How to make your data type reducible
Let's see how to make a vector-of-vector a reducible collection; i.e., a type that can be fed to mapfoldl.
struct VecOfVec{T}
vectors::Vector{Vector{T}}
endWe need next and complete to invoke the reducing function rf and @return_if_reduced to support early termination.
using Transducers
using Transducers: next, complete, @return_if_reducedSupporting mapfoldl and similar only requires Transducers.__foldl__:
function Transducers.__foldl__(rf, val, vov::VecOfVec)
for vector in vov.vectors
for x in vector
val = next(rf, val, x)
@return_if_reduced complete(rf, val)
end
end
return complete(rf, val)
endNote that it's often a good idea to implement Base.eltype:
Base.eltype(::VecOfVec{T}) where {T} = TIt can be then used as the input to the transducers:
vov = VecOfVec(collect.([1:n for n in 1:3]))
collect(Map(identity), vov)6-element Array{Int64,1}:
1
1
2
1
2
3Transducers.@return_if_reduced above is used to support terminating transducer like Take.
collect(Take(3), vov)3-element Array{Int64,1}:
1
1
2More complex example:
collect(PartitionBy(isequal(1)) |> Zip(Map(copy), Map(sum)), vov)4-element Array{Tuple{Array{Int64,1},Int64},1}:
([1, 1], 2)
([2], 2)
([1], 1)
([2, 3], 5)Notice that writing Transducers.__foldl__ is very straightforward comparing to how to define an iterator:
function Base.iterate(vov::VecOfVec, state=nothing)
if state === nothing
i, j = 1, 1
else
i, j = state
end
i > length(vov.vectors) && return nothingIf j is in bound, we are iterating the same sub-vector:
vi = vov.vectors[i]
if j <= length(vi)
return vi[j], (i, j + 1)
endOtherwise, find the next non-empty sub-vector and start iterating it:
for k in i + 1:length(vov.vectors)
vk = vov.vectors[k]
if !isempty(vk)
return vk[1], (k, 2) # i=k, j=2
end
end
return nothing
end
Base.length(vov::VecOfVec) = sum(length, vov.vectors)
collect(vov)6-element Array{Int64,1}:
1
1
2
1
2
3This page was generated using Literate.jl.