Merge pull request #28 from UofTActuarial/dev-0.3.2

Dev 0.3.2

.github/workflows/ci.yml

@@ -1,9 +1,6 @@
 name: CI
 on:
   pull_request:
-    branches:
-      - main
-      - dev
   push:
     branches:
       - main
@@ -17,6 +14,7 @@ jobs:
       fail-fast: false
       matrix:
         version:
+          - '1.6'
           - '1.4' # Replace this with the minimum Julia version that your package supports. E.g. if your package requires Julia 1.5 or higher, change this to '1.5'.
           - '1' # Leave this line unchanged. '1' will automatically expand to the latest stable 1.x release of Julia.
           - 'nightly'

Project.toml

@@ -1,7 +1,7 @@
 name = "LRMoE"
 uuid = "b6847444-3cfa-4637-857f-9c8c615f8351"
 authors = ["Spark Tseung <[email protected]>"]
-version = "0.3.1"
+version = "0.3.2"
 
 [deps]
 Clustering = "aaaa29a8-35af-508c-8bc3-b662a17a0fe5"

src/LRMoE.jl

@@ -79,15 +79,10 @@ export
     # loglikelihood functions
     pdf, logpdf,
     cdf, logcdf,
-    rowlogsumexp, expert_ll_pos,
-    expert_tn_pos,
-    expert_tn_bar_pos,
+    rowlogsumexp,
     expert_ll,
     expert_tn,
     expert_tn_bar,
-    expert_ll_pos_list,
-    expert_tn_pos_list,
-    expert_tn_bar_pos_list,
     expert_ll_list,
     expert_tn_list,
     expert_tn_bar_list, loglik_aggre_dim,

src/experts/continuous/burr.jl

@@ -134,9 +134,14 @@ mean(d::BurrExpert) = mean(LRMoE.Burr(d.k, d.c, d.λ))
 var(d::BurrExpert) = var(LRMoE.Burr(d.k, d.c, d.λ))
 quantile(d::BurrExpert, p) = quantile(LRMoE.Burr(d.k, d.c, d.λ), p)
 function lev(d::BurrExpert, u)
-    uu = 1 / (1 + (u / d.λ)^d.c)
-    return d.λ * gamma(float(1 + 1 / d.c)) * gamma(float(d.k - 1 / d.c)) /
-           gamma(float(d.k)) * beta_inc(1 + 1 / d.c, d.k - 1 / d.c, 1 - uu)[1] + u * uu^d.k
+    if isinf(u)
+        return mean(d)
+    else
+        uu = 1 / (1 + (u / d.λ)^d.c)
+        return d.λ * gamma(float(1 + 1 / d.c)) * gamma(float(d.k - 1 / d.c)) /
+               gamma(float(d.k)) * beta_inc(1 + 1 / d.c, d.k - 1 / d.c, 1 - uu)[1] +
+               u * uu^d.k
+    end
 end
 excess(d::BurrExpert, u) = mean(d) - lev(d, u)
 

src/experts/continuous/gamma.jl

@@ -25,7 +25,7 @@ end
 ## Outer constructors
 GammaExpert(k::Real, θ::Real) = GammaExpert(promote(k, θ)...)
 GammaExpert(k::Integer, θ::Integer) = GammaExpert(float(k), float(θ))
-GammaExpert() = GammaExpert(1.0, 1.0)
+GammaExpert() = GammaExpert(2.0, 1.0)
 
 ## Conversion
 function convert(::Type{GammaExpert{T}}, k::S, θ::S) where {T<:Real,S<:Real}
@@ -135,8 +135,12 @@ mean(d::GammaExpert) = mean(Distributions.Gamma(d.k, d.θ))
 var(d::GammaExpert) = var(Distributions.Gamma(d.k, d.θ))
 quantile(d::GammaExpert, p) = quantile(Distributions.Gamma(d.k, d.θ), p)
 function lev(d::GammaExpert, u)
-    return d.θ * d.k * gamma_inc(float(d.k + 1), u / d.θ, 0)[1] +
-           u * (1 - gamma_inc(float(d.k), u / d.θ, 0)[1])
+    if isinf(u)
+        return mean(d)
+    else
+        return d.θ * d.k * gamma_inc(float(d.k + 1), u / d.θ, 0)[1] +
+               u * (1 - gamma_inc(float(d.k), u / d.θ, 0)[1])
+    end
 end
 excess(d::GammaExpert, u) = mean(d) - lev(d, u)
 

src/experts/continuous/inversegaussian.jl

@@ -141,10 +141,14 @@ mean(d::InverseGaussianExpert) = mean(Distributions.InverseGaussian(d.μ, d.λ))
 var(d::InverseGaussianExpert) = var(Distributions.InverseGaussian(d.μ, d.λ))
 quantile(d::InverseGaussianExpert, p) = quantile(Distributions.InverseGaussian(d.μ, d.λ), p)
 function lev(d::InverseGaussianExpert, u)
-    z = (u - d.μ) / d.μ
-    y = (u + d.μ) / d.μ
-    return u - d.μ * z * cdf.(Normal(0, 1), z * (d.λ / u)^0.5) -
-           d.μ * y * exp(2 * d.λ / d.μ) * cdf.(Normal(0, 1), -y * (d.λ / u)^0.5)
+    if isinf(u)
+        return mean(d)
+    else
+        z = (u - d.μ) / d.μ
+        y = (u + d.μ) / d.μ
+        return u - d.μ * z * cdf.(Normal(0, 1), z * (d.λ / u)^0.5) -
+               d.μ * y * exp(2 * d.λ / d.μ) * cdf.(Normal(0, 1), -y * (d.λ / u)^0.5)
+    end
 end
 excess(d::InverseGaussianExpert, u) = mean(d) - lev(d, u)
 

src/experts/continuous/lognormal.jl

@@ -111,8 +111,12 @@ mean(d::LogNormalExpert) = mean(Distributions.LogNormal(d.μ, d.σ))
 var(d::LogNormalExpert) = var(Distributions.LogNormal(d.μ, d.σ))
 quantile(d::LogNormalExpert, p) = quantile(Distributions.LogNormal(d.μ, d.σ), p)
 function lev(d::LogNormalExpert, u)
-    return exp(d.μ + 0.5 * d.σ^2) * cdf.(Normal(d.μ + d.σ^2, d.σ), log(u)) +
-           u * (1 - cdf.(Normal(d.μ, d.σ), log(u)))
+    if isinf(u)
+        return mean(d)
+    else
+        return exp(d.μ + 0.5 * d.σ^2) * cdf.(Normal(d.μ + d.σ^2, d.σ), log(u)) +
+               u * (1 - cdf.(Normal(d.μ, d.σ), log(u)))
+    end
 end
 excess(d::LogNormalExpert, u) = mean(d) - lev(d, u)
 
@@ -131,7 +135,7 @@ function _int_obs_logY(d::LogNormalExpert, yl, yu, expert_ll)
         return log(yl)
     else
         return exp(-expert_ll) * (
-            d.σ * invsqrt2π * 0.5 * _diff_dens_series(d, yl, yu) +
+            d.σ * invsqrt2π * _diff_dens_series(d, yl, yu) +
             d.μ * _diff_dist_series(d, yl, yu)
         )
     end
@@ -140,7 +144,7 @@ end
 function _int_lat_logY(d::LogNormalExpert, tl, tu, expert_tn_bar)
     return exp(-expert_tn_bar) * (
         d.μ - (
-            d.σ * invsqrt2π * 0.5 * _diff_dens_series(d, tl, tu) +
+            d.σ * invsqrt2π * _diff_dens_series(d, tl, tu) +
             d.μ * _diff_dist_series(d, tl, tu)
         )
     )

src/experts/continuous/weibull.jl

@@ -129,8 +129,12 @@ mean(d::WeibullExpert) = mean(Distributions.Weibull(d.k, d.θ))
 var(d::WeibullExpert) = var(Distributions.Weibull(d.k, d.θ))
 quantile(d::WeibullExpert, p) = quantile(Distributions.Weibull(d.k, d.θ), p)
 function lev(d::WeibullExpert, u)
-    return d.θ * gamma_inc(float(1 / d.k + 1), (u / d.θ)^d.k, 0)[1] *
-           gamma(float(1 / d.k + 1)) + u * exp(-(u / d.θ)^d.k)
+    if isinf(u)
+        return mean(d)
+    else
+        return d.θ * gamma_inc(float(1 / d.k + 1), (u / d.θ)^d.k, 0)[1] *
+               gamma(float(1 / d.k + 1)) + u * exp(-(u / d.θ)^d.k)
+    end
 end
 excess(d::WeibullExpert, u) = mean(d) - lev(d, u)