Add rules and tests for GammaShapeRate node

src/nodes/predefined/gamma_mixture.jl

@@ -183,6 +183,8 @@ struct GammaShapeLikelihood{T <: Real} <: ContinuousUnivariateDistribution
  γ::T # p * β
 end
 
+Distributions.params(distribution::GammaShapeLikelihood) = (distribution.p, distribution.γ)
+
 Distributions.@distr_support GammaShapeLikelihood 0.0 Inf
 
 BayesBase.support(dist::GammaShapeLikelihood) = Distributions.RealInterval(0.0, Inf)

src/rules/gamma/marginals.jl

@@ -1,5 +1,5 @@
 export marginalrule
 
-@marginalrule Gamma(:out_α_θ) (m_out::Gamma, m_α::PointMass, m_θ::PointMass) = begin
+@marginalrule Gamma(:out_α_θ) (m_out::GammaDistributionsFamily, m_α::PointMass, m_θ::PointMass) = begin
  return (out = prod(ClosedProd(), Gamma(mean(m_α), mean(m_θ)), m_out), α = m_α, θ = m_θ)
 end

src/rules/gamma_shape_rate/a.jl

@@ -1,5 +1,7 @@
 import DomainSets
 
-@rule GammaShapeRate(:α, Marginalisation) (q_out::Gamma, q_β::Gamma) = begin
- return ContinuousUnivariateLogPdf(DomainSets.HalfLine(), (α) -> α * mean(log, q_β) + (α - 1) * mean(log, q_out) - loggamma(α))
+@rule GammaShapeRate(:α, Marginalisation) (q_out::Any, q_β::GammaDistributionsFamily) = begin
+ γ = mean(log, q_β) + mean(log, q_out)
+ params = promote(1, γ)
+ return GammaShapeLikelihood(params...)
 end

src/rules/gamma_shape_rate/b.jl

@@ -0,0 +1,3 @@
+export rule
+
+@rule GammaShapeRate(:β, Marginalisation) (q_out::Any, q_α::Any) = GammaShapeRate(1 + mean(q_α), mean(q_out))

src/rules/gamma_shape_rate/marginals.jl

@@ -1,5 +1,5 @@
 export marginalrule
 
-@marginalrule GammaShapeRate(:out_α_β) (m_out::Gamma, m_α::PointMass, m_β::PointMass) = begin
+@marginalrule GammaShapeRate(:out_α_β) (m_out::GammaDistributionsFamily, m_α::PointMass, m_β::PointMass) = begin
  return (out = prod(ClosedProd(), GammaShapeRate(mean(m_α), mean(m_β)), m_out), α = m_α, β = m_β)
 end

src/rules/gamma_shape_rate/out.jl

@@ -2,4 +2,4 @@ export rule
 
 @rule GammaShapeRate(:out, Marginalisation) (m_α::PointMass, m_β::PointMass) = GammaShapeRate(mean(m_α), mean(m_β))
 
-@rule GammaShapeRate(:out, Marginalisation) (q_α::PointMass, q_β::PointMass) = GammaShapeRate(mean(q_α), mean(q_β))
+@rule GammaShapeRate(:out, Marginalisation) (q_α::Any, q_β::Any) = GammaShapeRate(mean(q_α), mean(q_β))

src/rules/predefined.jl

@@ -27,6 +27,7 @@ include("gamma_inverse/marginals.jl")
 include("gamma_shape_rate/out.jl")
 include("gamma_shape_rate/marginals.jl")
 include("gamma_shape_rate/a.jl")
+include("gamma_shape_rate/b.jl")
 
 include("beta/out.jl")
 include("beta/marginals.jl")

test/rules/gamma_shape_rate/a_tests.jl

@@ -0,0 +1,12 @@
+@testitem "rules:GammaShapeRate:α" begin
+ using ReactiveMP, BayesBase, Random, ExponentialFamily, Distributions
+
+ import ReactiveMP: @test_rules, GammaShapeLikelihood
+
+ @testset "Variational Message Passing: (q_out::Any, q_β::GammaDistributionsFamily)" begin
+ @test_rules [check_type_promotion = true] GammaShapeRate(:α, Marginalisation) [
+ (input = (q_out = GammaShapeRate(1.0, 1.0), q_β = GammaShapeRate(1.0, 1.0)), output = GammaShapeLikelihood(1.0, 2.0 * -0.5772156649015315)),
+ (input = (q_out = PointMass(1.0), q_β = GammaShapeRate(1.0, 1.0)), output = GammaShapeLikelihood(1.0, -0.5772156649015315))
+ ]
+ end
+end

test/rules/gamma_shape_rate/b_tests.jl

@@ -0,0 +1,14 @@
+@testitem "rules:GammaShapeRate:β" begin
+ using ReactiveMP, BayesBase, Random, ExponentialFamily, Distributions
+
+ import ReactiveMP: @test_rules, GammaShapeLikelihood
+
+ @testset "Variational Message Passing: (q_out::Any, q_α::Any)" begin
+ @test_rules [check_type_promotion = true] GammaShapeRate(:β, Marginalisation) [
+ (input = (q_out = GammaShapeRate(1.0, 1.0), q_α = GammaShapeRate(1.0, 1.0)), output = GammaShapeRate(2.0, 1.0)),
+ (input = (q_out = PointMass(1.0), q_α = GammaShapeRate(1.0, 1.0)), output = GammaShapeRate(2.0, 1.0)),
+ (input = (q_out = GammaShapeScale(1.0, 1.0), q_α = PointMass(10.0)), output = GammaShapeRate(11.0, 1.0)),
+ (input = (q_out = GammaShapeScale(1.0, 10.0), q_α = GammaShapeRate(1.0, 1.0)), output = GammaShapeRate(2, 10))
+ ]
+ end
+end

test/rules/gamma_shape_rate/marginals_tests.jl

@@ -0,0 +1,22 @@
+@testitem "marginalrules:GammaShapeRate" begin
+ using ReactiveMP, BayesBase, Random, ExponentialFamily, Distributions
+
+ import ReactiveMP: @test_marginalrules
+
+ @testset "out_α_β: (m_out::GammaDistributionsFamily, m_α::PointMass, m_β::PointMass)" begin
+ @test_marginalrules [check_type_promotion = false] GammaShapeRate(:out_α_β) [
+ (
+ input = (m_out = GammaShapeRate(1.0, 2.0), m_α = PointMass(1.0), m_β = PointMass(2.0)),
+ output = (out = GammaShapeRate(1.0, 4.0), α = PointMass(1.0), β = PointMass(2.0))
+ ),
+ (
+ input = (m_out = GammaShapeScale(2.0, 2.0), m_α = PointMass(2.0), m_β = PointMass(3.0)),
+ output = (out = GammaShapeRate(3.0, 3.5), α = PointMass(2.0), β = PointMass(3.0))
+ ),
+ (
+ input = (m_out = GammaShapeRate(2.0, 3.0), m_α = PointMass(1.0), m_β = PointMass(3.0)),
+ output = (out = GammaShapeRate(2.0, 6.0), α = PointMass(1.0), β = PointMass(3.0))
+ )
+ ]
+ end
+end

test/rules/gamma_shape_rate/out_tests.jl

@@ -0,0 +1,21 @@
+
+@testitem "rules:GammaShapeRate:out" begin
+ using ReactiveMP, BayesBase, Random, ExponentialFamily, Distributions
+
+ import ReactiveMP: @test_rules
+
+ @testset "Belief Propagation: (m_α::Any, m_θ::Any)" begin
+ @test_rules [check_type_promotion = true] GammaShapeRate(:out, Marginalisation) [
+ (input = (m_α = PointMass(1.0), m_β = PointMass(2.0)), output = GammaShapeRate(1.0, 2.0)),
+ (input = (m_α = PointMass(3.0), m_β = PointMass(3.0)), output = GammaShapeRate(3.0, 3.0)),
+ (input = (m_α = PointMass(42.0), m_β = PointMass(42.0)), output = GammaShapeRate(42.0, 42.0))
+ ]
+ end
+
+ @testset "Variational Message Passing: (q_α::Any, q_β::Any)" begin
+ @test_rules [check_type_promotion = true] GammaShapeRate(:out, Marginalisation) [
+ (input = (q_α = PointMass(1.0), q_β = PointMass(2.0)), output = GammaShapeRate(1.0, 2.0)),
+ (input = (q_α = GammaShapeScale(1.0, 1.0), q_β = GammaShapeRate(1.0, 1.0)), output = GammaShapeRate(1.0, 1.0))
+ ]
+ end
+end # testset