add convenience constructors for LocationScaleLowRank

src/AdvancedVI.jl

@@ -191,10 +191,14 @@ include("objectives/elbo/repgradelbo.jl")
 export
     MvLocationScale,
     MeanFieldGaussian,
-    FullRankGaussian,
-    MvLocationScaleLowRank
+    FullRankGaussian
 
 include("families/location_scale.jl")
+
+export
+    MvLocationScaleLowRank,
+    LowRankGaussian
+
 include("families/location_scale_low_rank.jl")
 
 

src/families/location_scale.jl

@@ -120,13 +120,13 @@ function Distributions.cov(q::MvLocationScale)
 end
 
 """
-    FullRankGaussian(location, scale; check_args = true)
+    FullRankGaussian(μ, L; check_args = true)
 
 Construct a Gaussian variational approximation with a dense covariance matrix.
 
 # Arguments
-- `location::AbstractVector{T}`: Mean of the Gaussian.
-- `scale::LinearAlgebra.AbstractTriangular{T}`: Cholesky factor of the covariance of the Gaussian.
+- `μ::AbstractVector{T}`: Mean of the Gaussian.
+- `L::LinearAlgebra.AbstractTriangular{T}`: Cholesky factor of the covariance of the Gaussian.
 
 # Keyword Arguments
 - `check_args`: Check the conditioning of the initial scale (default: `true`).
@@ -142,13 +142,13 @@ function FullRankGaussian(
 end
 
 """
-    MeanFieldGaussian(location, scale; check_args = true)
+    MeanFieldGaussian(μ, L; check_args = true)
 
 Construct a Gaussian variational approximation with a diagonal covariance matrix.
 
 # Arguments
-- `location::AbstractVector{T}`: Mean of the Gaussian.
-- `scale::Diagonal{T}`: Diagonal Cholesky factor of the covariance of the Gaussian.
+- `μ::AbstractVector{T}`: Mean of the Gaussian.
+- `L::Diagonal{T}`: Diagonal Cholesky factor of the covariance of the Gaussian.
 
 # Keyword Arguments
 - `check_args`: Check the conditioning of the initial scale (default: `true`).

src/families/location_scale_low_rank.jl

@@ -3,6 +3,7 @@
     MvLocationLowRankScale(location, scale_diag, scale_factors, dist) <: ContinuousMultivariateDistribution
 
 Variational family with a covariance in the form of a diagonal matrix plus a squared low-rank matrix.
+The rank is given by `size(scale_factors, 2)`.
 
 It generally represents any distribution for which the sampling path can be
 represented as follows:
@@ -135,3 +136,27 @@ function update_variational_params!(
 
     opt_st, params
 end
+
+"""
+    LowRankGaussian(location, scale_diag, scale_factors; check_args = true)
+
+Construct a Gaussian variational approximation with a diagonal plus low-rank covariance matrix.
+
+# Arguments
+- `μ::AbstractVector{T}`: Mean of the Gaussian.
+- `D::Vector{T}`: Diagonal of the scale.
+- `U::Matrix{T}`: Low-rank factors of the scale, where `size(U,2)` is the rank.
+
+# Keyword Arguments
+- `check_args`: Check the conditioning of the initial scale (default: `true`).
+"""
+function LowRankGaussian(
+    μ::AbstractVector{T},
+    D::Vector{T},
+    U::Matrix{T};
+    scale_eps::T = sqrt(eps(T))
+) where {T <: Real}
+    @assert minimum(D) ≥ sqrt(scale_eps) "Initial scale is too small (smallest diagonal scale value is $(minimum(D)). This might result in unstable optimization behavior."
+    q_base = Normal{T}(zero(T), one(T))
+    MvLocationScaleLowRank(μ, D, U, q_base, scale_eps)
+end

test/families/location_scale_low_rank.jl

@@ -14,9 +14,7 @@
         Σ = Diagonal(D.^2) + U*U'
 
         q = if basedist == :gaussian
-            MvLocationScaleLowRank(
-                μ, D, U, Normal{realtype}(zero(realtype), one(realtype))
-            )
+            LowRankGaussian(μ, D, U)
         end
         q_true = if basedist == :gaussian
             MvNormal(μ, Σ)