MeasureBase 0.14 + other updates

.github/workflows/Breakage.yml

@@ -15,8 +15,8 @@ jobs:
       fail-fast: false
       matrix:
         pkg: [
-          "cscherrer/Soss.jl",
-          "mschauer/Mitosis.jl"
+          # "cscherrer/Soss.jl",
+          # "mschauer/Mitosis.jl"
         ]
         pkgversion: [latest, stable]
 

Project.toml

@@ -1,7 +1,7 @@
 name = "MeasureTheory"
 uuid = "eadaa1a4-d27c-401d-8699-e962e1bbc33b"
 authors = ["Chad Scherrer <chad.scherrer@gmail.com> and contributors"]
-version = "0.18.4"
+version = "0.19.0"
 
 [deps]
 Accessors = "7d9f7c33-5ae7-4f3b-8dc6-eff91059b697"
@@ -10,7 +10,6 @@ Compat = "34da2185-b29b-5c13-b0c7-acf172513d20"
 ConcreteStructs = "2569d6c7-a4a2-43d3-a901-331e8e4be471"
 ConstructionBase = "187b0558-2788-49d3-abe0-74a17ed4e7c9"
 DensityInterface = "b429d917-457f-4dbc-8f4c-0cc954292b1d"
-DistributionMeasures = "35643b39-bfd4-4670-843f-16596ca89bf3"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
 DynamicIterators = "6c76993d-992e-5bf1-9e63-34920a5a5a38"
 FillArrays = "1a297f60-69ca-5386-bcde-b61e274b549b"
@@ -47,30 +46,29 @@ Compat = "3.42, 4"
 ConcreteStructs = "0.2"
 ConstructionBase = "1.3"
 DensityInterface = "0.4"
-DistributionMeasures = "0.2"
 Distributions = "0.25"
 DynamicIterators = "0.4"
-FillArrays = "0.12, 0.13"
+FillArrays = "1"
 IfElse = "0.1"
 Infinities = "0.1"
 InverseFunctions = "0.1"
 KeywordCalls = "0.2"
-LazyArrays = "0.22"
+LazyArrays = "1"
 LogExpFunctions = "0.3.3"
 MLStyle = "0.4"
 MacroTools = "0.5"
 MappedArrays = "0.4"
-MeasureBase = "0.13"
+MeasureBase = "0.14"
 NamedTupleTools = "0.13, 0.14"
 PositiveFactorizations = "0.2"
 PrettyPrinting = "0.3, 0.4"
 Reexport = "1"
 SpecialFunctions = "1, 2"
-Static = "0.5, 0.6"
+Static = "0.8"
 StaticArraysCore = "1"
-StatsBase = "0.32, 0.33"
+StatsBase = "0.34"
 StatsFuns = "0.9, 1"
-TransformVariables = "0.5, 0.6, 0.7"
+TransformVariables = "0.8"
 Tricks = "0.1"
 julia = "1.6"
 

src/MeasureTheory.jl

@@ -8,7 +8,6 @@ using MLStyle
 import TransformVariables
 const TV = TransformVariables
 
-using DistributionMeasures
 using TransformVariables: asℝ₊, as𝕀, asℝ, transform
 
 import Base
@@ -27,7 +26,7 @@ export Lebesgue
 export ℝ, ℝ₊, 𝕀
 export ⊙
 export SpikeMixture
-export CountingMeasure
+export Counting
 export TrivialMeasure
 export Likelihood
 export testvalue
@@ -60,7 +59,7 @@ import MeasureBase:
     𝒹,
     ∫exp
 import MeasureBase: ≪
-using MeasureBase: BoundedInts, BoundedReals, CountingMeasure, IntegerDomain, IntegerNumbers
+using MeasureBase: BoundedInts, BoundedReals, CountingBase, IntegerDomain, IntegerNumbers
 using MeasureBase: weightedmeasure, restrict
 using MeasureBase: AbstractTransitionKernel
 
@@ -97,7 +96,7 @@ using MeasureBase: kernel
 using MeasureBase: Returns
 import MeasureBase: proxy, @useproxy
 import MeasureBase: basemeasure_depth
-using MeasureBase: LebesgueMeasure
+using MeasureBase: LebesgueBase
 
 import DensityInterface: logdensityof
 import DensityInterface: densityof
@@ -119,6 +118,9 @@ xlog1py(x, y) = x * log(1 + y)
 
 as(args...; kwargs...) = TV.as(args...; kwargs...)
 
+# Type piracy until https://github.com/JuliaMath/MeasureBase.jl/issues/127 is fixed
+MeasureBase.rand(::FixedRNG, ::Type{Bool}) = true
+
 include("utils.jl")
 include("const.jl")
 include("combinators/for.jl")

src/combinators/affine.jl

@@ -282,7 +282,7 @@ end
 @inline function basemeasure(d::AffinePushfwd{N,L}) where {N,L<:Lebesgue}
     weightedmeasure(-logjac(d), d.parent)
 end
-@inline function basemeasure(d::AffinePushfwd{N,L}) where {N,L<:LebesgueMeasure}
+@inline function basemeasure(d::AffinePushfwd{N,L}) where {N,L<:LebesgueBase}
     weightedmeasure(-logjac(d), d.parent)
 end
 

src/distributions.jl

@@ -39,4 +39,3 @@ function as(d::Dists.Product, _data::NamedTuple = NamedTuple())
     as(Vector, as(v[1]), n)
 end
 
-as(m::DistributionMeasures.DistributionMeasure) = as(m.d)

src/parameterized/bernoulli.jl

@@ -13,7 +13,7 @@ import Base
 
 Bernoulli(p) = Bernoulli((p = p,))
 
-basemeasure(::Bernoulli) = CountingMeasure()
+basemeasure(::Bernoulli) = CountingBase()
 
 testvalue(::Bernoulli) = true
 

src/parameterized/beta.jl

@@ -21,7 +21,7 @@ end
 
 @inline function basemeasure(d::Beta{(:α, :β)})
     ℓ = -logbeta(d.α, d.β)
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end
 
 Base.rand(rng::AbstractRNG, T::Type, μ::Beta) = rand(rng, Dists.Beta(μ.α, μ.β))

src/parameterized/betabinomial.jl

@@ -6,7 +6,7 @@ using SpecialFunctions
 
 @parameterized BetaBinomial(n, α, β)
 
-basemeasure(d::BetaBinomial) = CountingMeasure()
+basemeasure(d::BetaBinomial) = CountingBase()
 
 testvalue(::BetaBinomial) = 0
 

src/parameterized/binomial.jl

@@ -9,7 +9,7 @@ probit(p) = sqrt2 * erfinv(2p - 1)
 
 @parameterized Binomial(n, p)
 
-basemeasure(d::Binomial) = CountingMeasure()
+basemeasure(d::Binomial) = CountingBase()
 
 testvalue(::Binomial) = 0
 

src/parameterized/cauchy.jl

@@ -14,7 +14,7 @@
 
 logdensity_def(d::Cauchy, x) = logdensity_def(proxy(d), x)
 
-basemeasure(d::Cauchy) = WeightedMeasure(static(-logπ), LebesgueMeasure())
+basemeasure(d::Cauchy) = WeightedMeasure(static(-logπ), LebesgueBase())
 
 # @affinepars Cauchy
 
@@ -28,6 +28,8 @@
 
 Base.rand(rng::AbstractRNG, T::Type, μ::Cauchy{()}) = randn(rng, T) / randn(rng, T)
 
+Base.rand(::FixedRNG, ::Type{T}, μ::Cauchy{()}) where {T} = zero(T)
+
 for N in AFFINEPARS
     @eval begin
         proxy(d::Cauchy{$N}) = affine(params(d), Cauchy())
@@ -46,6 +48,8 @@
 
 HalfCauchy(σ) = HalfCauchy(σ = σ)
 
+Base.rand(::FixedRNG, ::Type{T}, μ::Half{<:Cauchy}) where {T} = one(T)
+
 proxy(d::Cauchy{()}) = Dists.Cauchy()
 
 insupport(::Cauchy, x) = true

src/parameterized/exponential.jl

@@ -6,7 +6,7 @@ export Exponential
 @parameterized Exponential(β)
 
 insupport(::Exponential, x) = x ≥ 0
-basemeasure(::Exponential) = LebesgueMeasure()
+basemeasure(::Exponential) = LebesgueBase()
 
 @kwstruct Exponential()
 

src/parameterized/gamma.jl

@@ -19,7 +19,7 @@
 
 function basemeasure(d::Gamma{(:k,)})
     ℓ = -loggamma(d.k)
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end
 
 @kwstruct Gamma(k, σ)
@@ -57,11 +57,11 @@
     ϕ = d.ϕ
     ϕinv = inv(ϕ)
     ℓ = -ϕinv * log(ϕ) - first(logabsgamma(ϕinv))
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end
 
 function basemeasure(d::Gamma{(:μ, :ϕ),Tuple{M,StaticFloat64{ϕ}}}) where {M,ϕ}
     ϕinv = inv(ϕ)
     ℓ = static(-ϕinv * log(ϕ) - first(logabsgamma(ϕinv)))
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end

src/parameterized/gumbel.jl

@@ -4,7 +4,7 @@ export Gumbel
 
 @parameterized Gumbel()
 
-basemeasure(::Gumbel{()}) = LebesgueMeasure()
+basemeasure(::Gumbel{()}) = LebesgueBase()
 
 @kwstruct Gumbel()
 
@@ -28,8 +28,8 @@ end
 
 import Base
 
-function Base.rand(rng::AbstractRNG, d::Gumbel{()})
-    u = rand(rng)
+function Base.rand(rng::AbstractRNG, ::Type{T}, d::Gumbel{()}) where {T}
+    u = rand(rng, T)
     -log(-log(u))
 end
 

src/parameterized/inverse-gaussian.jl

@@ -49,5 +49,5 @@
 
 function basemeasure(d::InverseGaussian{(:μ, :ϕ)})
     ℓ = static(-0.5) * (static(float(log2π)) + log(d.ϕ))
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end

src/parameterized/laplace.jl

@@ -27,13 +27,14 @@ end
 
 logdensity_def(d::Laplace, x) = logdensity_def(proxy(d), x)
 
-basemeasure(::Laplace{()}) = WeightedMeasure(static(-logtwo), LebesgueMeasure())
+basemeasure(::Laplace{()}) = WeightedMeasure(static(-logtwo), LebesgueBase())
 
 # @affinepars Laplace
 
 function Base.rand(rng::AbstractRNG, ::Type{T}, μ::Laplace{()}) where {T}
-    rand(rng, Dists.Laplace())
+    T(rand(rng, Dists.Laplace()))
 end
+
 Base.rand(rng::AbstractRNG, ::Type{T}, μ::Laplace) where {T} = Base.rand(rng, T, proxy(μ))
 
 ≪(::Laplace, ::Lebesgue{X}) where {X<:Real} = true

src/parameterized/lkj-cholesky.jl

@@ -81,14 +81,14 @@
 
 @inline function basemeasure(d::LKJCholesky{(:k, :η)})
     t = as(d)
-    base = Pushforward(t, LebesgueMeasure()^TV.dimension(t), False())
+    base = Pushforward(t, LebesgueBase()^TV.dimension(t), False())
     WeightedMeasure(Dists.lkj_logc0(d.k, d.η), base)
 end
 
 @inline function basemeasure(d::LKJCholesky{(:k, :logη)})
     t = as(d)
     η = exp(d.logη)
-    base = Pushforward(t, LebesgueMeasure()^TV.dimension(t), False())
+    base = Pushforward(t, LebesgueBase()^TV.dimension(t), False())
     WeightedMeasure(Dists.lkj_logc0(d.k, η), base)
 end
 

src/parameterized/multinomial.jl

@@ -4,7 +4,7 @@ export Multinomial
 
 @parameterized Multinomial(n, p)
 
-basemeasure(d::Multinomial) = CountingMeasure()
+basemeasure(d::Multinomial) = CountingBase()
 
 @inline function insupport(d::Multinomial{(:n, :p)}, x)
     length(x) == length(d.p) || return false

src/parameterized/negativebinomial.jl

@@ -7,7 +7,7 @@ import Base
 
 insupport(::NegativeBinomial, x) = isinteger(x) && x ≥ 0
 
-basemeasure(::NegativeBinomial) = CountingMeasure()
+basemeasure(::NegativeBinomial) = CountingBase()
 
 testvalue(::NegativeBinomial) = 0
 

src/parameterized/normal.jl

@@ -37,7 +37,7 @@
 insupport(d::Normal) = Returns(true)
 
 @inline logdensity_def(d::Normal{()}, x) = -x^2 / 2
-@inline basemeasure(::Normal{()}) = WeightedMeasure(static(-0.5 * log2π), LebesgueMeasure())
+@inline basemeasure(::Normal{()}) = WeightedMeasure(static(-0.5 * log2π), LebesgueBase())
 
 @kwstruct Normal(μ)
 @kwstruct Normal(σ)
@@ -147,7 +147,7 @@
 
 @inline function basemeasure(d::Normal{(:σ²,)})
     ℓ = static(-0.5) * (static(float(log2π)) + log(d.σ²))
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end
 
 proxy(d::Normal{(:μ, :σ²)}) = affine((μ = d.μ,), Normal((σ² = d.σ²,)))
@@ -163,7 +163,7 @@
 
 @inline function basemeasure(d::Normal{(:τ,)})
     ℓ = static(-0.5) * (static(float(log2π)) - log(d.τ))
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end
 
 proxy(d::Normal{(:μ, :τ)}) = affine((μ = d.μ,), Normal((τ = d.τ,)))

src/parameterized/snedecorf.jl

@@ -16,7 +16,7 @@
 
 @inline function basemeasure(d::SnedecorF{(:ν1, :ν2)})
     ℓ = -logbeta(d.ν1 / 2, d.ν2 / 2)
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end
 
 xform(::SnedecorF) = asℝ₊

src/parameterized/studentt.jl

@@ -50,7 +50,7 @@ end
 
 @inline function basemeasure(d::StudentT{(:ν,)})
     ℓ = loggamma((d.ν + 1) / 2) - loggamma(d.ν / 2) - log(π * d.ν) / 2
-    weightedmeasure(ℓ, LebesgueMeasure())
+    weightedmeasure(ℓ, LebesgueBase())
 end
 
 xform(::StudentT) = asℝ

src/parameterized/uniform.jl

@@ -13,7 +13,7 @@ export Uniform
 insupport(::Uniform{()}) = in𝕀
 insupport(::Uniform{()}, x) = in𝕀(x)
 
-@inline basemeasure(::Uniform{()}) = LebesgueMeasure()
+@inline basemeasure(::Uniform{()}) = LebesgueBase()
 
 proxy(::Uniform{()}) = Dists.Uniform()
 

src/transforms/corrcholesky.jl

@@ -30,9 +30,9 @@
     return Cholesky(U, 'U', 0), ℓ, index
 end
 
-TV.inverse_eltype(t::CorrCholesky, x::AbstractMatrix) = TV.extended_eltype(x)
+TV.inverse_eltype(t::CorrCholesky, x::AbstractMatrix) = eltype(x)
 
-TV.inverse_eltype(t::CorrCholesky, x::Cholesky) = TV.extended_eltype(x)
+TV.inverse_eltype(t::CorrCholesky, x::Cholesky) = eltype(x)
 
 function TV.inverse_at!(x::AbstractVector, index, t::CorrCholesky, L::LowerTriangular)
     return TV.inverse_at!(x, index, CorrCholeskyUpper(t.n), L')

src/transforms/corrcholeskylower.jl

@@ -38,7 +38,7 @@
     return U', ℓ, index
 end
 
-TV.inverse_eltype(t::CorrCholeskyLower, L::LowerTriangular) = TV.extended_eltype(L)
+TV.inverse_eltype(t::CorrCholeskyLower, L::LowerTriangular) = eltype(L)
 
 function TV.inverse_at!(x::AbstractVector, index, t::CorrCholeskyLower, L::LowerTriangular)
     return TV.inverse_at!(x, index, CorrCholeskyUpper(t.n), L')

src/transforms/ordered.jl

@@ -49,7 +49,7 @@
     return (y, ℓ, index)
 end
 
-TV.inverse_eltype(t::Ordered, y::AbstractVector) = TV.extended_eltype(y)
+TV.inverse_eltype(t::Ordered, y::AbstractVector) = eltype(y)
 
 Ordered(n::Int) = Ordered(asℝ, n)