Deepcopy adaptor before starting sampling

This avoids the unintuitive behaviour seen in #379
TuringLang · Nov 5, 2024 · 2a5731c · 2a5731c
1 parent a6f0621
commit 2a5731c
Show file tree

Hide file tree

Showing 3 changed files with 68 additions and 1 deletion.
diff --git a/src/sampler.jl b/src/sampler.jl
@@ -166,6 +166,42 @@ function sample(
     verbose::Bool = true,
     progress::Bool = false,
     (pm_next!)::Function = pm_next!,
+) where {T<:AbstractVecOrMat{<:AbstractFloat}}
+    # Prevent adaptor from being mutated
+    adaptor = deepcopy(adaptor)
+    # Then call sample_mutating_adaptor with the same arguments
+    return sample_mutating_adaptor(
+        rng,
+        h,
+        κ,
+        θ,
+        n_samples,
+        adaptor,
+        n_adapts;
+        drop_warmup = drop_warmup,
+        verbose = verbose,
+        progress = progress,
+        (pm_next!) = pm_next!,
+    )
+end
+
+"""
+    sample_mutating_adaptor(args...; kwargs...)
+
+The same as `sample`, but mutates the `adaptor` argument.
+"""
+function sample_mutating_adaptor(
+    rng::Union{AbstractRNG,AbstractVector{<:AbstractRNG}},
+    h::Hamiltonian,
+    κ::HMCKernel,
+    θ::T,
+    n_samples::Int,
+    adaptor::AbstractAdaptor = NoAdaptation(),
+    n_adapts::Int = min(div(n_samples, 10), 1_000);
+    drop_warmup = false,
+    verbose::Bool = true,
+    progress::Bool = false,
+    (pm_next!)::Function = pm_next!,
 ) where {T<:AbstractVecOrMat{<:AbstractFloat}}
     @assert !(drop_warmup && (adaptor isa Adaptation.NoAdaptation)) "Cannot drop warmup samples if there is no adaptation phase."
     # Prepare containers to store sampling results

diff --git a/test/adaptation.jl b/test/adaptation.jl
@@ -14,7 +14,8 @@ function runnuts(ℓπ, metric; n_samples = 3_000)
     integrator = AdvancedHMC.make_integrator(nuts, step_size)
     κ = AdvancedHMC.make_kernel(nuts, integrator)
     adaptor = AdvancedHMC.make_adaptor(nuts, metric, integrator)
-    samples, stats = sample(h, κ, θ_init, n_samples, adaptor, n_adapts; verbose = false)
+    # Use mutating version of sample() here
+    samples, stats = AdvancedHMC.sample_mutating_adaptor(rng, h, κ, θ_init, n_samples, adaptor, n_adapts; verbose = false)
     return (samples = samples, stats = stats, adaptor = adaptor)
 end
 

diff --git a/test/sampler.jl b/test/sampler.jl
@@ -62,6 +62,7 @@ end
     n_steps = 10
     n_samples = 22_000
     n_adapts = 4_000
+
     @testset "$metricsym" for (metricsym, metric) in Dict(
         :UnitEuclideanMetric => UnitEuclideanMetric(D),
         :DiagEuclideanMetric => DiagEuclideanMetric(D),
@@ -157,6 +158,7 @@ end
             end
         end
     end
+
     @testset "drop_warmup" begin
         nuts = NUTS(0.8)
         metric = DiagEuclideanMetric(D)
@@ -191,4 +193,32 @@ end
         @test length(samples) == n_samples
         @test length(stats) == n_samples
     end
+
+    @testset "reproducibility" begin
+        # Multiple calls to sample() should yield the same results
+        nuts = NUTS(0.8)
+        metric = DiagEuclideanMetric(D)
+        h = Hamiltonian(metric, ℓπ, ∂ℓπ∂θ)
+        integrator = Leapfrog(ϵ)
+        κ = AdvancedHMC.make_kernel(nuts, integrator)
+        adaptor = AdvancedHMC.make_adaptor(nuts, metric, integrator)
+
+        all_samples = []
+        for i = 1:5
+            samples, stats = sample(
+                Random.MersenneTwister(42),
+                h,
+                κ,
+                θ_init,
+                100, # n_samples -- don't need so many
+                adaptor,
+                50, # n_adapts -- likewise
+                verbose = false,
+                progress = false,
+                drop_warmup = true,
+            )
+            push!(all_samples, samples)
+        end
+        @test all(map(s -> s ≈ all_samples[1], all_samples[2:end]))
+    end
 end