JuliennedArrays.jl

Align(slices, alongs::Int...)

Alternative syntax: alongs is which dimensions will be taken up by the inner arrays.

julia> using JuliennedArrays

julia> input = reshape(1:8, 2, 2, 2)
2×2×2 reshape(::UnitRange{Int64}, 2, 2, 2) with eltype Int64:
[:, :, 1] =
 1  3
 2  4

[:, :, 2] =
 5  7
 6  8

julia> slices = Slices(input, 1, 3)
2-element Slices{SubArray{Int64, 2}, 1}:
 [1 5; 2 6]
 [3 7; 4 8]

julia> Align(slices, 1, 3)
2×2×2 Align{Int64, 3} with eltype Int64:
[:, :, 1] =
 1  3
 2  4

[:, :, 2] =
 5  7
 6  8

You must include one along for each inner dimension. alongs should also be strictly increasing.

julia> Align(slices, 1)
ERROR: ArgumentError: (1,) is not of length inner dimensions (2).
[...]

Julia can infer the result if alongs is constant.

julia> using Test: @inferred

julia> align_1_3(x) = Align(x, 1, 3);

julia> @inferred align_1_3(slices)
2×2×2 Align{Int64, 3} with eltype Int64:
[:, :, 1] =
 1  3
 2  4

[:, :, 2] =
 5  7
 6  8 

slice_axes is the axes of one slice. If slices is empty, you can specify slice_axes manually.

julia> Align([rand(3) for _ in 1:0], 1)
ERROR: BoundsError: attempt to access 0-element Vector{Vector{Float64}} at index [1]
[...]

julia> Align([rand(3) for _ in 1:0], 1; slice_axes = axes(rand(3)))
3×0 Align{Float64, 2} with eltype Float64

Align(slices, alongs::TypedBool...; slice_axes = axes(first(slices)))

Align an array of arrays, all with the same size.

alongs, made of True and False objects, shows which dimensions will be taken up by the inner arrays. Inverse of Slices.

julia> using JuliennedArrays

julia> slices = [[1, 2], [3, 4]];

julia> aligned = Align(slices, False(), True())
2×2 Align{Int64, 2} with eltype Int64:
 1  2
 3  4

julia> aligned[1, :] == slices[1]
true

julia> aligned[1, 1] = 0;

julia> slices
2-element Vector{Vector{Int64}}:
 [0, 2]
 [3, 4]

Will throw an error if you try to align slices with different axes. Use @inbounds to skip this check.

julia> unequal = [[1], [1, 2]];

julia> Align(unequal, False(), True())
ERROR: ArgumentError: Slice [1, 2] does not have slice_axes (Base.OneTo(1),)
[...]

slice_axes is the axes of one slice. If slices is empty, you can specify slice_axes manually.

julia> Align([rand(3) for _ in 1:0], True(), False())
ERROR: BoundsError: attempt to access 0-element Vector{Vector{Float64}} at index [1]
[...]

julia> Align([rand(3) for _ in 1:0], True(), False(); slice_axes = axes(rand(3)))
3×0 Align{Float64, 2} with eltype Float64

struct False

Typed false

Slices(whole, alongs::TypedBool...)

Slice whole into views.

alongs, made of True and False objects, shows which dimensions will be replaced with : when slicing.

julia> using JuliennedArrays

julia> whole = [1 2; 3 4];

julia> slices = Slices(whole, False(), True())
2-element Slices{SubArray{Int64, 1}, 1}:
 [1, 2]
 [3, 4]

julia> slices[1] == whole[1, :]
true

julia> slices[1] = [2, 1];

julia> whole
2×2 Matrix{Int64}:
 2  1
 3  4

julia> larger = rand(5, 5, 5);

julia> larger_slices = Slices(larger, True(), False(), False());

julia> size(first(larger_slices))
(5,)

You must include one along for each dimension of whole.

julia> using Test: @test_throws

julia> @test_throws MethodError Slices(whole, True());

Slices(whole, alongs::Int...)

Alternative syntax: alongs is which dimensions will be replaced with : when slicing.

julia> using JuliennedArrays

julia> input = reshape(1:8, 2, 2, 2)
2×2×2 reshape(::UnitRange{Int64}, 2, 2, 2) with eltype Int64:
[:, :, 1] =
 1  3
 2  4

[:, :, 2] =
 5  7
 6  8

julia> s = Slices(input, 1, 3)
2-element Slices{SubArray{Int64, 2}, 1}:
 [1 5; 2 6]
 [3 7; 4 8]

julia> map(sum, s)
2-element Vector{Int64}:
 14
 22

No along should be greater than the number of dimensions of whole. alongs should also be strictly increasing.

julia> Slices(input, 4)
ERROR: ArgumentError: 4, a dimension number, is out of bounds or out of order.
[...]

You can infer the result if the dimensions are constant.

julia> using Test: @inferred

julia> slices_1_3(x) = Slices(x, 1, 3);

julia> @inferred slices_1_3(input)
2-element Slices{SubArray{Int64, 2}, 1}:
 [1 5; 2 6]
 [3 7; 4 8]

struct True

Typed true