Adding eachslice for AbstractDimStack

[sethaxen]

I know #418 will add post-1.9 support for eachslice(::AbstractDimStack; dims), but I wonder if we could right now add support that would work for pre-1.9:

julia> using DimensionalData

julia> function Base.eachslice(ds::AbstractDimStack; dims)
           slicedims = Dimensions.dims(ds, dims)
           return (view(ds, d...) for d in DimIndices(slicedims))
       end;

julia> ds = DimStack((
           x=DimArray(randn(2, 3, 4), (Dim{:a}(5:6), :b, :c)),
           y=DimArray(randn(2, 3, 5), (Dim{:a}(5:6), :b, :d))
       ));

julia> collect(eachslice(ds; dims=:a))[1] |> refdims
Dim{:a} Sampled{Int64} 5:5 ForwardOrdered Regular Points

julia> collect(eachslice(ds; dims=:a))[2] |> refdims
Dim{:a} Sampled{Int64} 6:6 ForwardOrdered Regular Points

As a bonus, unlike the version in Base, this already supports multiple dimensions.

julia> collect(eachslice(ds; dims=(:a, :b)))[1] |> refdims
Dim{:a} Sampled{Int64} 5:5 ForwardOrdered Regular Points,
Dim{:b}

julia> collect(eachslice(ds; dims=(:a, :b)))[2] |> refdims
Dim{:a} Sampled{Int64} 6:6 ForwardOrdered Regular Points,
Dim{:b}

julia> collect(eachslice(ds; dims=(:a, :b)))[3] |> refdims
Dim{:a} Sampled{Int64} 5:5 ForwardOrdered Regular Points,
Dim{:b}

julia> eachslice(ds; dims=(:a, :b)) |> collect |> size
(2, 3)

julia> eachslice(ds; dims=(:a, :b)) |> collect |> typeof
DimArray{DimStack{NamedTuple{(:x, :y), Tuple{SubArray{Float64, 1, Array{Float64, 3}, Tuple{Int64, Int64, Base.Slice{Base.OneTo{Int64}}}, true}, SubArray{Float64, 1, Array{Float64, 3}, Tuple{Int64, Int64, Base.Slice{Base.OneTo{Int64}}}, true}}}, Tuple{Dim{:c, DimensionalData.Dimensions.LookupArrays.NoLookup{Base.OneTo{Int64}}}, Dim{:d, DimensionalData.Dimensions.LookupArrays.NoLookup{Base.OneTo{Int64}}}}, Tuple{Dim{:a, DimensionalData.Dimensions.LookupArrays.Sampled{Int64, UnitRange{Int64}, DimensionalData.Dimensions.LookupArrays.ForwardOrdered, DimensionalData.Dimensions.LookupArrays.Regular{Int64}, DimensionalData.Dimensions.LookupArrays.Points, DimensionalData.Dimensions.LookupArrays.NoMetadata}}, Dim{:b, DimensionalData.Dimensions.LookupArrays.NoLookup{UnitRange{Int64}}}}, NamedTuple{(:x, :y), Tuple{Tuple{Dim{:c, Colon}}, Tuple{Dim{:d, Colon}}}}, DimensionalData.Dimensions.LookupArrays.NoMetadata, NamedTuple{(:x, :y), Tuple{DimensionalData.Dimensions.LookupArrays.NoMetadata, DimensionalData.Dimensions.LookupArrays.NoMetadata}}}, 2, Tuple{Dim{:a, DimensionalData.Dimensions.LookupArrays.Sampled{Int64, UnitRange{Int64}, DimensionalData.Dimensions.LookupArrays.ForwardOrdered, DimensionalData.Dimensions.LookupArrays.Regular{Int64}, DimensionalData.Dimensions.LookupArrays.Points, DimensionalData.Dimensions.LookupArrays.NoMetadata}}, Dim{:b, DimensionalData.Dimensions.LookupArrays.NoLookup{Base.OneTo{Int64}}}}, Tuple{}, Matrix{DimStack{NamedTuple{(:x, :y), Tuple{SubArray{Float64, 1, Array{Float64, 3}, Tuple{Int64, Int64, Base.Slice{Base.OneTo{Int64}}}, true}, SubArray{Float64, 1, Array{Float64, 3}, Tuple{Int64, Int64, Base.Slice{Base.OneTo{Int64}}}, true}}}, Tuple{Dim{:c, DimensionalData.Dimensions.LookupArrays.NoLookup{Base.OneTo{Int64}}}, Dim{:d, DimensionalData.Dimensions.LookupArrays.NoLookup{Base.OneTo{Int64}}}}, Tuple{Dim{:a, DimensionalData.Dimensions.LookupArrays.Sampled{Int64, UnitRange{Int64}, DimensionalData.Dimensions.LookupArrays.ForwardOrdered, DimensionalData.Dimensions.LookupArrays.Regular{Int64}, DimensionalData.Dimensions.LookupArrays.Points, DimensionalData.Dimensions.LookupArrays.NoMetadata}}, Dim{:b, DimensionalData.Dimensions.LookupArrays.NoLookup{UnitRange{Int64}}}}, NamedTuple{(:x, :y), Tuple{Tuple{Dim{:c, Colon}}, Tuple{Dim{:d, Colon}}}}, DimensionalData.Dimensions.LookupArrays.NoMetadata, NamedTuple{(:x, :y), Tuple{DimensionalData.Dimensions.LookupArrays.NoMetadata, DimensionalData.Dimensions.LookupArrays.NoMetadata}}}}, DimensionalData.NoName, DimensionalData.Dimensions.LookupArrays.NoMetadata}

[rafaqz]

Good idea! Thats such a concise implementation.

[sethaxen]

Is there a good reason to not do something similar for AbstractDimArrays? Currently they only support single dimensions.

[rafaqz]

No I actually assumed you meant both.