Add getparams and setparams!! following AbstractMCMC v5.5 and v…

…5.6 (#103)

* add `getparams` and `setparams!!`

* add BangBang as dep, add functions for `GradientTransition`

* increase atol

* Update test/runtests.jl

Co-authored-by: Penelope Yong <[email protected]>

* update functions with `model` arguments

* fix test errors

* remove BangBang dep

---------

Co-authored-by: Penelope Yong <[email protected]>

Project.toml

@@ -1,6 +1,6 @@
 name = "AdvancedMH"
 uuid = "5b7e9947-ddc0-4b3f-9b55-0d8042f74170"
-version = "0.8.3"
+version = "0.8.4"
 
 [deps]
 AbstractMCMC = "80f14c24-f653-4e6a-9b94-39d6b0f70001"
@@ -23,18 +23,18 @@ AdvancedMHMCMCChainsExt = "MCMCChains"
 AdvancedMHStructArraysExt = "StructArrays"
 
 [compat]
-AbstractMCMC = "5"
+AbstractMCMC = "5.6"
 DiffResults = "1"
 Distributions = "0.25"
 FillArrays = "1"
 ForwardDiff = "0.10"
+LinearAlgebra = "1.6"
 LogDensityProblems = "2"
 MCMCChains = "6.0.4"
+Random = "1.6"
 Requires = "1"
 StructArrays = "0.6"
 julia = "1.6"
-LinearAlgebra = "1.6"
-Random = "1.6"
 
 [extras]
 DiffResults = "163ba53b-c6d8-5494-b064-1a9d43ac40c5"

src/AdvancedMH.jl

@@ -140,6 +140,20 @@ function __init__()
     end
 end
 
+# AbstractMCMC.jl interface
+function AbstractMCMC.getparams(t::Transition)
+    return t.params
+end
+
+# TODO (sunxd): remove `DensityModel` in favor of `AbstractMCMC.LogDensityModel`
+function AbstractMCMC.setparams!!(model::DensityModelOrLogDensityModel, t::Transition, params)
+    return Transition(
+        params,
+        logdensity(model, params),
+        t.accepted
+    )
+end
+
 # Include inference methods.
 include("proposal.jl")
 include("mh-core.jl")

src/MALA.jl

@@ -11,7 +11,7 @@ MALA(d::RandomWalkProposal) = MALA{typeof(d)}(d)
 MALA(d) = MALA(RandomWalkProposal(d))
 
 
-struct GradientTransition{T<:Union{Vector, Real, NamedTuple}, L<:Real, G<:Union{Vector, Real, NamedTuple}} <: AbstractTransition
+struct GradientTransition{T<:Union{Vector,Real,NamedTuple},L<:Real,G<:Union{Vector,Real,NamedTuple}} <: AbstractTransition
     params::T
     lp::L
     gradient::G
@@ -20,6 +20,20 @@ end
 
 logdensity(model::DensityModelOrLogDensityModel, t::GradientTransition) = t.lp
 
+function AbstractMCMC.getparams(t::GradientTransition)
+    return t.params
+end
+
+function AbstractMCMC.setparams!!(model::DensityModelOrLogDensityModel, t::GradientTransition, params)
+    lp, gradient = logdensity_and_gradient(model, params)
+    return GradientTransition(
+        params,
+        lp,
+        gradient,
+        t.accepted
+    )
+end
+
 propose(::Random.AbstractRNG, ::MALA, ::DensityModelOrLogDensityModel) = error("please specify initial parameters")
 function transition(sampler::MALA, model::DensityModelOrLogDensityModel, params, accepted)
     return GradientTransition(params, logdensity_and_gradient(model, params)..., accepted)
@@ -88,6 +102,6 @@ logdensity_and_gradient(::DensityModelOrLogDensityModel, ::Any)
 function logdensity_and_gradient(model::AbstractMCMC.LogDensityModel, params)
     check_capabilities(model)
     return LogDensityProblems.logdensity_and_gradient(model.logdensity, params)
- end
+end
 
 

test/runtests.jl

@@ -1,4 +1,5 @@
 using AdvancedMH
+using AbstractMCMC
 using DiffResults
 using Distributions
 using ForwardDiff
@@ -33,6 +34,25 @@ include("util.jl")
     LogDensityProblems.logdensity(::typeof(density), θ) = density(θ)
     LogDensityProblems.dimension(::typeof(density)) = 2
 
+    @testset "getparams/setparams!! (AbstractMCMC interface)" begin
+        t1, _ = AbstractMCMC.step(Random.default_rng(), model, StaticMH([Normal(0, 1), Normal(0, 1)]))
+        t2, _ = AbstractMCMC.step(Random.default_rng(), model, MALA(x -> MvNormal(x, I)); initial_params=ones(2))
+        for t in [t1, t2]
+            @test AbstractMCMC.getparams(model, t) == t.params
+
+            new_transition = AbstractMCMC.setparams!!(model, t, AbstractMCMC.getparams(model, t))
+            @test new_transition.lp == t.lp
+            @test new_transition.accepted == t.accepted
+            @test new_transition.params == t.params
+            if hasfield(typeof(t), :gradient)
+                @test new_transition.gradient == t.gradient
+            end
+
+            t_replaced = AbstractMCMC.setparams!!(model, t, [1.0, 2.0])
+            @test t_replaced.params == [1.0, 2.0]
+        end
+    end
+
     @testset "StaticMH" begin
         # Set up our sampler with initial parameters.
         spl1 = StaticMH([Normal(0,1), Normal(0, 1)])
@@ -69,7 +89,7 @@ include("util.jl")
         @test mean(chain1.σ) ≈ 1.0 atol=0.1
         @test mean(chain2.μ) ≈ 0.0 atol=0.1
         @test mean(chain2.σ) ≈ 1.0 atol=0.1
-        @test mean(chain3.μ) ≈ 0.0 atol=0.1
+        @test mean(chain3.μ) ≈ 0.0 atol=0.15
         @test mean(chain3.σ) ≈ 1.0 atol=0.1
     end