Fix deprecations (#318)

Fixes deprecations of `MvNormal` constructors in Distributions 0.25.11.

Tests pass locally.

Author: devmotion

src/compiler.jl

@@ -133,8 +133,8 @@ This is used mainly to unwrap `NamedDist` distributions and adjust the indices o
 variables.
 
 # Example
-```jldoctest; setup=:(using Distributions)
-julia> _, _, vns = DynamicPPL.unwrap_right_left_vns(MvNormal([1.0, 1.0], [1.0 0.0; 0.0 1.0]), randn(2, 2), @varname(x)); string(vns[end])
+```jldoctest; setup=:(using Distributions, LinearAlgebra)
+julia> _, _, vns = DynamicPPL.unwrap_right_left_vns(MvNormal(ones(2), I), randn(2, 2), @varname(x)); string(vns[end])
 "x[:,2]"
 
 julia> _, _, vns = DynamicPPL.unwrap_right_left_vns(Normal(), randn(1, 2), @varname(x[:])); string(vns[end])

src/loglikelihoods.jl

@@ -145,7 +145,7 @@ y .~ Normal(μ, σ)
 ```
 3. using `MvNormal`:
 ```julia
-y ~ MvNormal(fill(μ, n), Diagonal(fill(σ, n)))
+y ~ MvNormal(fill(μ, n), σ^2 * I)
 ```
 
 In (1) and (2), `y` will be treated as a collection of `n` i.i.d. 1-dimensional variables,

src/varinfo.jl

@@ -34,7 +34,7 @@ Let `md` be an instance of `Metadata`:
 To make `md::Metadata` type stable, all the `md.vns` must have the same symbol
 and distribution type. However, one can have a Julia variable, say `x`, that is a
 matrix or a hierarchical array sampled in partitions, e.g.
-`x[1][:] ~ MvNormal(zeros(2), 1.0); x[2][:] ~ MvNormal(ones(2), 1.0)`, and is managed by
+`x[1][:] ~ MvNormal(zeros(2), I); x[2][:] ~ MvNormal(ones(2), I)`, and is managed by
 a single `md::Metadata` so long as all the distributions on the RHS of `~` are of the
 same type. Type unstable `Metadata` will still work but will have inferior performance.
 When sampling, the first iteration uses a type unstable `Metadata` for all the

test/Project.toml

@@ -22,7 +22,7 @@ Zygote = "e88e6eb3-aa80-5325-afca-941959d7151f"
 AbstractMCMC = "2.1, 3.0"
 AbstractPPL = "0.2"
 Bijectors = "0.9.5"
-Distributions = "< 0.25.11"
+Distributions = "0.25"
 DistributionsAD = "0.6.3"
 Documenter = "0.26.1, 0.27"
 ForwardDiff = "0.10.12"

test/compiler.jl

@@ -206,7 +206,7 @@ end
         # test DPPL#61
         @model function testmodel_missing5(z)
             m ~ Normal()
-            z[1:end] ~ MvNormal(fill(m, length(z)), 1.0)
+            z[1:end] ~ MvNormal(fill(m, length(z)), I)
             return m
         end
         model = testmodel_missing5(rand(10))
@@ -379,7 +379,7 @@ end
 
         # AR1 model. Dynamic prefixing.
         @model function AR1(num_steps, α, μ, σ, ::Type{TV}=Vector{Float64}) where {TV}
-            η ~ MvNormal(num_steps, 1.0)
+            η ~ MvNormal(zeros(num_steps), I)
             δ = sqrt(1 - α^2)
 
             x = TV(undef, num_steps)
@@ -400,7 +400,7 @@ end
             num_obs = length(y)
             @inbounds for i in 1:num_obs
                 x = @submodel $(Symbol("ar1_$i")) AR1(num_steps, α, μ, σ)
-                y[i] ~ MvNormal(x, 0.1)
+                y[i] ~ MvNormal(x, 0.01 * I)
             end
         end
 
@@ -419,7 +419,7 @@ end
         x = Normal()
         @test DynamicPPL.check_tilde_rhs(x) === x
 
-        x = [Laplace(), Normal(), MvNormal(3, 1.0)]
+        x = [Laplace(), Normal(), MvNormal(zeros(3), I)]
         @test DynamicPPL.check_tilde_rhs(x) === x
     end
     @testset "isliteral" begin
@@ -436,14 +436,14 @@ end
         # Verify that we indeed can parse this.
         @test @model(function array_literal_model()
             # `assume` and literal `observe`
-            m ~ MvNormal(2, 1.0)
-            return [10.0, 10.0] ~ MvNormal(m, 0.5 * ones(2))
+            m ~ MvNormal(zeros(2), I)
+            return [10.0, 10.0] ~ MvNormal(m, 0.25 * I)
         end) isa Function
 
         @model function array_literal_model2()
             # `assume` and literal `observe`
-            m ~ MvNormal(2, 1.0)
-            return [10.0, 10.0] ~ MvNormal(m, 0.5 * ones(2))
+            m ~ MvNormal(zeros(2), I)
+            return [10.0, 10.0] ~ MvNormal(m, 0.25 * I)
         end
 
         @test array_literal_model2()() == [10.0, 10.0]

test/context_implementations.jl

@@ -2,10 +2,9 @@
     # https://github.com/TuringLang/DynamicPPL.jl/issues/129
     @testset "#129" begin
         @model function test(x)
-            μ ~ MvNormal(fill(0, 2), 2.0)
+            μ ~ MvNormal(zeros(2), 4 * I)
             z = Vector{Int}(undef, length(x))
-            # `z .~ Categorical.(ps)` cannot be parsed by Julia 1.0
-            (.~)(z, Categorical.(fill([0.5, 0.5], length(x))))
+            z .~ Categorical.(fill([0.5, 0.5], length(x)))
             for i in 1:length(x)
                 x[i] ~ Normal(μ[z[i]], 0.1)
             end

test/loglikelihoods.jl

@@ -1,28 +1,28 @@
 # A collection of models for which the mean-of-means for the posterior should
 # be same.
-@model function gdemo1(x=10 * ones(2), ::Type{TV}=Vector{Float64}) where {TV}
+@model function gdemo1(x=[10.0, 10.0], ::Type{TV}=Vector{Float64}) where {TV}
     # `dot_assume` and `observe`
     m = TV(undef, length(x))
     m .~ Normal()
-    return x ~ MvNormal(m, 0.5)
+    return x ~ MvNormal(m, 0.25 * I)
 end
 
-@model function gdemo2(x=10 * ones(2), ::Type{TV}=Vector{Float64}) where {TV}
+@model function gdemo2(x=[10.0, 10.0], ::Type{TV}=Vector{Float64}) where {TV}
     # `assume` with indexing and `observe`
     m = TV(undef, length(x))
     for i in eachindex(m)
         m[i] ~ Normal()
     end
-    return x ~ MvNormal(m, 0.5)
+    return x ~ MvNormal(m, 0.25 * I)
 end
 
-@model function gdemo3(x=10 * ones(2))
+@model function gdemo3(x=[10.0, 10.0])
     # Multivariate `assume` and `observe`
-    m ~ MvNormal(length(x), 1.0)
-    return x ~ MvNormal(m, 0.5)
+    m ~ MvNormal(zero(x), I)
+    return x ~ MvNormal(m, 0.25 * I)
 end
 
-@model function gdemo4(x=10 * ones(2), ::Type{TV}=Vector{Float64}) where {TV}
+@model function gdemo4(x=[10.0, 10.0], ::Type{TV}=Vector{Float64}) where {TV}
     # `dot_assume` and `observe` with indexing
     m = TV(undef, length(x))
     m .~ Normal()
@@ -33,16 +33,16 @@ end
 
 # Using vector of `length` 1 here so the posterior of `m` is the same
 # as the others.
-@model function gdemo5(x=10 * ones(1))
+@model function gdemo5(x=[10.0])
     # `assume` and `dot_observe`
     m ~ Normal()
     return x .~ Normal(m, 0.5)
 end
 
 @model function gdemo6(::Type{TV}=Vector{Float64}) where {TV}
     # `assume` and literal `observe`
-    m ~ MvNormal(2, 1.0)
-    return [10.0, 10.0] ~ MvNormal(m, 0.5)
+    m ~ MvNormal(zeros(2), I)
+    return [10.0, 10.0] ~ MvNormal(m, 0.25 * I)
 end
 
 @model function gdemo7(::Type{TV}=Vector{Float64}) where {TV}
@@ -76,23 +76,23 @@ end
 end
 
 @model function _likelihood_dot_observe(m, x)
-    return x ~ MvNormal(m, 0.5)
+    return x ~ MvNormal(m, 0.25 * I)
 end
 
-@model function gdemo10(x=10 * ones(2), ::Type{TV}=Vector{Float64}) where {TV}
+@model function gdemo10(x=[10.0, 10.0], ::Type{TV}=Vector{Float64}) where {TV}
     m = TV(undef, length(x))
     m .~ Normal()
 
     # Submodel likelihood
     @submodel _likelihood_dot_observe(m, x)
 end
 
-@model function gdemo11(x=10 * ones(2, 1), ::Type{TV}=Vector{Float64}) where {TV}
+@model function gdemo11(x=fill(10.0, 2, 1), ::Type{TV}=Vector{Float64}) where {TV}
     m = TV(undef, length(x))
     m .~ Normal()
 
     # Dotted observe for `Matrix`.
-    return x .~ MvNormal(m, 0.5)
+    return x .~ MvNormal(m, 0.25 * I)
 end
 
 const gdemo_models = (

test/prob_macro.jl

@@ -30,15 +30,15 @@
     @testset "vector" begin
         n = 5
         @model function demo(x, n=n)
-            m ~ MvNormal(n, 1.0)
-            return x ~ MvNormal(m, 1.0)
+            m ~ MvNormal(zeros(n), I)
+            return x ~ MvNormal(m, I)
         end
         mval = rand(n)
         xval = rand(n)
         iters = 1000
 
-        logprior = logpdf(MvNormal(n, 1.0), mval)
-        loglike = logpdf(MvNormal(mval, 1.0), xval)
+        logprior = logpdf(MvNormal(zeros(n), I), mval)
+        loglike = logpdf(MvNormal(mval, I), xval)
         logjoint = logprior + loglike
 
         model = demo(xval)

test/turing/compiler.jl

@@ -175,12 +175,12 @@
 
         res = sample(vdemo1b(x), alg, 250)
 
-        D = 2
         @model function vdemo2(x)
-            μ ~ MvNormal(zeros(D), ones(D))
-            @. x ~ $(MvNormal(μ, ones(D)))
+            μ ~ MvNormal(zeros(size(x, 1)), I)
+            @. x ~ $(MvNormal(μ, I))
         end
 
+        D = 2
         alg = HMC(0.01, 5)
         res = sample(vdemo2(randn(D, 100)), alg, 250)
 
@@ -206,7 +206,7 @@
 
         t_vec = @elapsed res = sample(vdemo4(), alg, 1000)
 
-        @model vdemo5() = x ~ MvNormal(zeros(N), 2 * ones(N))
+        @model vdemo5() = x ~ MvNormal(zeros(N), 4 * I)
 
         t_mv = @elapsed res = sample(vdemo5(), alg, 1000)
 
@@ -243,12 +243,12 @@
         x = randn(100)
         res = sample(vdemo1(x), alg, 250)
 
-        D = 2
         @model function vdemo2(x)
-            μ ~ MvNormal(zeros(D), ones(D))
-            return x .~ MvNormal(μ, ones(D))
+            μ ~ MvNormal(zeros(size(x, 1)), I)
+            return x .~ MvNormal(μ, I)
         end
 
+        D = 2
         alg = HMC(0.01, 5)
         res = sample(vdemo2(randn(D, 100)), alg, 250)
 
@@ -274,7 +274,7 @@
 
         t_vec = @elapsed res = sample(vdemo4(), alg, 1000)
 
-        @model vdemo5() = x ~ MvNormal(zeros(N), 2 * ones(N))
+        @model vdemo5() = x ~ MvNormal(zeros(N), 4 * I)
 
         t_mv = @elapsed res = sample(vdemo5(), alg, 1000)
 

test/turing/model.jl

@@ -14,7 +14,7 @@
         end
 
         @model function demo2(xs)
-            m ~ MvNormal(2, 1.0)
+            m ~ MvNormal(zeros(2), I)
 
             for i in eachindex(xs)
                 xs[i] ~ Normal(m[1], 1.0)
@@ -50,7 +50,7 @@
         @model function demo3(xs, ::Type{TV}=Vector{Float64}) where {TV}
             m = Vector{TV}(undef, 2)
             for i in 1:length(m)
-                m[i] ~ MvNormal(2, 1.0)
+                m[i] ~ MvNormal(zeros(2), I)
             end
 
             for i in eachindex(xs)
@@ -63,7 +63,7 @@
         @model function demo4(xs, ::Type{TV}=Vector{Vector{Float64}}) where {TV}
             m = TV(undef, 2)
             for i in 1:length(m)
-                m[i] ~ MvNormal(2, 1.0)
+                m[i] ~ MvNormal(zeros(2), I)
             end
 
             for i in eachindex(xs)