A unified interface for sequential, threaded, and distributed fold
77 Stars
Updated Last
9 Months Ago
Started In
June 2020

Folds: sequential, threaded, and distributed fold interface for Julia

Stable Dev CI

Folds.jl provides a unified interface for sequential, threaded, and distributed folds.

julia> using Folds

julia> Folds.sum(1:10)

julia> Folds.sum(1:10, ThreadedEx())  # equivalent to above

julia> Folds.sum(1:10, DistributedEx())

Iterator transforms and transducers

Most of the functions can be used with iterator comprehensions:

julia> Folds.sum(y for x in 1:10 if isodd(x) for y in 1:x^2)

and Transducers.jl:

julia> using Transducers

julia> 1:10 |> Filter(isodd) |> MapCat(x -> 1:x^2) |> Folds.sum

Package interop

Folds.jl interoperates with various packages. For example, StructArrays.jl can be used as an input and/or output:

julia> using StructArrays

julia> table = StructVector(
           x = [:a, :a, :b, :a, :b],
           y = [1, 2, 3, 4, 5],

julia> Folds.copy(StructVector, (row for row in table if row.x === :a))
3-element StructArray(::Vector{Symbol}, ::Vector{Int64}) with eltype NamedTuple{(:x, :y), Tuple{Symbol, Int64}}:
 (x = :a, y = 1)
 (x = :a, y = 2)
 (x = :a, y = 4)

It also works with OnlineStats.jl by treating it as a reducing function (or more precisely a monoid):

julia> using OnlineStats

julia> Folds.reduce(Mean(), 1:10)
Mean: n=10 | value=5.5

Extensible execution mechanism

Folds.jl decouples the implementation and the execution mechanism ("executor"). Additional executors can be installed from FoldsThreads.jl, FoldsCUDA.jl (rather WIP), and FoldsDagger.jl (very WIP).