Merge pull request #236 from ReactiveBayes/remove-lv

Remove LV from the package

Author: bvdmitri

.github/workflows/CI.yml

@@ -30,6 +30,7 @@ jobs:
       - uses: actions/checkout@v2
       - uses: julia-actions/setup-julia@v2
         with:
+          include-all-prereleases: true
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}
       - uses: julia-actions/cache@v2
@@ -50,6 +51,8 @@ jobs:
     steps:
       - uses: actions/checkout@v2
       - uses: julia-actions/setup-julia@v2
+        with: 
+          version: '1'
       - uses: actions/cache@v4
         env:
           cache-name: cache-artifacts

Project.toml

@@ -17,7 +17,6 @@ IntervalSets = "8197267c-284f-5f27-9208-e0e47529a953"
 IrrationalConstants = "92d709cd-6900-40b7-9082-c6be49f344b6"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 LogExpFunctions = "2ab3a3ac-af41-5b50-aa03-7779005ae688"
-LoopVectorization = "bdcacae8-1622-11e9-2a5c-532679323890"
 PositiveFactorizations = "85a6dd25-e78a-55b7-8502-1745935b8125"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
@@ -42,7 +41,6 @@ IntervalSets = "0.6, 0.7"
 IrrationalConstants = "0.1, 0.2"
 LinearAlgebra = "1.9"
 LogExpFunctions = "0.3"
-LoopVectorization = "0.12"
 PositiveFactorizations = "0.2"
 Random = "1.9"
 RecursiveArrayTools = "3.26"

src/distributions/categorical.jl

@@ -4,7 +4,6 @@ export logpartition
 import Distributions: Categorical, probs
 import LogExpFunctions: logsumexp
 import FillArrays: OneElement
-using LoopVectorization
 
 BayesBase.vague(::Type{<:Categorical}, dims::Int) = Categorical(ones(dims) ./ dims)
 BayesBase.convert_paramfloattype(::Type{T}, distribution::Categorical) where {T <: Real} = Categorical(convert(AbstractVector{T}, probs(distribution)))
@@ -49,7 +48,7 @@ end
 function (::MeanToNatural{Categorical})(tuple_of_θ::Tuple{Any}, _)
     (p,) = tuple_of_θ
     pₖ = p[end]
-    return (LoopVectorization.vmap(pᵢ -> log(pᵢ / pₖ), p),)
+    return (map(pᵢ -> log(pᵢ / pₖ), p),)
 end
 
 function (::NaturalToMean{Categorical})(tuple_of_η::Tuple{V}, _) where {V <: Vector}
@@ -101,8 +100,8 @@ getgradlogpartition(::NaturalParametersSpace, ::Type{Categorical}, conditioner)
                 )
             )
         end
-        sumη = vmapreduce(exp, +, η)
-        return vmap(d -> exp(d) / sumη, η)
+        sumη = mapreduce(exp, +, η)
+        return map(d -> exp(d) / sumη, η)
     end
 
 getfisherinformation(::NaturalParametersSpace, ::Type{Categorical}, conditioner) =

src/distributions/dirichlet.jl

@@ -5,7 +5,6 @@ import SpecialFunctions: digamma, loggamma, trigamma
 import Base.Broadcast: BroadcastFunction
 
 using FillArrays
-using LoopVectorization
 using StaticArrays
 using LinearAlgebra
 using LogExpFunctions
@@ -20,7 +19,7 @@ function BayesBase.prod(::PreserveTypeProd{Distribution}, left::Dirichlet, right
 end
 
 BayesBase.probvec(dist::Dirichlet) = params(dist)[1]
-BayesBase.std(dist::Dirichlet)     = vmap(sqrt, var(dist))
+BayesBase.std(dist::Dirichlet)     = map(sqrt, var(dist))
 
 BayesBase.mean(::BroadcastFunction{typeof(log)}, dist::Dirichlet)      = digamma.(probvec(dist)) .- digamma(sum(probvec(dist)))
 BayesBase.mean(::BroadcastFunction{typeof(clamplog)}, dist::Dirichlet) = digamma.((clamp(p, tiny, typemax(p)) for p in probvec(dist))) .- digamma(sum(probvec(dist)))
@@ -54,7 +53,7 @@ unpack_parameters(::Type{Dirichlet}, packed) = (packed,)
 isbasemeasureconstant(::Type{Dirichlet}) = ConstantBaseMeasure()
 
 getbasemeasure(::Type{Dirichlet}) = (x) -> one(Float64)
-getsufficientstatistics(::Type{Dirichlet}) = (x -> vmap(log, x),)
+getsufficientstatistics(::Type{Dirichlet}) = (x -> map(log, x),)
 
 getlogpartition(::NaturalParametersSpace, ::Type{Dirichlet}) = (η) -> begin
     (η1,) = unpack_parameters(Dirichlet, η)

src/distributions/dirichlet_collection.jl

@@ -7,7 +7,6 @@ using Distributions
 using SpecialFunctions, LogExpFunctions
 
 import FillArrays: Ones, Eye
-import LoopVectorization: vmap, vmapreduce
 using LinearAlgebra, Random
 
 using BlockArrays: BlockDiagonal
@@ -74,7 +73,7 @@ isbasemeasureconstant(::Type{DirichletCollection}) = ConstantBaseMeasure()
 getbasemeasure(::Type{DirichletCollection}, conditioner) = (x) -> one(Float64)
 getlogbasemeasure(::Type{DirichletCollection}, conditioner) = (x) -> zero(Float64)
 
-getsufficientstatistics(::Type{DirichletCollection}, conditioner) = (x -> vmap(log, x),)
+getsufficientstatistics(::Type{DirichletCollection}, conditioner) = (x -> map(log, x),)
 
 BayesBase.mean(dist::DirichletCollection) = dist.α ./ dist.α0
 BayesBase.mean(::BroadcastFunction{typeof(log)}, dist::DirichletCollection) = digamma.(dist.α) .- digamma.(dist.α0)

src/distributions/laplace.jl

@@ -41,7 +41,7 @@ function BayesBase.prod!(
     (η_right, conditioner_right) = (getnaturalparameters(right), getconditioner(right))
 
     if isapprox(conditioner_left, conditioner_right) && isapprox(conditioner_left, conditioner_container)
-        LoopVectorization.vmap!(+, η_container, η_left, η_right)
+        map!(+, η_container, η_left, η_right)
         return container
     end
 

src/distributions/normal_family/normal_family.jl

@@ -25,7 +25,6 @@ import Random: rand!
 import Distributions: logpdf
 import StatsFuns: invsqrt2π
 
-using LoopVectorization
 using StatsFuns: log2π
 using LinearAlgebra
 using SpecialFunctions
@@ -237,7 +236,7 @@ BayesBase.promote_variate_type(::Type{Multivariate}, ::Type{<:NormalWeightedMean
 
 function Base.convert(::Type{Normal{T}}, dist::UnivariateNormalDistributionsFamily) where {T <: Real}
     mean, std = mean_std(dist)
-    return Normal(convert(T, mean), convert(T, std))
+    return Normal(convert(T, T(mean)), convert(T, T(std)))
 end
 
 function Base.convert(::Type{Normal}, dist::UnivariateNormalDistributionsFamily{T}) where {T <: Real}
@@ -249,7 +248,7 @@ function Base.convert(
     dist::MultivariateNormalDistributionsFamily
 ) where {T <: Real, C <: Distributions.PDMats.PDMat{T, Matrix{T}}, M <: AbstractVector{T}}
     mean, cov = mean_cov(dist)
-    return MvNormal(convert(M, mean), Distributions.PDMats.PDMat(convert(AbstractMatrix{T}, cov)))
+    return MvNormal(convert(M, M(mean)), Distributions.PDMats.PDMat(convert(AbstractMatrix{T}, cov)))
 end
 
 function Base.convert(::Type{MvNormal{T}}, dist::MultivariateNormalDistributionsFamily) where {T <: Real}
@@ -268,7 +267,7 @@ end
 
 function Base.convert(::Type{NormalMeanVariance{T}}, dist::UnivariateNormalDistributionsFamily) where {T <: Real}
     mean, var = mean_var(dist)
-    return NormalMeanVariance(convert(T, mean), convert(T, var))
+    return NormalMeanVariance(convert(T, T(mean)), convert(T, T(var)))
 end
 
 function Base.convert(::Type{MvNormalMeanCovariance{T}}, dist::MultivariateNormalDistributionsFamily) where {T <: Real}
@@ -287,7 +286,7 @@ function Base.convert(
     dist::MultivariateNormalDistributionsFamily
 ) where {T <: Real, M <: AbstractArray{T}, P <: AbstractArray{T}}
     mean, cov = mean_cov(dist)
-    return MvNormalMeanCovariance(convert(M, mean), convert(P, cov))
+    return MvNormalMeanCovariance(convert(M, M(mean)), convert(P, P(cov)))
 end
 
 function Base.convert(::Type{NormalMeanVariance}, dist::UnivariateNormalDistributionsFamily{T}) where {T <: Real}
@@ -302,7 +301,7 @@ end
 
 function Base.convert(::Type{NormalMeanPrecision{T}}, dist::UnivariateNormalDistributionsFamily) where {T <: Real}
     mean, precision = mean_precision(dist)
-    return NormalMeanPrecision(convert(T, mean), convert(T, precision))
+    return NormalMeanPrecision(convert(T, T(mean)), convert(T, T(precision)))
 end
 
 function Base.convert(::Type{MvNormalMeanPrecision{T}}, dist::MultivariateNormalDistributionsFamily) where {T <: Real}
@@ -321,7 +320,7 @@ function Base.convert(
     dist::MultivariateNormalDistributionsFamily
 ) where {T <: Real, M <: AbstractArray{T}, P <: AbstractArray{T}}
     mean, precision = mean_precision(dist)
-    return MvNormalMeanPrecision(convert(M, mean), convert(P, precision))
+    return MvNormalMeanPrecision(convert(M, M(mean)), convert(P, P(precision)))
 end
 
 function Base.convert(::Type{NormalMeanPrecision}, dist::UnivariateNormalDistributionsFamily{T}) where {T <: Real}
@@ -339,7 +338,7 @@ function Base.convert(
     dist::UnivariateNormalDistributionsFamily
 ) where {T <: Real}
     weightedmean, precision = weightedmean_precision(dist)
-    return NormalWeightedMeanPrecision(convert(T, weightedmean), convert(T, precision))
+    return NormalWeightedMeanPrecision(convert(T, T(weightedmean)), convert(T, T(precision)))
 end
 
 function Base.convert(
@@ -361,7 +360,7 @@ function Base.convert(
     dist::MultivariateNormalDistributionsFamily
 ) where {T <: Real, M <: AbstractArray{T}, P <: AbstractArray{T}}
     weightedmean, precision = weightedmean_precision(dist)
-    return MvNormalWeightedMeanPrecision(convert(M, weightedmean), convert(P, precision))
+    return MvNormalWeightedMeanPrecision(convert(M, M(weightedmean)), convert(P, P(precision)))
 end
 
 function Base.convert(
@@ -497,7 +496,7 @@ function BayesBase.rand!(
 ) where {T <: Real}
     randn!(rng, container)
     μ, σ = mean_std(dist)
-    @turbo for i in eachindex(container)
+    for i in eachindex(container)
         container[i] = μ + σ * container[i]
     end
     container

src/distributions/pareto.jl

@@ -41,7 +41,7 @@ function BayesBase.prod!(
     (η_right, conditioner_right) = (getnaturalparameters(right), getconditioner(right))
 
     if isapprox(conditioner_left, conditioner_right) && isapprox(conditioner_left, conditioner_container)
-        LoopVectorization.vmap!(+, η_container, η_left, η_right)
+        map!(+, η_container, η_left, η_right)
         return container
     end
 

src/distributions/wip/multinomial.jl

@@ -3,7 +3,6 @@ export Multinomial
 import Distributions: Multinomial, probs
 import StableRNGs: StableRNG
 using StaticArrays
-using LoopVectorization
 using LogExpFunctions
 
 vague(::Type{<:Multinomial}, n::Int, dims::Int) = Multinomial(n, ones(dims) ./ dims)
@@ -58,7 +57,7 @@ end
 
 function pack_naturalparameters(dist::Multinomial)
     @inbounds p = params(dist)[2]
-    return vmap(log, p / p[end])
+    return map(log, p / p[end])
 end
 
 unpack_naturalparameters(ef::ExponentialFamilyDistribution{Multinomial}) = (getnaturalparameters(ef),)
@@ -69,7 +68,7 @@ function Base.convert(::Type{ExponentialFamilyDistribution}, dist::Multinomial)
 end
 
 function Base.convert(::Type{Distribution}, exponentialfamily::ExponentialFamilyDistribution{Multinomial})
-    expη = vmap(exp, getnaturalparameters(exponentialfamily))
+    expη = map(exp, getnaturalparameters(exponentialfamily))
     p = expη / sum(expη)
     return Multinomial(getconditioner(exponentialfamily), p)
 end
@@ -110,7 +109,7 @@ function fisherinformation(expfamily::ExponentialFamilyDistribution{Multinomial}
     η = getnaturalparameters(expfamily)
     n = getconditioner(expfamily)
     I = Matrix{Float64}(undef, length(η), length(η))
-    seη = sum(vmap(exp, η))
+    seη = mapreduce(exp, +, η)
     @inbounds for i in 1:length(η)
         I[i, i] = exp(η[i]) * (seη - exp(η[i])) / (seη)^2
         @inbounds for j in 1:i-1

src/distributions/wishart.jl

@@ -5,7 +5,6 @@ import Base: ndims, size, convert
 import LinearAlgebra
 import SpecialFunctions: digamma
 import StatsFuns: logmvgamma
-using LoopVectorization
 
 """
     WishartFast{T <: Real, A <: AbstractMatrix{T}} <: ContinuousMatrixDistribution
@@ -38,7 +37,7 @@ BayesBase.params(dist::WishartFast)  = (dist.ν, cholinv(dist.invS))
 BayesBase.mean(dist::WishartFast)    = mean(convert(Wishart, dist))
 BayesBase.var(dist::WishartFast)     = var(convert(Wishart, dist))
 BayesBase.cov(dist::WishartFast)     = cov(convert(Wishart, dist))
-BayesBase.std(dist::WishartFast)     = vmap(sqrt, var(dist))
+BayesBase.std(dist::WishartFast)     = map(sqrt, var(dist))
 BayesBase.mode(dist::WishartFast)    = mode(convert(Wishart, dist))
 BayesBase.entropy(dist::WishartFast) = entropy(convert(Wishart, dist))