Skip to content

array generation: allow specifying dims as distributions #10

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
Open

array generation: allow specifying dims as distributions #10

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
79e7023
allow specifying dims as distributions in make(Array, ...)
rfourquet Mar 21, 2020
33d0203
allow specifying dims as distributions in rand(..., Array, dims...)
rfourquet Mar 21, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,6 +178,18 @@ julia> julia> rand(Bernoulli(0.2), BitVector, 10) # using the Bernoulli distribu
false
true

julia> rand(Int8, Array, 3, 5) # more explicit syntax than rand(Int8, 3, 5) ...
3×5 Array{Int8,2}:
-4 -110 70 -4 63
-38 4 55 -86 -106
-56 -124 46 118 114

julia> rand(Int8, Array, 1:3, 3:5) # ... but dimensions can be specified with a distribution!
# the size will be computed as (rand(1:3), rand(3:5))
2×5 Array{Int8,2}:
-2 -43 106 -74 18
-117 -97 2 -126 125

julia> rand(1:3, NTuple{3}) # NTuple{3} considered as a container, equivalent to rand(make(NTuple{3}, 1:3))
(3, 3, 1)

Expand Down
53 changes: 33 additions & 20 deletions src/containers.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,33 +95,46 @@ end

macro make_array_container(Cont)
definitions =
[ :(rand(rng::AbstractRNG, $Cont, dims::Dims) = rand(rng, _make_cont(t, dims))),
:(rand( $Cont, dims::Dims) = rand(GLOBAL_RNG, _make_cont(t, dims))),
:(rand(rng::AbstractRNG, $Cont, dims::Integer...) = rand(rng, _make_cont(t, Dims(dims)))),
:(rand( $Cont, dims::Integer...) = rand(GLOBAL_RNG, _make_cont(t, Dims(dims)))),

:(rand(rng::AbstractRNG, X, $Cont, dims::Dims) = rand(rng, _make_cont(t, X, dims))),
:(rand( X, $Cont, dims::Dims) = rand(GLOBAL_RNG, _make_cont(t, X, dims))),
:(rand(rng::AbstractRNG, X, $Cont, dims::Integer...) = rand(rng, _make_cont(t, X, Dims(dims)))),
:(rand( X, $Cont, dims::Integer...) = rand(GLOBAL_RNG, _make_cont(t, X, Dims(dims)))),

:(rand(rng::AbstractRNG, ::Type{X}, $Cont, dims::Dims) where {X} = rand(rng, _make_cont(t, X, dims))),
:(rand( ::Type{X}, $Cont, dims::Dims) where {X} = rand(GLOBAL_RNG, _make_cont(t, X, dims))),
:(rand(rng::AbstractRNG, ::Type{X}, $Cont, dims::Integer...) where {X} = rand(rng, _make_cont(t, X, Dims(dims)))),
:(rand( ::Type{X}, $Cont, dims::Integer...) where {X} = rand(GLOBAL_RNG, _make_cont(t, X, Dims(dims)))),
[ :(rand(rng::AbstractRNG, $Cont, dims...) = rand(rng, _make_cont(t, _default_sampling(t), dims...))),
:(rand( $Cont, dims...) = rand(GLOBAL_RNG, _make_cont(t, _default_sampling(t), dims...))),

:(rand(rng::AbstractRNG, X, $Cont, dims...) = rand(rng, _make_cont(t, X, dims...))),
:(rand( X, $Cont, dims...) = rand(GLOBAL_RNG, _make_cont(t, X, dims...))),

:(rand(rng::AbstractRNG, ::Type{X}, $Cont, dims...) where {X} = rand(rng, _make_cont(t, X, dims...))),
:(rand( ::Type{X}, $Cont, dims...) where {X} = rand(GLOBAL_RNG, _make_cont(t, X, dims...))),

# remove ambiguities with Random
:(rand(rng::AbstractRNG, $Cont, dims::Integer...) = rand(rng, _make_cont(t, _default_sampling(t), dims...))),
:(rand( $Cont, dims::Integer...) = rand(GLOBAL_RNG, _make_cont(t, _default_sampling(t), dims...))),

:(rand(rng::AbstractRNG, X, $Cont, dims::Integer...) = rand(rng, _make_cont(t, X, dims...))),
:(rand( X, $Cont, dims::Integer...) = rand(GLOBAL_RNG, _make_cont(t, X, dims...))),

:(rand(rng::AbstractRNG, ::Type{X}, $Cont, dims::Integer...) where {X} = rand(rng, _make_cont(t, X, dims...))),
:(rand( ::Type{X}, $Cont, dims::Integer...) where {X} = rand(GLOBAL_RNG, _make_cont(t, X, dims...))),
]
esc(Expr(:block, definitions...))
end

_make_cont(args...) = make(args...)

@make_array_container(t::Type{<:Array})
@make_array_container(t::Type{<:BitArray})
@make_array_container(t::AbstractFloat)
_make_cont(t::AbstractFloat, x, dims::Dims{1}) = make(SparseVector, x, t, dims)
_make_cont(t::AbstractFloat, dims::Dims{1}) = make(SparseVector, t, dims)
_make_cont(t::AbstractFloat, x, dims::Dims{2}) = make(SparseMatrixCSC, x, t, dims)
_make_cont(t::AbstractFloat, dims::Dims{2}) = make(SparseMatrixCSC, t, dims)

_make_cont(args...) = make(args...)
_default_sampling(t) = default_sampling(t)

_make_cont(t::AbstractFloat, x, dims::Dims{1}) = make(SparseVector, x, t, dims)
_make_cont(t::AbstractFloat, x, d1::Integer) = make(SparseVector, x, t, d1)
_make_cont(t::AbstractFloat, x, dims::Dims{2}) = make(SparseMatrixCSC, x, t, dims)
_make_cont(t::AbstractFloat, x, d1::Integer, d2::Integer) = make(SparseMatrixCSC, x, t, d1, d2)

_default_sampling(t::AbstractFloat) = default_sampling(SparseVector)

# ambiguities
rand(rng::AbstractRNG, t::AbstractFloat, dims::Dims) = rand(rng, _make_cont(t, _default_sampling(t), dims))
rand( t::AbstractFloat, dims::Dims) = rand( _make_cont(t, _default_sampling(t), dims))


## sets/dicts

Expand Down
12 changes: 12 additions & 0 deletions src/distributions.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,6 +57,18 @@ _deduce_type(::Type{T}, ::Val{true}, ::Type{X}) where {T,X} = T
_deduce_type(::Type{T}, ::Val{false}, ::Type{X}) where {T,X} = T{X}


## Const

# distribution always yielding the same value
struct Const{T} <: Distribution{T}
x::T
end

Base.getindex(c::Const) = c.x

rand(::AbstractRNG, c::SamplerTrivial{<:Const}) = c[][]


## Uniform

abstract type Uniform{T} <: Distribution{T} end
Expand Down
58 changes: 41 additions & 17 deletions src/sampling.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -426,40 +426,57 @@ rand(rng::AbstractRNG, sp::SamplerTag{Cont{S}}) where {S<:Base.ImmutableDict} =
default_sampling(::Type{<:AbstractArray{T}}) where {T} = Uniform(T)
default_sampling(::Type{<:AbstractArray}) = Uniform(Float64)

make(A::Type{<:AbstractArray}, X, d1::Integer, dims::Integer...) = make(A, X, Dims((d1, dims...)))
make(A::Type{<:AbstractArray}, ::Type{X}, d1::Integer, dims::Integer...) where {X} = make(A, X, Dims((d1, dims...)))
_makedim(d::Integer) = Const(Int(d))
_makedim(d) = _makedim(d, gentype(d))
_makedim(d, ::Type{Int}) = d
_makedim(d, T) = throw(ArgumentError("the passed dimension distribution doesn't yield an Int"))

make(A::Type{<:AbstractArray}, dims::Dims) = make(A, default_sampling(A), dims)
make(A::Type{<:AbstractArray}, d1::Integer, dims::Integer...) = make(A, default_sampling(A), Dims((d1, dims...)))
make(A::Type{<:AbstractArray}, d1::Integer, dims::Integer...) =
make(A, default_sampling(A), Const(Dims((d1, dims...))))

if VERSION < v"1.1.0"
# to resolve ambiguity
make(A::Type{<:AbstractArray}, X, d1::Integer) = make(A, X, Dims((d1,)))
make(A::Type{<:AbstractArray}, X, d1::Integer, d2::Integer) = make(A, X, Dims((d1, d2)))
end
make(A::Type{<:AbstractArray}, dims::Dims) = make(A, default_sampling(A), Const(dims))
make(A::Type{<:AbstractArray}, dims::Distribution) = make(A, default_sampling(A), dims)

# make sure that dims is interpreted as dimensions, not as an Int distribution
make(A::Type{<:AbstractArray}, X, dims::Dims) = make(A, X, Const(dims))
make(A::Type{<:AbstractArray}, ::Type{X}, dims::Dims) where {X} = make(A, X, Const(dims))

make(A::Type{<:AbstractArray}, X, d1, dims...) =
make(A, X, make(Tuple, _makedim(d1), _makedim.(dims)...))

make(A::Type{<:AbstractArray}, ::Type{X}, d1, dims...) where {X} =
make(A, X, make(Tuple, _makedim(d1), _makedim.(dims)...))

# stop the recursion
make(A::Type{<:AbstractArray}, X, dims::Distribution) =
Make{maketype(A, X, dims)}(X, dims)

make(A::Type{<:AbstractArray}, ::Type{X}, dims::Distribution) where {X} =
Make{maketype(A, X, dims)}(X, dims)

Sampler(RNG::Type{<:AbstractRNG}, c::Make2{A}, n::Repetition) where {A<:AbstractArray} =
SamplerTag{A}((sampler(RNG, c[1], n), c[2]))
SamplerTag{A}((sampler(RNG, c[1], Val(Inf)), # values
Sampler(RNG, c[2], n))) # dimensions

rand(rng::AbstractRNG, sp::SamplerTag{A}) where {A<:AbstractArray} =
rand!(rng, A(undef, sp.data[2]), sp.data[1])
rand!(rng, A(undef, rand(rng, sp.data[2])), sp.data[1])


#### Array

# cf. inference bug https://github.com/JuliaLang/julia/issues/28762
# we have to write out all combinations for getting proper inference
maketype(A::Type{Array{T}}, _, ::Dims{N}) where {T, N} = Array{T, N}
maketype(A::Type{Array{T,N}}, _, ::Dims{N}) where {T, N} = Array{T, N}
maketype(A::Type{Array{T,N} where T}, X, ::Dims{N}) where {N} = Array{val_gentype(X), N}
maketype(A::Type{Array}, X, ::Dims{N}) where {N} = Array{val_gentype(X), N}
maketype(A::Type{Array{T}}, _, ::Distribution{Dims{N}}) where {T, N} = Array{T, N}
maketype(A::Type{Array{T,N}}, _, ::Distribution{Dims{N}}) where {T, N} = Array{T, N}
maketype(A::Type{Array{T,N} where T}, X, ::Distribution{Dims{N}}) where {N} = Array{val_gentype(X), N}
maketype(A::Type{Array}, X, ::Distribution{Dims{N}}) where {N} = Array{val_gentype(X), N}

#### BitArray

default_sampling(::Type{<:BitArray}) = Uniform(Bool)

maketype(::Type{BitArray{N}}, _, ::Dims{N}) where {N} = BitArray{N}
maketype(::Type{BitArray}, _, ::Dims{N}) where {N} = BitArray{N}
maketype(::Type{BitArray{N}}, _, ::Distribution{Dims{N}}) where {N} = BitArray{N}
maketype(::Type{BitArray}, _, ::Distribution{Dims{N}}) where {N} = BitArray{N}


#### sparse vectors & matrices
Expand All @@ -482,6 +499,13 @@ make(T::Type{SparseMatrixCSC}, p::AbstractFloat, d1::Integer, d2::Integer) = mak
make(T::Type{SparseVector}, p::AbstractFloat, dims::Dims{1}) = make(T, default_sampling(T), p, dims)
make(T::Type{SparseMatrixCSC}, p::AbstractFloat, dims::Dims{2}) = make(T, default_sampling(T), p, dims)

# stop the recursion
make(T::Type{<:AbstractSparseArray}, X, p::AbstractFloat, dims::Dims) =
Make{maketype(T, X, p, dims)}(X, p, dims)

make(T::Type{<:AbstractSparseArray}, ::Type{X}, p::AbstractFloat, dims::Dims) where {X} =
Make{maketype(T, X, p, dims)}(X, p, dims)


Sampler(RNG::Type{<:AbstractRNG}, c::Make3{A}, n::Repetition) where {A<:AbstractSparseArray} =
SamplerTag{Cont{A}}((sp = sampler(RNG, c[1], n),
Expand Down
37 changes: 31 additions & 6 deletions test/runtests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,7 +79,20 @@ const spString = Sampler(MersenneTwister, String)
a = rand(rng..., Int8, A, 10)
@test a isa Vector{AT}
@test all(in(rInt8), a)
a = rand(rng..., Int8, A, 10:20)
@test a isa Vector{AT}
@test all(in(rInt8), a)
@test length(a) ∈ 10:20
end
a = rand(rng..., Int8, Array, make(Tuple, 1:3, 1:4))
@test a isa Matrix{Int8}
@test all(issubset.(size(a), (1:3, 1:4)))
a = rand(rng..., Int8, Array, 3, 1:4)
@test a isa Matrix{Int8}
@test all(issubset.(size(a), (3:3, 1:4)))
a = rand(rng..., Array, 1:3, 4)
@test a isa Matrix{Float64}
@test all(issubset.(size(a), (1:3, 4:4)))
end

# Set
Expand Down Expand Up @@ -478,24 +491,36 @@ end
@testset "rand(make(Array/BitArray, ...))" begin
for (T, Arr) = (Bool => BitArray, Float64 => Array{Float64}),
k = ([], [T], [Bernoulli(T, 0.3)]),
(d, dim) = ([(6,)] => 1,
[(2,3)] => 2,
[6] => 1,
[2, 3] => 2,
[Int8(2), Int16(3)] => 2),
(needk, d, sz, dim) = ((false, [(6,)], (6,), 1),
(false, [(2,3)], (2,3), 2),
(false, [6], (6,), 1),
(false, [2, 3], (2, 3), 2),
(false, [Int8(2), Int16(3)], (2, 3), 2),

(false, [make(Tuple, 1:3)], (1:3,), 1),
(false, [make(Tuple, 1:3, 1:9)], (1:3, 1:9), 2),
(true, [1:3], (1:3,), 1),
(true, [1:3, Uniform(1:9)], (1:3, 1:9), 2),
(true, [1:3, 9], (1:3, 9), 2),
# (true, [0x1:0x3, 9], (1:3, 9), 2),
(true, [3, Uniform(1:9)], (3, 1:9), 2),
),
A = (T == Bool ?
(BitArray, BitArray{dim}) :
(Array, Array{Float64}, Array{Float64,dim}, Array{U,dim} where U))

needk && isempty(k) && continue
s = rand(make(A, k..., d...))
@test s isa Arr{dim}
@test length(s) == 6
@test all(issubset.(size(s), sz))
end
@test_throws MethodError rand(make(Matrix, 2))
@test_throws MethodError rand(make(Vector, 2, 3))
@test_throws MethodError rand(make(BitMatrix, 2))
@test_throws MethodError rand(make(BitVector, 2, 3))

@test_throws ArgumentError rand(make(Vector, Int, 2, 'a':'c'))

@test rand(make(Array, spString, 9)) isa Array{String}
@test rand(make(BitArray, Sampler(MersenneTwister, [0, 0, 0, 1]), 9)) isa BitArray
# TODO: below was testing without explicit `Array` as 1st argument, so this test
Expand Down