Remove Turing integration tests

.github/workflows/CI.yml

@@ -73,7 +73,6 @@ jobs:
 
       - uses: julia-actions/julia-runtest@v1
         env:
-          GROUP: All
           JULIA_NUM_THREADS: ${{ matrix.runner.num_threads }}
 
       - uses: julia-actions/julia-processcoverage@v1

.github/workflows/CompatHelper.yml

@@ -14,4 +14,4 @@ jobs:
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           COMPATHELPER_PRIV: ${{ secrets.DOCUMENTER_KEY }}
-        run: julia -e 'using CompatHelper; CompatHelper.main(; subdirs = ["", "docs", "test", "test/turing"])'
+        run: julia -e 'using CompatHelper; CompatHelper.main(; subdirs = ["", "docs", "test"])'

.github/workflows/JuliaPre.yml

@@ -25,5 +25,3 @@ jobs:
       - uses: julia-actions/cache@v2
       - uses: julia-actions/julia-buildpkg@v1
       - uses: julia-actions/julia-runtest@v1
-        env:
-          GROUP: DynamicPPL

Project.toml

@@ -1,6 +1,6 @@
 name = "DynamicPPL"
 uuid = "366bfd00-2699-11ea-058f-f148b4cae6d8"
-version = "0.31.5"
+version = "0.32.0"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"

docs/src/api.md

@@ -279,12 +279,9 @@ VarInfo
 TypedVarInfo
 ```
 
-One main characteristic of [`VarInfo`](@ref) is that samples are stored in a linearized form.
-
-```@docs
-link!
-invlink!
-```
+One main characteristic of [`VarInfo`](@ref) is that samples are transformed to unconstrained Euclidean space and stored in a linearized form, as described in the [transformation page](internals/transformations.md).
+The [Transformations section below](#Transformations) describes the methods used for this.
+In the specific case of `VarInfo`, it keeps track of whether samples have been transformed by setting flags on them, using the following functions.
 
 ```@docs
 set_flag!

src/test_utils/models.jl

@@ -323,28 +323,30 @@ function varnames(model::Model{typeof(demo_assume_dot_observe)})
     return [@varname(s), @varname(m)]
 end
 
-@model function demo_assume_observe_literal()
-    # `assume` and literal `observe`
+@model function demo_assume_multivariate_observe_literal()
+    # multivariate `assume` and literal `observe`
     s ~ product_distribution([InverseGamma(2, 3), InverseGamma(2, 3)])
     m ~ MvNormal(zeros(2), Diagonal(s))
     [1.5, 2.0] ~ MvNormal(m, Diagonal(s))
 
     return (; s=s, m=m, x=[1.5, 2.0], logp=getlogp(__varinfo__))
 end
-function logprior_true(model::Model{typeof(demo_assume_observe_literal)}, s, m)
+function logprior_true(model::Model{typeof(demo_assume_multivariate_observe_literal)}, s, m)
     s_dist = product_distribution([InverseGamma(2, 3), InverseGamma(2, 3)])
     m_dist = MvNormal(zeros(2), Diagonal(s))
     return logpdf(s_dist, s) + logpdf(m_dist, m)
 end
-function loglikelihood_true(model::Model{typeof(demo_assume_observe_literal)}, s, m)
+function loglikelihood_true(
+    model::Model{typeof(demo_assume_multivariate_observe_literal)}, s, m
+)
     return logpdf(MvNormal(m, Diagonal(s)), [1.5, 2.0])
 end
 function logprior_true_with_logabsdet_jacobian(
-    model::Model{typeof(demo_assume_observe_literal)}, s, m
+    model::Model{typeof(demo_assume_multivariate_observe_literal)}, s, m
 )
     return _demo_logprior_true_with_logabsdet_jacobian(model, s, m)
 end
-function varnames(model::Model{typeof(demo_assume_observe_literal)})
+function varnames(model::Model{typeof(demo_assume_multivariate_observe_literal)})
     return [@varname(s), @varname(m)]
 end
 
@@ -377,26 +379,50 @@ function varnames(model::Model{typeof(demo_dot_assume_observe_index_literal)})
     return [@varname(s[1]), @varname(s[2]), @varname(m[1]), @varname(m[2])]
 end
 
-@model function demo_assume_literal_dot_observe()
+@model function demo_assume_observe_literal()
+    # univariate `assume` and literal `observe`
+    s ~ InverseGamma(2, 3)
+    m ~ Normal(0, sqrt(s))
+    1.5 ~ Normal(m, sqrt(s))
+    2.0 ~ Normal(m, sqrt(s))
+
+    return (; s=s, m=m, x=[1.5, 2.0], logp=getlogp(__varinfo__))
+end
+function logprior_true(model::Model{typeof(demo_assume_observe_literal)}, s, m)
+    return logpdf(InverseGamma(2, 3), s) + logpdf(Normal(0, sqrt(s)), m)
+end
+function loglikelihood_true(model::Model{typeof(demo_assume_observe_literal)}, s, m)
+    return logpdf(Normal(m, sqrt(s)), 1.5) + logpdf(Normal(m, sqrt(s)), 2.0)
+end
+function logprior_true_with_logabsdet_jacobian(
+    model::Model{typeof(demo_assume_observe_literal)}, s, m
+)
+    return _demo_logprior_true_with_logabsdet_jacobian(model, s, m)
+end
+function varnames(model::Model{typeof(demo_assume_observe_literal)})
+    return [@varname(s), @varname(m)]
+end
+
+@model function demo_assume_dot_observe_literal()
     # `assume` and literal `dot_observe`
     s ~ InverseGamma(2, 3)
     m ~ Normal(0, sqrt(s))
     [1.5, 2.0] .~ Normal(m, sqrt(s))
 
     return (; s=s, m=m, x=[1.5, 2.0], logp=getlogp(__varinfo__))
 end
-function logprior_true(model::Model{typeof(demo_assume_literal_dot_observe)}, s, m)
+function logprior_true(model::Model{typeof(demo_assume_dot_observe_literal)}, s, m)
     return logpdf(InverseGamma(2, 3), s) + logpdf(Normal(0, sqrt(s)), m)
 end
-function loglikelihood_true(model::Model{typeof(demo_assume_literal_dot_observe)}, s, m)
+function loglikelihood_true(model::Model{typeof(demo_assume_dot_observe_literal)}, s, m)
     return loglikelihood(Normal(m, sqrt(s)), [1.5, 2.0])
 end
 function logprior_true_with_logabsdet_jacobian(
-    model::Model{typeof(demo_assume_literal_dot_observe)}, s, m
+    model::Model{typeof(demo_assume_dot_observe_literal)}, s, m
 )
     return _demo_logprior_true_with_logabsdet_jacobian(model, s, m)
 end
-function varnames(model::Model{typeof(demo_assume_literal_dot_observe)})
+function varnames(model::Model{typeof(demo_assume_dot_observe_literal)})
     return [@varname(s), @varname(m)]
 end
 
@@ -574,8 +600,9 @@ const DemoModels = Union{
     Model{typeof(demo_assume_multivariate_observe)},
     Model{typeof(demo_dot_assume_observe_index)},
     Model{typeof(demo_assume_dot_observe)},
-    Model{typeof(demo_assume_literal_dot_observe)},
+    Model{typeof(demo_assume_dot_observe_literal)},
     Model{typeof(demo_assume_observe_literal)},
+    Model{typeof(demo_assume_multivariate_observe_literal)},
     Model{typeof(demo_dot_assume_observe_index_literal)},
     Model{typeof(demo_assume_submodel_observe_index_literal)},
     Model{typeof(demo_dot_assume_observe_submodel)},
@@ -585,7 +612,9 @@ const DemoModels = Union{
 }
 
 const UnivariateAssumeDemoModels = Union{
-    Model{typeof(demo_assume_dot_observe)},Model{typeof(demo_assume_literal_dot_observe)}
+    Model{typeof(demo_assume_dot_observe)},
+    Model{typeof(demo_assume_dot_observe_literal)},
+    Model{typeof(demo_assume_observe_literal)},
 }
 function posterior_mean(model::UnivariateAssumeDemoModels)
     return (s=49 / 24, m=7 / 6)
@@ -609,7 +638,7 @@ const MultivariateAssumeDemoModels = Union{
     Model{typeof(demo_assume_index_observe)},
     Model{typeof(demo_assume_multivariate_observe)},
     Model{typeof(demo_dot_assume_observe_index)},
-    Model{typeof(demo_assume_observe_literal)},
+    Model{typeof(demo_assume_multivariate_observe_literal)},
     Model{typeof(demo_dot_assume_observe_index_literal)},
     Model{typeof(demo_assume_submodel_observe_index_literal)},
     Model{typeof(demo_dot_assume_observe_submodel)},
@@ -759,9 +788,10 @@ const DEMO_MODELS = (
     demo_assume_multivariate_observe(),
     demo_dot_assume_observe_index(),
     demo_assume_dot_observe(),
-    demo_assume_observe_literal(),
+    demo_assume_multivariate_observe_literal(),
     demo_dot_assume_observe_index_literal(),
-    demo_assume_literal_dot_observe(),
+    demo_assume_dot_observe_literal(),
+    demo_assume_observe_literal(),
     demo_assume_submodel_observe_index_literal(),
     demo_dot_assume_observe_submodel(),
     demo_dot_assume_dot_observe_matrix(),

src/threadsafe.jl

@@ -79,8 +79,6 @@ setval!(vi::ThreadSafeVarInfo, val, vn::VarName) = setval!(vi.varinfo, val, vn)
 keys(vi::ThreadSafeVarInfo) = keys(vi.varinfo)
 haskey(vi::ThreadSafeVarInfo, vn::VarName) = haskey(vi.varinfo, vn)
 
-link!(vi::ThreadSafeVarInfo, spl::AbstractSampler) = link!(vi.varinfo, spl)
-invlink!(vi::ThreadSafeVarInfo, spl::AbstractSampler) = invlink!(vi.varinfo, spl)
 islinked(vi::ThreadSafeVarInfo, spl::AbstractSampler) = islinked(vi.varinfo, spl)
 
 function link!!(

src/varinfo.jl

@@ -1221,27 +1221,6 @@ function link!!(
     return Accessors.@set vi.varinfo = DynamicPPL.link!!(t, vi.varinfo, spl, model)
 end
 
-"""
-    link!(vi::VarInfo, spl::Sampler)
-
-Transform the values of the random variables sampled by `spl` in `vi` from the support
-of their distributions to the Euclidean space and set their corresponding `"trans"`
-flag values to `true`.
-"""
-function link!(vi::VarInfo, spl::AbstractSampler)
-    Base.depwarn(
-        "`link!(varinfo, sampler)` is deprecated, use `link!!(varinfo, sampler, model)` instead.",
-        :link!,
-    )
-    return _link!(vi, spl)
-end
-function link!(vi::VarInfo, spl::AbstractSampler, spaceval::Val)
-    Base.depwarn(
-        "`link!(varinfo, sampler, spaceval)` is deprecated, use `link!!(varinfo, sampler, model)` instead.",
-        :link!,
-    )
-    return _link!(vi, spl, spaceval)
-end
 function _link!(vi::UntypedVarInfo, spl::AbstractSampler)
     # TODO: Change to a lazy iterator over `vns`
     vns = _getvns(vi, spl)
@@ -1319,29 +1298,6 @@ function maybe_invlink_before_eval!!(vi::VarInfo, context::AbstractContext, mode
     return maybe_invlink_before_eval!!(t, vi, context, model)
 end
 
-"""
-    invlink!(vi::VarInfo, spl::AbstractSampler)
-
-Transform the values of the random variables sampled by `spl` in `vi` from the
-Euclidean space back to the support of their distributions and sets their corresponding
-`"trans"` flag values to `false`.
-"""
-function invlink!(vi::VarInfo, spl::AbstractSampler)
-    Base.depwarn(
-        "`invlink!(varinfo, sampler)` is deprecated, use `invlink!!(varinfo, sampler, model)` instead.",
-        :invlink!,
-    )
-    return _invlink!(vi, spl)
-end
-
-function invlink!(vi::VarInfo, spl::AbstractSampler, spaceval::Val)
-    Base.depwarn(
-        "`invlink!(varinfo, sampler, spaceval)` is deprecated, use `invlink!!(varinfo, sampler, model)` instead.",
-        :invlink!,
-    )
-    return _invlink!(vi, spl, spaceval)
-end
-
 function _invlink!(vi::UntypedVarInfo, spl::AbstractSampler)
     vns = _getvns(vi, spl)
     if istrans(vi, vns[1])

test/Project.toml

@@ -20,6 +20,7 @@ LogDensityProblemsAD = "996a588d-648d-4e1f-a8f0-a84b347e47b1"
 MCMCChains = "c7f686f2-ff18-58e9-bc7b-31028e88f75d"
 MacroTools = "1914dd2f-81c6-5fcd-8719-6d5c9610ff09"
 Mooncake = "da2b9cff-9c12-43a0-ae48-6db2b0edb7d6"
+OrderedCollections = "bac558e1-5e72-5ebc-8fee-abe8a469f55d"
 Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 ReverseDiff = "37e2e3b7-166d-5795-8a7a-e32c996b4267"

test/ad.jl

@@ -34,4 +34,43 @@
             end
         end
     end
+
+    @testset "Turing#2151: ReverseDiff compilation & eltype(vi, spl)" begin
+        # Failing model
+        t = 1:0.05:8
+        σ = 0.3
+        y = @. rand(sin(t) + Normal(0, σ))
+        @model function state_space(y, TT, ::Type{T}=Float64) where {T}
+            # Priors 
+            α ~ Normal(y[1], 0.001)
+            τ ~ Exponential(1)
+            η ~ filldist(Normal(0, 1), TT - 1)
+            σ ~ Exponential(1)
+            # create latent variable
+            x = Vector{T}(undef, TT)
+            x[1] = α
+            for t in 2:TT
+                x[t] = x[t - 1] + η[t - 1] * τ
+            end
+            # measurement model
+            y ~ MvNormal(x, σ^2 * I)
+            return x
+        end
+        model = state_space(y, length(t))
+
+        # Dummy sampling algorithm for testing. The test case can only be replicated
+        # with a custom sampler, it doesn't work with SampleFromPrior(). We need to
+        # overload assume so that model evaluation doesn't fail due to a lack
+        # of implementation
+        struct MyEmptyAlg end
+        DynamicPPL.getspace(::DynamicPPL.Sampler{MyEmptyAlg}) = ()
+        DynamicPPL.assume(rng, ::DynamicPPL.Sampler{MyEmptyAlg}, dist, vn, vi) =
+            DynamicPPL.assume(dist, vn, vi)
+
+        # Compiling the ReverseDiff tape used to fail here
+        spl = Sampler(MyEmptyAlg())
+        vi = VarInfo(model)
+        ldf = DynamicPPL.LogDensityFunction(vi, model, SamplingContext(spl))
+        @test LogDensityProblemsAD.ADgradient(AutoReverseDiff(; compile=true), ldf) isa Any
+    end
 end