Dont check for proper in division of to accomodate inference

ext/ReactiveMPProjectionExt/ReactiveMPProjectionExt.jl

@@ -5,6 +5,13 @@ using ReactiveMP, ExponentialFamily, Distributions, ExponentialFamilyProjection,
 struct DivisionOf{A, B}
     numerator::A
     denumerator::B
+    function DivisionOf(numerator::A, denumerator::B) where {A, B}
+        if variate_form(A) == variate_form(B)
+            return new{A, B}(numerator, denumerator)
+        else
+            error("DivisionOf does not support arguments of different variate forms: $(variate_form(A)) and $(variate_form(B))")
+        end
+    end
 end
 
 (divisionof::DivisionOf)(x) = logpdf(divisionof, x)
@@ -44,6 +51,8 @@ include("layout/cvi_projection.jl")
 include("rules/in.jl")
 include("rules/out.jl")
 include("rules/marginals.jl")
+include("divisionof/univariate_gaussian.jl")
+include("divisionof/multivariate_gaussian.jl")
 
 # This will enable the extension and make `CVIProjection` compatible with delta nodes 
 # Otherwise it should throw an error suggesting users to install `ExponentialFamilyProjection`

ext/ReactiveMPProjectionExt/divisionof/multivariate_gaussian.jl

@@ -0,0 +1,28 @@
+function BayesBase.prod(
+    ::GenericProd, something::C, division::DivisionOf{A, B}
+) where {A <: MultivariateGaussianDistributionsFamily, B <: MultivariateGaussianDistributionsFamily, C <: MultivariateGaussianDistributionsFamily}
+    d_numerator = convert(MvNormalMeanPrecision, division.numerator)
+    d_denumerator = convert(MvNormalMeanPrecision, division.denumerator)
+    d_something = convert(MvNormalMeanPrecision, something)
+
+    ef_a = convert(ExponentialFamilyDistribution, d_numerator)
+    ef_b = convert(ExponentialFamilyDistribution, d_denumerator)
+    ef_c = convert(ExponentialFamilyDistribution, d_something)
+
+    ef_a_typetag = ExponentialFamily.exponential_family_typetag(ef_a)
+
+    resulting_nat_params = ExponentialFamily.getnaturalparameters(ef_a) - ExponentialFamily.getnaturalparameters(ef_b) + ExponentialFamily.getnaturalparameters(ef_c)
+    ef_resulting = ExponentialFamily.ExponentialFamilyDistribution(ef_a_typetag, resulting_nat_params, nothing, nothing)
+
+    if !ExponentialFamily.isproper(ef_resulting)
+        @warn "The product of $(something) and $(division.numerator) divided by $(division.denumerator) is not proper" maxlog=1
+    end
+
+    return convert(Distribution, ef_resulting)
+end
+
+function BayesBase.prod(
+    prodtype::GenericProd, division::DivisionOf{A, B}, something::C
+) where {A <: MultivariateGaussianDistributionsFamily, B <: MultivariateGaussianDistributionsFamily, C <: MultivariateGaussianDistributionsFamily}
+    return prod(prodtype, something, division)
+end

ext/ReactiveMPProjectionExt/divisionof/univariate_gaussian.jl

@@ -0,0 +1,24 @@
+function BayesBase.prod(
+    ::GenericProd, something::C, division::DivisionOf{A, B}
+) where {A <: UnivariateGaussianDistributionsFamily, B <: UnivariateGaussianDistributionsFamily, C <: UnivariateGaussianDistributionsFamily}
+    ef_a = convert(ExponentialFamilyDistribution, division.numerator)
+    ef_b = convert(ExponentialFamilyDistribution, division.denumerator)
+    ef_c = convert(ExponentialFamilyDistribution, something)
+
+    ef_a_typetag = ExponentialFamily.exponential_family_typetag(ef_a)
+
+    resulting_nat_params = ExponentialFamily.getnaturalparameters(ef_a) - ExponentialFamily.getnaturalparameters(ef_b) + ExponentialFamily.getnaturalparameters(ef_c)
+    ef_resulting = ExponentialFamily.ExponentialFamilyDistribution(ef_a_typetag, resulting_nat_params, nothing, nothing)
+
+    if !ExponentialFamily.isproper(ef_resulting)
+        @warn "The product of $(something) and $(division.numerator) divided by $(division.denumerator) is not proper" maxlog=1
+    end
+
+    return convert(Distribution, ef_resulting)
+end
+
+function BayesBase.prod(
+    prodtype::GenericProd, division::DivisionOf{A, B}, something::C
+) where {A <: UnivariateGaussianDistributionsFamily, B <: UnivariateGaussianDistributionsFamily, C <: UnivariateGaussianDistributionsFamily}
+    return prod(prodtype, something, division)
+end

test/ext/ReactiveMPProjectionExt/ReactiveMPProjectionExt_tests.jl

@@ -15,6 +15,105 @@
     @test ext.DivisionOf(d1, d2) == prod(GenericProd(), missing, ext.DivisionOf(d1, d2))
 end
 
+@testitem "Check warning when DivisionOf is not proper" begin
+    using ExponentialFamily, ExponentialFamilyProjection, BayesBase
+    ext = Base.get_extension(ReactiveMP, :ReactiveMPProjectionExt)
+    @test !isnothing(ext)
+    using .ext
+
+    @testset "Check warning when DivisionOf is not proper" begin
+        d1 = NormalMeanVariance(0, 1)
+        d2 = NormalMeanVariance(0, 0.01)
+        d3 = NormalMeanVariance(0, 0.5)
+        @test_logs (:warn, "The product of $(d3) and $(d1) divided by $(d2) is not proper") prod(GenericProd(), ext.DivisionOf(d1, d2), d3)
+    end
+
+    @testset "Check warning when DivisionOf is not proper" begin
+        d1 = MvNormalMeanCovariance([0.0, 0.0], [1.0 0.0; 0.0 1.0])
+        d2 = MvNormalMeanCovariance([1.0, 1.0], [0.01 0.0; 0.0 0.01])
+        d3 = MvNormalMeanCovariance([2.0, 2.0], [3.0 0.0; 0.0 3.0])
+        @test_logs (:warn, "The product of $(d3) and $(d1) divided by $(d2) is not proper") prod(GenericProd(), ext.DivisionOf(d1, d2), d3)
+    end
+
+    @testset "Check no warning when DivisionOf is proper" begin
+        d1 = NormalMeanVariance(0, 1)
+        d2 = NormalMeanVariance(0, 2)
+        d3 = NormalMeanVariance(0, 1)
+        @test_logs prod(GenericProd(), ext.DivisionOf(d1, d2), d3)
+    end
+end
+
+@testitem "DivisionOf(Gaussian, Gaussian) x Gaussian" begin
+    using ExponentialFamily, ExponentialFamilyProjection, BayesBase
+
+    # `DivisionOf` is internal to the extension
+    ext = Base.get_extension(ReactiveMP, :ReactiveMPProjectionExt)
+    @test !isnothing(ext)
+    using .ext
+
+    d1 = NormalMeanVariance(0, 1)
+    d2 = NormalMeanVariance(1, 2)
+    d3 = NormalMeanVariance(2, 3)
+    result = NormalMeanVariance(0.2, 1.2)
+
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ prod(GenericProd(), d3, ext.DivisionOf(d1, d2))
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ result
+
+    d1 = NormalMeanVariance(0, 1)
+    d2 = convert(NormalWeightedMeanPrecision, d2)
+    d3 = convert(NormalMeanPrecision, d3)
+
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ prod(GenericProd(), d3, ext.DivisionOf(d1, d2))
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ result
+end
+
+@testitem "DivisionOf(MvGaussian, MvGaussian) x MvGaussian" begin
+    using ExponentialFamily, ExponentialFamilyProjection, BayesBase
+
+    # `DivisionOf` is internal to the extension
+    ext = Base.get_extension(ReactiveMP, :ReactiveMPProjectionExt)
+    @test !isnothing(ext)
+    using .ext
+
+    d1 = MvNormalMeanCovariance([0.0, 0.0], [1.0 0.0; 0.0 1.0])
+    d2 = MvNormalMeanCovariance([1.0, 1.0], [2.0 0.0; 0.0 2.0])
+    d3 = MvNormalMeanCovariance([2.0, 2.0], [3.0 0.0; 0.0 3.0])
+    result = MvNormalMeanCovariance([0.2, 0.2], [1.2 0.0; 0.0 1.2])
+
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ prod(GenericProd(), d3, ext.DivisionOf(d1, d2))
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ result
+
+    d1 = MvNormalMeanCovariance([0.0, 0.0], [1.0 0.0; 0.0 1.0])
+    d2 = convert(MvNormalWeightedMeanPrecision, d2)
+    d3 = convert(MvNormalMeanPrecision, d3)
+
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ prod(GenericProd(), d3, ext.DivisionOf(d1, d2))
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ result
+
+    d1 = MvNormalMeanCovariance([0.0, 0.0], [1.0 0.0; 0.0 1.0])
+    d2 = MvNormalMeanScalePrecision([1.0, 1.0], 1 / 2)
+    d3 = convert(MvNormalMeanPrecision, d3)
+
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ prod(GenericProd(), d3, ext.DivisionOf(d1, d2))
+    @test prod(GenericProd(), ext.DivisionOf(d1, d2), d3) ≈ result
+end
+
+@testitem "Raise error when DivisionOf of Univarive and Multivariate" begin
+    using ExponentialFamily, ExponentialFamilyProjection, BayesBase
+
+    # `DivisionOf` is internal to the extension
+    ext = Base.get_extension(ReactiveMP, :ReactiveMPProjectionExt)
+    @test !isnothing(ext)
+    using .ext
+
+    d1 = NormalMeanVariance(0, 1)
+    d2 = MvNormalMeanCovariance([0.0, 0.0], [1.0 0.0; 0.0 1.0])
+    d3 = NormalMeanVariance(0, 1)
+
+    @test_throws "DivisionOf does not support arguments of different variate forms" prod(GenericProd(), ext.DivisionOf(d1, d2), d2)
+    @test_throws "DivisionOf does not support arguments of different variate forms" prod(GenericProd(), d2, ext.DivisionOf(d1, d2))
+end
+
 @testitem "create_project_to_ins type stability" begin
     using ExponentialFamily, ExponentialFamilyProjection, BayesBase, Test
     using ReactiveMP: CVIProjection