remove FillArrays, LogDensityProblemsAD

The removal FillArrays will increase memory consumption for univariate
slice samplers since we now maintain `d` copies of the univariate slice sampler object.
However, since slice samplers are not expected to work for large `d`,
this should be fine.

Author: Red-Portal

Project.toml

@@ -6,10 +6,8 @@ version = "0.7.4"
 AbstractMCMC = "80f14c24-f653-4e6a-9b94-39d6b0f70001"
 Accessors = "7d9f7c33-5ae7-4f3b-8dc6-eff91059b697"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
-FillArrays = "1a297f60-69ca-5386-bcde-b61e274b549b"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 LogDensityProblems = "6fdf6af0-433a-55f7-b3ed-c6c6e0b8df7c"
-LogDensityProblemsAD = "996a588d-648d-4e1f-a8f0-a84b347e47b1"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
 
@@ -23,10 +21,8 @@ SliceSamplingTuringExt = ["Turing"]
 AbstractMCMC = "4, 5"
 Accessors = "0.1"
 Distributions = "0.25"
-FillArrays = "1"
 LinearAlgebra = "1"
 LogDensityProblems = "2"
-LogDensityProblemsAD = "1"
 Random = "1"
 Requires = "1"
 Turing = "0.37, 0.38"

ext/SliceSamplingTuringExt.jl

@@ -2,12 +2,10 @@
 module SliceSamplingTuringExt
 
 if isdefined(Base, :get_extension)
-    using LogDensityProblemsAD
     using Random
     using SliceSampling
     using Turing
 else
-    using ..LogDensityProblemsAD
     using ..Random
     using ..SliceSampling
     using ..Turing

src/SliceSampling.jl

@@ -1,18 +1,13 @@
 
 module SliceSampling
 
-using AbstractMCMC
 using Accessors
+using AbstractMCMC
 using Distributions
-using FillArrays
 using LinearAlgebra
 using LogDensityProblems
 using Random
 
-# The following is necessary because Turing wraps all models with
-# LogDensityProblemsAD by default. So we need access to these types.
-using LogDensityProblemsAD
-
 # reexports
 using AbstractMCMC: sample, MCMCThreads, MCMCDistributed, MCMCSerial
 export sample, MCMCThreads, MCMCDistributed, MCMCSerial
@@ -59,14 +54,6 @@ function initial_sample(::Random.AbstractRNG, ::Any)
     )
 end
 
-# If target is from `LogDensityProblemsAD`, unwrap target before calling `initial_sample`.
-# This is necessary since Turing wraps `DynamicPPL.Model`s when passed to an `externalsampler`.
-function initial_sample(
-    rng::Random.AbstractRNG, wrap::LogDensityProblemsAD.ADGradientWrapper
-)
-    return initial_sample(rng, parent(wrap))
-end
-
 function exceeded_max_prop(max_prop::Int)
     return error(
         "Exceeded maximum number of proposal $(max_prop), ",

src/multivariate/randpermgibbs.jl

@@ -73,7 +73,7 @@ function AbstractMCMC.step(
     unislices       = if sampler.unislice isa AbstractVector
         sampler.unislice
     else
-        Fill(sampler.unislice, d)
+        fill(sampler.unislice, d)
     end
 
     props = zeros(Int, d)