Merge branch 'master' into DistributionMeasures

Author: cscherrer

Project.toml

@@ -1,7 +1,7 @@
 name = "MeasureTheory"
 uuid = "eadaa1a4-d27c-401d-8699-e962e1bbc33b"
 authors = ["Chad Scherrer <[email protected]> and contributors"]
-version = "0.16.6"
+version = "0.17.2"
 
 [deps]
 Accessors = "7d9f7c33-5ae7-4f3b-8dc6-eff91059b697"
@@ -27,14 +27,13 @@ MacroTools = "1914dd2f-81c6-5fcd-8719-6d5c9610ff09"
 MappedArrays = "dbb5928d-eab1-5f90-85c2-b9b0edb7c900"
 MeasureBase = "fa1605e6-acd5-459c-a1e6-7e635759db14"
 NamedTupleTools = "d9ec5142-1e00-5aa0-9d6a-321866360f50"
-NestedTuples = "a734d2a7-8d68-409b-9419-626914d4061d"
 PositiveFactorizations = "85a6dd25-e78a-55b7-8502-1745935b8125"
 PrettyPrinting = "54e16d92-306c-5ea0-a30b-337be88ac337"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Reexport = "189a3867-3050-52da-a836-e630ba90ab69"
 SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
 Static = "aedffcd0-7271-4cad-89d0-dc628f76c6d3"
-StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
+StaticArraysCore = "1e83bf80-4336-4d27-bf5d-d5a4f845583c"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
 StatsFuns = "4c63d2b9-4356-54db-8cca-17b64c39e42c"
@@ -44,10 +43,11 @@ Tricks = "410a4b4d-49e4-4fbc-ab6d-cb71b17b3775"
 [compat]
 Accessors = "0.1"
 ChangesOfVariables = "0.1"
-Compat = "3.35, 4"
+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"
@@ -60,15 +60,14 @@ LogExpFunctions = "0.3.3"
 MLStyle = "0.4"
 MacroTools = "0.5"
 MappedArrays = "0.4"
-MeasureBase = "0.12"
+MeasureBase = "0.13"
 NamedTupleTools = "0.13, 0.14"
-NestedTuples = "0.3"
 PositiveFactorizations = "0.2"
 PrettyPrinting = "0.3, 0.4"
 Reexport = "1"
 SpecialFunctions = "1, 2"
 Static = "0.5, 0.6"
-StaticArrays = "1.3"
+StaticArraysCore = "1"
 StatsBase = "0.32, 0.33"
 StatsFuns = "0.9, 1"
 TransformVariables = "0.5, 0.6"

src/MeasureTheory.jl

@@ -1,13 +1,14 @@
 module MeasureTheory
 
+using Compat
 using Random
 
 using MeasureBase
 using MLStyle
-using NestedTuples
 import TransformVariables
 const TV = TransformVariables
 
+using DistributionMeasures
 using TransformVariables: asℝ₊, as𝕀, asℝ, transform
 
 import Base
@@ -65,13 +66,12 @@ using MeasureBase: AbstractTransitionKernel
 
 import Statistics: mean, var, std
 
-import MeasureBase: likelihood
-export likelihood
+import MeasureBase: likelihoodof
+export likelihoodof
 export log_likelihood_ratio
 
 using DistributionMeasures
-
-using StaticArrays
+using StaticArraysCore
 
 import PrettyPrinting
 

src/combinators/exponential-families.jl

@@ -97,7 +97,7 @@ end
 
 # end
 
-function MeasureBase.likelihood(fam::ExponentialFamily, y)
+function MeasureBase.likelihoodof(fam::ExponentialFamily, y)
     c = logdensityof(fam.base, y)
     t = ApplyArray(vcat, (f.(y) for f in fam.t)...)
     tᵀx = t' * fam.x

src/combinators/product.jl

@@ -2,6 +2,10 @@ function as(d::PowerMeasure)
     as(Array, as(d.parent), length.(d.axes)...)
 end
 
+function as(d::ProductMeasure{<:AbstractArray{<:Dirac}})
+    return asConst(testvalue.(marginals(d)))
+end
+
 function as(d::ProductMeasure{A}) where {A<:AbstractArray}
     mar = marginals(d)
     ts = map(as, mar)

src/combinators/tweedie.jl

@@ -145,7 +145,7 @@ struct TweedieLikelihood{C,Θ,H,T,A} <: AbstractLikelihood
     a::A
 end
 
-function likelihood(fam::Tweedie, x)
+function likelihoodof(fam::Tweedie, x)
     c = logdensityof(fam.base, x)
     t = fam.t(x)
     TweedieLikelihood(c, fam.θ, fam.η, t, fam.a)

src/distributions.jl

@@ -3,43 +3,43 @@ import Distributions
 export Dists
 const Dists = Distributions
 
-function Measure(dist::Dists.Distribution{F,S}) where {F,S}
-    X = Base.eltype(dist)
-    DistributionMeasure{F,S,X}(dist)
-end
 
-struct DistributionMeasure{F,S,X} <: AbstractMeasure
-    dist::Dists.Distribution{F,S}
-end
 
-@inline function basemeasure(μ::DistributionMeasure{F,S,X}) where {F,S<:Dists.Continuous,X}
-    return Lebesgue(X)
-end
+@inline function as(d, _data::NamedTuple)
+    if hasmethod(Dists.support, (typeof(d),))
+        return asTransform(Dists.support(d)) 
+    end
 
-@inline function logdensity_def(μ::DistributionMeasure{F,S,X}, x::X) where {F,S,X}
-    return Dists.logpdf(μ.dist, x)
+    error("Not implemented:\nas($d)")
 end
 
-function Base.rand(rng::AbstractRNG, μ::DistributionMeasure{F,S,X}) where {F,S,X}
-    return rand(rng, μ.dist)
-end
+using TransformVariables: ShiftedExp, ScaledShiftedLogistic
 
-basemeasure(d::Dists.AbstractMvNormal) = Lebesgue(ℝ)^size(d)
+function asTransform(supp:: Dists.RealInterval) 
+    (lb, ub) = (supp.lb, supp.ub)
 
-@inline function basemeasure(μ::Dists.Distribution{Dists.Univariate,Dists.Continuous})
-    return Lebesgue(ℝ)
+    (lb, ub) == (-Inf, Inf) && (return asℝ)
+    isinf(ub) && return ShiftedExp(true,lb)
+    isinf(lb) && return ShiftedExp(false,lb)
+    return ScaledShiftedLogistic(ub-lb, lb)
 end
 
-@inline function basemeasure(μ::Dists.Distribution{Dists.Univariate,Dists.Discrete})
-    return CountingMeasure(ℤ)
-end
+as(μ::AbstractMeasure,  _data::NamedTuple) = as(μ)
 
-∫(::typeof(identity), ::Dists.Distribution) = 1.0
+as(d::Dists.AbstractMvNormal, _data::NamedTuple = NamedTuple()) = TV.as(Array, size(d))
 
-logdensity_def(μ::Dists.Distribution, x) = Dists.logpdf(μ, x)
 
-density_def(μ::Dists.Distribution, x) = Dists.pdf(μ, x)
+function as(d::Dists.Distribution{Dists.Univariate}, _data::NamedTuple=NamedTuple())
+    sup = Dists.support(d)
+    lo = isinf(sup.lb) ? -TV.∞ : sup.lb
+    hi = isinf(sup.ub) ? TV.∞ : sup.ub
+    as(Real, lo,hi)
+end
 
-testvalue(d::Dists.Distribution) = rand(d)
+function as(d::Dists.Product, _data::NamedTuple=NamedTuple())
+    n = length(d)
+    v = d.v
+    as(Vector, as(v[1]), n)
+end
 
-insupport(d::Dists.Distribution, x) = logdensityof(d, x) > -Inf
+as(m::DistributionMeasures.DistributionMeasure) = as(m.d)

test/runtests.jl

@@ -275,6 +275,12 @@ end
     end
 end
 
+@testset "Product of Diracs" begin
+    x = randn(3)
+    t = as(productmeasure(Dirac.(x))) 
+    @test transform(t, []) == x
+end
+
 # @testset "Univariate chain" begin
 #     ξ0 = 1.
 #     x = 1.2