred_graphlet_counts,
                             max_graphlet_size = max_graphlet_size, 
                             neighbourhood_size = neighbourhood_size,
                             expected_ego_count_fn = expected_count_fn)

With furrr (takes ~0.5 seconds):

library(furrr)
plan(multiprocess)

centred_counts <- future_map(query_graphs, netdis_centred_graphlet_counts,
                             max_graphlet_size = max_graphlet_size, 
                             neighbourhood_size = neighbourhood_size,
                             expected_ego_count_fn = expected_count_fn)

NB: The default plan is to run sequentially, in which case it runs the same as purrr::map. Recommendation is to leave setting the plan to the user: https://www.rdocumentation.org/packages/future/versions/1.14.0/topics/plan