remove SimpleUnPack dependency

Project.toml

@@ -19,7 +19,6 @@ PDMats = "90014a1f-27ba-587c-ab20-58faa44d9150"
 ProgressMeter = "92933f4c-e287-5a05-a399-4b506db050ca"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
-SimpleUnPack = "ce78b400-467f-4804-87d8-8f486da07d0a"
 StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
 StatsFuns = "4c63d2b9-4356-54db-8cca-17b64c39e42c"
 

docs/Project.toml

@@ -9,7 +9,6 @@ LogDensityProblems = "6fdf6af0-433a-55f7-b3ed-c6c6e0b8df7c"
 Optimisers = "3bd65402-5787-11e9-1adc-39752487f4e2"
 PDMats = "90014a1f-27ba-587c-ab20-58faa44d9150"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
-SimpleUnPack = "ce78b400-467f-4804-87d8-8f486da07d0a"
 
 [compat]
 ADTypes = "0.1.6"

docs/src/advi.md

@@ -116,7 +116,6 @@ StickingTheLandingEntropy
 
 ```@setup stl
 using LogDensityProblems
-using SimpleUnPack
 using PDMats
 using Bijectors
 using LinearAlgebra
@@ -134,7 +133,7 @@ struct NormalLogNormal{MX,SX,MY,SY}
 end
 
 function LogDensityProblems.logdensity(model::NormalLogNormal, θ)
-    @unpack μ_x, σ_x, μ_y, Σ_y = model
+    (; μ_x, σ_x, μ_y, Σ_y) = model
     logpdf(LogNormal(μ_x, σ_x), θ[1]) + logpdf(MvNormal(μ_y, Σ_y), θ[2:end])
 end
 
@@ -151,17 +150,15 @@ n_dims = 10
 σ_x    = exp.(randn())
 μ_y    = randn(n_dims)
 σ_y    = exp.(randn(n_dims))
-model  = NormalLogNormal(μ_x, σ_x, μ_y, PDMats.PDiagMat(σ_y.^2));
+model  = NormalLogNormal(μ_x, σ_x, μ_y, Diagonal(σ_y.^2));
 
 d  = LogDensityProblems.dimension(model);
 μ  = randn(d);
 L  = Diagonal(ones(d));
 q0 = AVI.VIMeanFieldGaussian(μ, L)
 
-model  = NormalLogNormal(μ_x, σ_x, μ_y, PDMats.PDiagMat(σ_y.^2));
-
 function Bijectors.bijector(model::NormalLogNormal)
-    @unpack μ_x, σ_x, μ_y, Σ_y = model
+    (; μ_x, σ_x, μ_y, Σ_y) = model
     Bijectors.Stacked(
         Bijectors.bijector.([LogNormal(μ_x, σ_x), MvNormal(μ_y, Σ_y)]),
         [1:1, 2:1+length(μ_y)])

docs/src/started.md

@@ -27,7 +27,6 @@ ADVI with `Bijectors.Exp` bijectors is able to infer this model exactly.
 Using the `LogDensityProblems` interface, we the model can be defined as follows:
 ```@example advi
 using LogDensityProblems
-using SimpleUnPack
 
 struct NormalLogNormal{MX,SX,MY,SY}
     μ_x::MX
@@ -37,7 +36,7 @@ struct NormalLogNormal{MX,SX,MY,SY}
 end
 
 function LogDensityProblems.logdensity(model::NormalLogNormal, θ)
-    @unpack μ_x, σ_x, μ_y, Σ_y = model
+    (; μ_x, σ_x, μ_y, Σ_y) = model
     logpdf(LogNormal(μ_x, σ_x), θ[1]) + logpdf(MvNormal(μ_y, Σ_y), θ[2:end])
 end
 
@@ -51,14 +50,14 @@ end
 ```
 Let's now instantiate the model
 ```@example advi
-using PDMats
+using LinearAlgebra
 
 n_dims = 10
 μ_x    = randn()
 σ_x    = exp.(randn())
 μ_y    = randn(n_dims)
 σ_y    = exp.(randn(n_dims))
-model  = NormalLogNormal(μ_x, σ_x, μ_y, PDMats.PDiagMat(σ_y.^2));
+model  = NormalLogNormal(μ_x, σ_x, μ_y, Diagonal(σ_y.^2));
 ```
 
 Since the `y` follows a log-normal prior, its support is bounded to be the positive half-space ``\mathbb{R}_+``.
@@ -67,7 +66,7 @@ Thus, we will use [Bijectors](https://github.com/TuringLang/Bijectors.jl) to mat
 using Bijectors
 
 function Bijectors.bijector(model::NormalLogNormal)
-    @unpack μ_x, σ_x, μ_y, Σ_y = model
+    (; μ_x, σ_x, μ_y, Σ_y) = model
     Bijectors.Stacked(
         Bijectors.bijector.([LogNormal(μ_x, σ_x), MvNormal(μ_y, Σ_y)]),
         [1:1, 2:1+length(μ_y)])
@@ -94,8 +93,6 @@ objective   = AVI.ADVI(model, n_montecaro; invbij = b⁻¹)
 ```
 For the variational family, we will use the classic mean-field Gaussian family.
 ```@example advi
-using LinearAlgebra
-
 d = LogDensityProblems.dimension(model);
 μ = randn(d);
 L = Diagonal(ones(d));

src/AdvancedVI.jl

@@ -1,7 +1,6 @@
 
 module AdvancedVI
 
-using SimpleUnPack: @unpack, @pack!
 using Accessors
 
 using Random: AbstractRNG, default_rng

src/distributions/location_scale.jl

@@ -35,42 +35,42 @@ Base.length(q::VILocationScale) = length(q.location)
 Base.size(q::VILocationScale) = size(q.location)
 
 function StatsBase.entropy(q::VILocationScale)
-    @unpack  location, scale, dist = q
+    (; location, scale, dist) = q
     n_dims = length(location)
     n_dims*entropy(dist) + first(logabsdet(scale))
 end
 
 function logpdf(q::VILocationScale, z::AbstractVector{<:Real})
-    @unpack location, scale, dist = q
+    (; location, scale, dist) = q
     sum(zᵢ -> logpdf(dist, zᵢ), scale \ (z - location)) - first(logabsdet(scale))
 end
 
 function _logpdf(q::VILocationScale, z::AbstractVector{<:Real})
-    @unpack location, scale, dist = q
+    (; location, scale, dist) = q
     sum(zᵢ -> logpdf(dist, zᵢ), scale \ (z - location)) - first(logabsdet(scale))
 end
 
 function rand(q::VILocationScale)
-    @unpack location, scale, dist = q
+    (; location, scale, dist) = q
     n_dims = length(location)
     scale*rand(dist, n_dims) + location
 end
 
 function rand(rng::AbstractRNG, q::VILocationScale, num_samples::Int) 
-    @unpack location, scale, dist = q
+    (; location, scale, dist) = q
     n_dims = length(location)
     scale*rand(rng, dist, n_dims, num_samples) .+ location
 end
 
 function _rand!(rng::AbstractRNG, q::VILocationScale, x::AbstractVector{<:Real})
-    @unpack location, scale, dist = q
+    (; location, scale, dist) = q
     rand!(rng, dist, x)
     x .= scale*x
     return x += location
 end
 
 function _rand!(rng::AbstractRNG, q::VILocationScale, x::AbstractMatrix{<:Real})
-    @unpack location, scale, dist = q
+    (; location, scale, dist) = q
     rand!(rng, dist, x)
     x *= scale
     return x += location

test/Project.toml

@@ -14,7 +14,6 @@ Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Random123 = "74087812-796a-5b5d-8853-05524746bad3"
 ReTest = "e0db7c4e-2690-44b9-bad6-7687da720f89"
 ReverseDiff = "37e2e3b7-166d-5795-8a7a-e32c996b4267"
-SimpleUnPack = "ce78b400-467f-4804-87d8-8f486da07d0a"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 Tracker = "9f7883ad-71c0-57eb-9f7f-b5c9e6d3789c"
 Zygote = "e88e6eb3-aa80-5325-afca-941959d7151f"

test/advi_locscale.jl

@@ -8,7 +8,6 @@ using Optimisers
 using Distributions
 using PDMats
 using LinearAlgebra
-using SimpleUnPack: @unpack
 
 struct TestModel{M,L,S}
     model::M
@@ -48,7 +47,7 @@ include("models/utils.jl")
 
             T = 10000
             modelstats = modelconstr(realtype; rng)
-            @unpack model, μ_true, L_true, n_dims, is_meanfield = modelstats
+            (; model, μ_true, L_true, n_dims, is_meanfield) = modelstats
 
             b    = Bijectors.bijector(model)
             b⁻¹  = inverse(b)

test/models/normal.jl

@@ -5,7 +5,7 @@ struct TestMvNormal{M,S}
 end
 
 function LogDensityProblems.logdensity(model::TestMvNormal, θ)
-    @unpack μ, Σ = model
+    (; μ, Σ) = model
     logpdf(MvNormal(μ, Σ), θ)
 end
 

test/models/normallognormal.jl

@@ -7,7 +7,7 @@ struct NormalLogNormal{MX,SX,MY,SY}
 end
 
 function LogDensityProblems.logdensity(model::NormalLogNormal, θ)
-    @unpack μ_x, σ_x, μ_y, Σ_y = model
+    (; μ_x, σ_x, μ_y, Σ_y) = model
     logpdf(LogNormal(μ_x, σ_x), θ[1]) + logpdf(MvNormal(μ_y, Σ_y), θ[2:end])
 end
 
@@ -20,7 +20,7 @@ function LogDensityProblems.capabilities(::Type{<:NormalLogNormal})
 end
 
 function Bijectors.bijector(model::NormalLogNormal)
-    @unpack μ_x, σ_x, μ_y, Σ_y = model
+    (; μ_x, σ_x, μ_y, Σ_y) = model
     Bijectors.Stacked(
         Bijectors.bijector.([LogNormal(μ_x, σ_x), MvNormal(μ_y, Σ_y)]),
         [1:1, 2:1+length(μ_y)])
@@ -56,7 +56,8 @@ function normallognormal_meanfield(realtype; rng = default_rng())
     μ_y  = randn(rng, realtype, n_dims)
     σ_y  = log.(exp.(randn(rng, realtype, n_dims)) .+ 1)
 
-    model = NormalLogNormal(μ_x, σ_x, μ_y, PDMats.PDiagMat(σ_y.^2))
+    #model = NormalLogNormal(μ_x, σ_x, μ_y, PDMats.PDiagMat(σ_y.^2))
+    model = NormalLogNormal(μ_x, σ_x, μ_y, Diagonal(σ_y.^2))
 
     μ = vcat(μ_x, μ_y)
     L = vcat(σ_x, σ_y) |> Diagonal

test/optimize.jl

@@ -6,7 +6,6 @@ using Optimisers
 using Distributions
 using PDMats
 using LinearAlgebra
-using SimpleUnPack: @unpack
 
 struct TestModel{M,L,S}
     model::M