issue 492: Create distribution that returns <: Union{Int,BigInt} (#497

)

* Make RealUnivariateDistribution type

* test RealUnivariateDistribution

* update method

* reformat

* modify procf

* Explicit use of Distributions

* fix unit test fails

* Discrete valued dists expecting rand return of Union{int,BigInt}

* Make RealUnivariateDistribution type

* test RealUnivariateDistribution

* update method

* reformat

* modify procf

* Explicit use of Distributions

* fix unit test fails

* Discrete valued dists expecting rand return of Union{int,BigInt}

* Typo

* Revert CI test

---------

Co-authored-by: Sam Abbott <[email protected]>

EpiAware/src/EpiAwareUtils/EpiAwareUtils.jl

@@ -10,11 +10,13 @@ using DynamicPPL: Model, fix, condition, @submodel, @model
 using MCMCChains: Chains
 using Random: AbstractRNG, randexp
 using Tables: rowtable
+import Base: eltype
 
 using Distributions, DocStringExtensions, QuadGK, Statistics, Turing
 
 #Export Structures
-export HalfNormal, DirectSample, SafePoisson, SafeNegativeBinomial
+export HalfNormal, DirectSample, SafePoisson, SafeNegativeBinomial, SafeIntValued, SafeInt,
+       SafeDiscreteUnivariateDistribution
 
 #Export functions
 export scan, spread_draws, censored_cdf, censored_pmf, get_param_array, prefix_submodel, ∫F
@@ -32,6 +34,7 @@ include("turing-methods.jl")
 include("DirectSample.jl")
 include("post-inference.jl")
 include("get_param_array.jl")
+include("SafeInt.jl")
 include("SafePoisson.jl")
 include("SafeNegativeBinomial.jl")
 

EpiAware/src/EpiAwareUtils/SafeInt.jl

@@ -0,0 +1,16 @@
+const SafeInt = Union{Int, BigInt}
+
+"""
+A type to represent real-valued distributions, the purpose of this type is to avoid problems
+with the `eltype` function when having `rand` calls in the model.
+"""
+struct SafeIntValued <: Distributions.ValueSupport end
+function Base.eltype(::Type{<:Distributions.Sampleable{F, SafeIntValued}}) where {F}
+    SafeInt
+end
+
+"""
+A constant alias for `Distribution{Univariate, SafeIntValued}`. This type represents a univariate distribution with real-valued outcomes.
+"""
+const SafeDiscreteUnivariateDistribution = Distributions.Distribution{
+    Distributions.Univariate, SafeIntValued}

EpiAware/src/EpiAwareUtils/SafeNegativeBinomial.jl

@@ -65,7 +65,7 @@ var(d)
 2.4617291430060293e40
 ```
 "
-struct SafeNegativeBinomial{T <: Real} <: DiscreteUnivariateDistribution
+struct SafeNegativeBinomial{T <: Real} <: SafeDiscreteUnivariateDistribution
     r::T
     p::T
 

EpiAware/src/EpiAwareUtils/SafePoisson.jl

@@ -45,7 +45,7 @@ var(d)
 7.016735912097631e20
 ```
 "
-struct SafePoisson{T <: Real} <: DiscreteUnivariateDistribution
+struct SafePoisson{T <: Real} <: SafeDiscreteUnivariateDistribution
     λ::T
 
     SafePoisson{T}(λ::Real) where {T <: Real} = new{T}(λ)
@@ -86,7 +86,7 @@ Distributions.rate(d::SafePoisson) = d.λ
 ### Statistics
 
 Distributions.mean(d::SafePoisson) = d.λ
-Distributions.mode(d::SafePoisson) = _safe_int_floor(d.λ)
+Distributions.mode(d::SafePoisson) = floor(d.λ)
 Distributions.var(d::SafePoisson) = d.λ
 Distributions.skewness(d::SafePoisson) = one(typeof(d.λ)) / sqrt(d.λ)
 Distributions.kurtosis(d::SafePoisson) = one(typeof(d.λ)) / d.λ
@@ -229,7 +229,7 @@ function log1pmx(x::Float64)
 end
 
 # Procedure F
-function procf(λ, K::Int, s::Float64)
+function procf(λ, K::SafeInt, s::Float64)
     # can be pre-computed, but does not seem to affect performance
     ω = 0.3989422804014327 / s
     b1 = 0.041666666666666664 / λ

EpiAware/test/EpiAwareUtils/SafeInt.jl

@@ -0,0 +1,8 @@
+@testitem "SafeInt Type Tests" begin
+    using Distributions
+    struct DummySampleable <: Sampleable{Univariate, SafeIntValued} end
+
+    @test SafeIntValued <: Distributions.ValueSupport
+    @test eltype(DummySampleable) <: Union{Int, BigInt}
+    @test SafeDiscreteUnivariateDistribution == Distribution{Univariate, SafeIntValued}
+end

EpiAware/test/EpiAwareUtils/SafeNegativeBinomial.jl

@@ -70,7 +70,7 @@ end
 
     dist = SafeNegativeBinomial(r, p)
     @testset "Large value of mean samples a BigInt with SafePoisson" begin
-        @test rand(dist) isa BigInt
+        @test rand(dist) isa Union{Int, BigInt}
     end
     @testset "Large value of mean sample failure with Poisson" begin
         _dist = EpiAware.EpiAwareUtils._negbin(dist)

EpiAware/test/EpiAwareUtils/SafePoisson.jl

@@ -2,9 +2,9 @@
     λ = 10.0
     dist = SafePoisson(λ)
     @test typeof(dist) <: SafePoisson
-    @test rand(dist) isa Int
-    @test rand(dist, 10) isa Vector{Int}
-    @test rand(dist, 10, 10) isa Array{Int}
+    @test rand(dist) isa SafeInt
+    @test rand(dist, 10) isa Vector{SafeInt}
+    @test rand(dist, 10, 10) isa Array{SafeInt}
 end
 
 @testitem "Check distribution properties of SafePoisson" begin
@@ -54,7 +54,7 @@ end
     bigλ = exp(48.0) #Large value of λ
     dist = SafePoisson(bigλ)
     @testset "Large value of mean samples a BigInt with SafePoisson" begin
-        @test rand(dist) isa BigInt
+        @test rand(dist) isa SafeInt
     end
     @testset "Large value of mean sample failure with Poisson" begin
         _dist = Poisson(dist.λ)

EpiAware/test/EpiObsModels/modifiers/LatentDelay.jl

@@ -175,9 +175,7 @@ end
         ExpGrowthRate,
         Renewal] .|>
                 em_type -> em_type(
-        data = EpiData([0.2, 0.5, 0.3],
-            em_type == Renewal ? softplus : exp
-        ),
+        data = EpiData([0.2, 0.5, 0.3], exp),
         initialisation_prior = Normal(log(100.0), 0.01)
     )
 

EpiAware/test/EpiObsModels/modifiers/PrefixObservationModel.jl

@@ -10,5 +10,5 @@ end
 
     mdl = generate_observations(model, missing, 10)
     draw = rand(mdl)
-    @test typeof(draw[:var"Test.y_t[1]"]) <: Int
+    @test typeof(draw[:var"Test.y_t[1]"]) <: Real
 end