Skip to content

Adding a future backend and possibly refactoring parallelization backends #152

@coatless

Description

@coatless

Hi @ecpolley! Hope all is well.

I'm reaching out today to see if you'd be interested in receiving a PR that contains a {future} backend for SuperLearner. future is part of a collection of R packages by Henrik Bengtsson that seeks to unify the underlying parallelization frameworks within R. The reason for preferring a future-oriented backend is it allows us to easily work with remote and cluster-based resources for intense computations.

For a future backend to exist, very few changes are required. For example, the following diff based on a slight code change with mcSuperLearner (based on multicore) seems to work out quite well.

A screenshot showing the diff between a possible `futureSuperLearner()` function and the already existing `mcSuperLearner()` function

Let me know if you would be interested in receiving this backend.

Moreover, glancing at the code differences between snow*(), mc*(), and future*() SuperLearner backends, there is a lot of similar code that could potentially be combined with minimal disruption.

/ cc @SG-TLR

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions