Add DynamicHMC as a sampling backend

ext/ComradeDynamicHMCExt.jl

@@ -0,0 +1,62 @@
+module ComradeDynamicHMCExt
+
+using Comrade
+using DynamicHMC
+
+using DocStringExtensions
+using HypercubeTransform
+using Random
+using LogDensityProblems
+using Serialization
+using StatsBase
+
+
+function DynamicHMC.mcmc_with_warmup(
+        rng, post::Comrade.VLBIPosterior, N;
+        kwargs...
+    )
+
+    if isnothing(Comrade.admode(post))
+        throw(ArgumentError("You must specify an automatic differentiation type in VLBIPosterior with admode kwarg"))
+    else
+        tpost = asflat(post)
+    end
+
+    results = mcmc_with_warmup(rng, tpost, N; kwargs...)
+
+    return PosteriorSamples(
+        transform.(Ref(tpost), eachcol(results.posterior_matrix)),
+        results.tree_statistics;
+        metadata = Dict(
+            :sampler = :DynamicHMC, :ϵ => results.ϵ,
+            :mass_matrix => results.κ
+        )
+    )
+end
+
+function DynamicHMC.mcmc_with_warmup(
+        rng, post::Comrade.VLBIPosterior, N, output::DiskStore;
+        kwargs...
+    )
+
+    (; name, stride) = output
+    stride = min(stride, N)
+    nscans = nsamples ÷ output_stride + (nsamples % output_stride != 0 ? 1 : 0)
+    outbase = joinpath(name, "samples", "output_scan_")
+
+
+    tpost = asflat(post)
+    results = mcmc_with_warmup(rng, tpost, N; kwargs...)
+
+    return PosteriorSamples(
+        transform.(Ref(tpost), eachcol(results.posterior_matrix)),
+        results.tree_statistics;
+        metadata = Dict(
+            :sampler = :DynamicHMC, :ϵ => results.ϵ,
+            :mass_matrix => results.κ
+        )
+    )
+end
+
+
+end