Replace Turing.PROGRESS with AbstractMCMC.PROGRESS

HISTORY.md

@@ -2,6 +2,8 @@
 
 ## Breaking changes
 
+  - Removed Turing's local `PROGRESS` and `setprogress!`; use `AbstractMCMC.PROGRESS` and `AbstractMCMC.setprogress!` instead.
+
 ### DynamicPPL 0.42
 
 Turing.jl v0.46 brings with it all the underlying changes in DynamicPPL 0.42.

docs/src/api.md

@@ -144,12 +144,6 @@ See the [AD guide](https://turinglang.org/docs/tutorials/docs-10-using-turing-au
 | `AutoMooncake`    | [`ADTypes.AutoMooncake`](@extref)    | Mooncake.jl backend    |
 | `AutoReverseDiff` | [`ADTypes.AutoReverseDiff`](@extref) | ReverseDiff.jl backend |
 
-### Debugging
-
-```@docs
-setprogress!
-```
-
 ### Distributions
 
 These distributions are defined in Turing.jl, but not in Distributions.jl.

src/Turing.jl

@@ -24,21 +24,6 @@ using ADTypes: ADTypes, AutoForwardDiff, AutoReverseDiff, AutoMooncake, AutoEnzy
 
 const DEFAULT_ADTYPE = ADTypes.AutoForwardDiff()
 
-const PROGRESS = Ref(true)
-
-# TODO: remove `PROGRESS` and this function in favour of `AbstractMCMC.PROGRESS`
-"""
-    setprogress!(progress::Bool)
-
-Enable progress logging in Turing if `progress` is `true`, and disable it otherwise.
-"""
-function setprogress!(progress::Bool)
-    @info "[Turing]: progress logging is $(progress ? "enabled" : "disabled") globally"
-    PROGRESS[] = progress
-    AbstractMCMC.setprogress!(progress; silent=true)
-    return progress
-end
-
 # Random probability measures.
 include("stdlib/distributions.jl")
 include("stdlib/RandomMeasures.jl")
@@ -141,8 +126,6 @@ export
     AutoReverseDiff,
     AutoMooncake,
     AutoEnzyme,
-    # Debugging - Turing
-    setprogress!,
     # Distributions
     Flat,
     FlatPos,

src/mcmc/Inference.jl

@@ -18,7 +18,7 @@ using DynamicPPL:
 using Distributions, Libtask, Bijectors
 using FlexiChains: FlexiChains, VNChain
 using LinearAlgebra
-using ..Turing: PROGRESS, Turing
+using ..Turing: Turing
 using StatsFuns: logsumexp
 using Random: AbstractRNG
 using AbstractMCMC: AbstractModel, AbstractSampler

src/mcmc/hmc.jl

@@ -91,7 +91,7 @@ function AbstractMCMC.sample(
     chain_type=DEFAULT_CHAIN_TYPE,
     initial_params=Turing.Inference.init_strategy(sampler),
     initial_state=nothing,
-    progress=PROGRESS[],
+    progress=AbstractMCMC.PROGRESS[],
     nadapts=sampler.n_adapts,
     discard_adapt=true,
     discard_initial=-1,

src/mcmc/particle_mcmc.jl

@@ -134,7 +134,7 @@ function AbstractMCMC.sample(
     check_model=true,
     chain_type=DEFAULT_CHAIN_TYPE,
     initial_params=Turing.Inference.init_strategy(sampler),
-    progress=PROGRESS[],
+    progress=AbstractMCMC.PROGRESS[],
     discard_initial=0,
     thinning=1,
     verbose=false,

src/mcmc/repeat_sampler.jl

@@ -116,7 +116,7 @@ function AbstractMCMC.sample(
     check_model=true,
     initial_params=Turing.Inference.init_strategy(sampler),
     chain_type=DEFAULT_CHAIN_TYPE,
-    progress=PROGRESS[],
+    progress=AbstractMCMC.PROGRESS[],
     verbose=true,
     kwargs...,
 )
@@ -146,7 +146,7 @@ function AbstractMCMC.sample(
     check_model=true,
     initial_params=fill(Turing.Inference.init_strategy(sampler), n_chains),
     chain_type=DEFAULT_CHAIN_TYPE,
-    progress=PROGRESS[],
+    progress=AbstractMCMC.PROGRESS[],
     kwargs...,
 )
     check_model && Turing._check_model(model, sampler)

src/variational/Variational.jl

@@ -18,7 +18,8 @@ using DynamicPPL: DynamicPPL, LogDensityFunction
 using LinearAlgebra
 using LogDensityProblems: LogDensityProblems
 using Random
-using ..Turing: DEFAULT_ADTYPE, PROGRESS
+using ..Turing: DEFAULT_ADTYPE
+import AbstractMCMC
 
 export vi,
     q_locationscale,
@@ -94,7 +95,7 @@ function q_initialize_scale(
         energy = mean(
             map(1:num_samples) do _
                 z = rand(rng, q)
-                LogDensityProblems.logdensity(ldf, z)
+                return LogDensityProblems.logdensity(ldf, z)
             end,
         )
 
@@ -347,7 +348,7 @@ Base.rand(res::VIResult, sz::Integer...) = Base.rand(Random.default_rng(), res,
         ),
         unconstrained::Bool=requires_unconstrained_space(algorithm),
         fix_transforms::Bool=false,
-        show_progress::Bool = Turing.PROGRESS[],
+        show_progress::Bool = AbstractMCMC.PROGRESS[],
         kwargs...
     )
 
@@ -391,7 +392,7 @@ function vi(
     ),
     unconstrained::Bool=requires_unconstrained_space(algorithm),
     fix_transforms::Bool=false,
-    show_progress::Bool=PROGRESS[],
+    show_progress::Bool=AbstractMCMC.PROGRESS[],
     kwargs...,
 )
     transform_strategy = unconstrained ? DynamicPPL.LinkAll() : DynamicPPL.UnlinkAll()

test/mcmc/chains.jl

@@ -9,7 +9,7 @@ using Random: Random, Xoshiro
 using Test
 using Turing
 
-Turing.setprogress!(false)
+AbstractMCMC.setprogress!(false)
 
 # This sampler does nothing (it just stays at the existing state)
 struct StaticSampler <: AbstractMCMC.AbstractSampler end

test/runtests.jl

@@ -4,6 +4,7 @@ using Pkg
 using Random: seed!
 using Test
 using TimerOutputs: TimerOutputs, @timeit
+import AbstractMCMC
 import Turing
 
 # Fix the global Random.seed for reproducibility.
@@ -13,7 +14,7 @@ include("test_utils/models.jl")
 include("test_utils/numerical_tests.jl")
 include("test_utils/sampler.jl")
 
-Turing.setprogress!(false)
+AbstractMCMC.setprogress!(false)
 included_paths, excluded_paths = parse_args(ARGS)
 
 # Filter which tests to run and collect timing and allocations information to show in a

test/variational/vi.jl

@@ -3,7 +3,7 @@ module AdvancedVITests
 using ..Models: gdemo_default
 using ..NumericalTests: check_gdemo
 
-using AbstractMCMC: AbstractMCMC
+import AbstractMCMC
 using AdvancedVI
 using Bijectors: Bijectors
 using Distributions: Dirichlet, Normal
@@ -49,7 +49,9 @@ using Turing.Variational
 
     @testset "default interface" begin
         for q0 in [q_meanfield_gaussian, q_fullrank_gaussian]
-            result = vi(gdemo_default, q0, 100; show_progress=Turing.PROGRESS[], adtype)
+            result = vi(
+                gdemo_default, q0, 100; show_progress=AbstractMCMC.PROGRESS[], adtype
+            )
             @test result isa Turing.Variational.VIResult
             @test rand(result) isa DynamicPPL.VarNamedTuple
             @test rand(result, 2) isa Vector{<:DynamicPPL.VarNamedTuple}
@@ -76,7 +78,7 @@ using Turing.Variational
             q_fullrank_gaussian,
             T;
             algorithm,
-            show_progress=Turing.PROGRESS[],
+            show_progress=AbstractMCMC.PROGRESS[],
         )
         c2 = rand(result, 10)
         @test c2 isa Vector{<:DynamicPPL.VarNamedTuple}
@@ -105,7 +107,7 @@ using Turing.Variational
             q_fullrank_gaussian,
             T;
             algorithm,
-            show_progress=Turing.PROGRESS[],
+            show_progress=AbstractMCMC.PROGRESS[],
         )
 
         N = 1000