Add MGVI and rename several things in API

NEWS.md

@@ -14,6 +14,8 @@ Several algorithms have changed their names, but also their role:
     changed (no deprecation for the parameter changes). Tuning and
     sample weighting scheme selection have moved to `TransformedMCMC`.
 
+* `PriorToGaussian` has become `PriorToNormal`.
+
 Partial deprecations are available for the above, a lot of old code should
 run more or less unchanged (with deprecation warnings). Also:
 
@@ -27,6 +29,9 @@ run more or less unchanged (with deprecation warnings). Also:
 
 * `MCMCTuningAlgorithm` has been replaced by `MCMCTransformTuning`.
 
+* The `trafo` parameter of algorithms has been renamed to `pretransform`, the
+  `trafo` field in algorithm results has been renamed to `f_pretransform`.
+
 
 ### New features
 
@@ -39,11 +44,13 @@ run more or less unchanged (with deprecation warnings). Also:
   via tunable space transformations instead of tuning covariance matrices
   in proposal distributions.
 
-  MCMC tuning has been split into proposal tuning (algorithms of type
+* MCMC tuning has been split into proposal tuning (algorithms of type
   `MCMCProposalTuning`) and transform turning (algorithms of type
   `MCMCTransformTuning`). Proposal tuning has now a much more limited role
   and often may be `NoMCMCProposalTuning()` (e.g. for `RandomWalk`).
 
+* Added `MGVISampling` for Metric Gaussian Variational Inference.
+
 
 BAT.jl v3.0.0
 -------------

Project.toml

@@ -69,33 +69,36 @@ AdvancedHMC = "0bf59076-c3b1-5ca4-86bd-e02cd72cde3d"
 Cuba = "8a292aeb-7a57-582c-b821-06e4c11590b1"
 Folds = "41a02a25-b8f0-4f67-bc48-60067656b558"
 HDF5 = "f67ccb44-e63f-5c2f-98bd-6dc0ccc4ba2f"
+MGVI = "fdae7790-d271-4276-880d-f72bbddf129c"
 NestedSamplers = "41ceaf6f-1696-4a54-9b49-2e7a9ec3782e"
 Optim = "429524aa-4258-5aef-a3af-852621145aeb"
 Optimization = "7f7a1694-90dd-40f0-9382-eb1efda571ba"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 Transducers = "28d57a85-8fef-5791-bfe6-a80928e7c999"
 UltraNest = "6822f173-b0be-4018-9ee2-28bf56348d09"
 
+
 [extensions]
 BATAdvancedHMCExt = "AdvancedHMC"
 BATCubaExt = "Cuba"
 BATFoldsExt = ["Folds", "Transducers"]
 BATHDF5Ext = "HDF5"
+BATMGVIExt = "MGVI"
 BATNestedSamplersExt = "NestedSamplers"
 BATOptimExt = "Optim"
 BATOptimizationExt = ["Optimization", "ADTypes"]
 BATPlotsExt = "Plots"
 BATUltraNestExt = "UltraNest"
 
 [compat]
-ADTypes = "0.1, 0.2, 1"
 Accessors = "0.1"
 Adapt = "3, 4"
+ADTypes = "0.1, 0.2, 1"
 AdvancedHMC = "0.5, 0.6"
 AffineMaps = "0.2.3, 0.3"
 ArgCheck = "1, 2.0"
 ArraysOfArrays = "0.4, 0.5, 0.6"
-AutoDiffOperators = "0.1.8, 0.2"
+AutoDiffOperators = "0.2.1"
 ChainRulesCore = "0.9.44, 0.10, 1"
 ChangesOfVariables = "0.1.1"
 Clustering = "0.13, 0.14, 0.15"
@@ -132,13 +135,14 @@ MacroTools = "0.5"
 Markdown = "1"
 MeasureBase = "0.12, 0.13, 0.14"
 Measurements = "2"
+MGVI = "0.4"
 NamedArrays = "0.9, 0.10"
 NestedSamplers = "0.8"
 Optim = "0.19,0.20, 0.21, 0.22, 1"
 Optimization = "3, 4"
-PDMats = "0.9, 0.10, 0.11"
 ParallelProcessingTools = "0.4"
 Parameters = "0.12, 0.13"
+PDMats = "0.9, 0.10, 0.11"
 Plots = "1"
 PositiveFactorizations = "0.2"
 Printf = "1"

docs/Project.toml

@@ -9,6 +9,7 @@ IntervalSets = "8197267c-284f-5f27-9208-e0e47529a953"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Literate = "98b081ad-f1c9-55d3-8b20-4c87d4299306"
 Markdown = "d6f4376e-aef5-505a-96c1-9c027394607a"
+MGVI = "fdae7790-d271-4276-880d-f72bbddf129c"
 NestedSamplers = "41ceaf6f-1696-4a54-9b49-2e7a9ec3782e"
 Optim = "429524aa-4258-5aef-a3af-852621145aeb"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"

docs/src/list_of_algorithms.md

@@ -32,7 +32,7 @@ BAT sampling algorithm type: [`TransformedMCMC`](@ref), MCMC algorithm subtype:
 
 ```julia
 import AdvancedHMC, ForwardDiff
-set_batcontext(ad = ADSelector(ForwardDiff))
+set_batcontext(ad = ForwardDiff)
 bat_sample(target, TransformedMCMC(mcalg = HamiltonianMC()))
 ```
 Requires the [AdvancedHMC](https://github.com/TuringLang/AdvancedHMC.jl) Julia package to be loaded explicitly.
@@ -160,7 +160,7 @@ using Optim
 bat_findmode(target, OptimAlg(optalg = Optim.NelderMead()))
 
 import ForwardDiff
-set_batcontext(ad = ADSelector(ForwardDiff))
+set_batcontext(ad = ForwardDiff)
 bat_findmode(target, OptimAlg(optalg = Optim.LBFGS()))
 ```
 

docs/src/stable_api.md

@@ -44,6 +44,7 @@ bat_transform
 
 get_batcontext
 set_batcontext
+log_batdebug
 
 distbind
 distprod
@@ -69,6 +70,7 @@ EffSampleSizeAlgorithm
 EffSampleSizeFromAC
 EvaluatedMeasure
 ExplicitInit
+FixedMGVISchedule
 FixedNBins
 FreedmanDiaconisBinning
 GelmanRubinConvergence
@@ -90,6 +92,7 @@ MCMCInitAlgorithm
 MCMCMultiCycleBurnin
 MCMCProposalTuning
 MCMCTransformTuning
+MGVISampling
 ModeAsDefined
 NoMCMCProposalTuning
 NoMCMCTransformTuning
@@ -98,7 +101,7 @@ OptimizationAlg
 OrderedResampling
 PosteriorMeasure
 PriorSubstitution
-PriorToGaussian
+PriorToNormal
 PriorToUniform
 RAMTuning
 RandomWalk
@@ -120,4 +123,5 @@ BAT.AbstractMedianEstimator
 BAT.AbstractModeEstimator
 BAT.AbstractSamplingAlgorithm
 BAT.ConvergenceTest
+BAT.MGVISchedule
 ```

docs/src/tutorial_lit.jl

@@ -222,10 +222,10 @@ posterior = PosteriorMeasure(likelihood, prior)
 # possible parameter values for the histogram fit.
 #
 # To increase the verbosity level of BAT logging output, you may want to set
-# the Julia logging level for BAT to debug via `ENV["JULIA_DEBUG"] = "BAT"`.
+# the Julia logging level for BAT to debug via `bat_logdebug()`.
 
-#nb ENV["JULIA_DEBUG"] = "BAT"
-#jl ENV["JULIA_DEBUG"] = "BAT"
+#nb bat_logdebug()
+#jl bat_logdebug()
 
 # Now we can generate a set of MCMC samples via [`bat_sample`](@ref). We'll
 # use 4 MCMC chains with 10^5 MC steps in each chain (after tuning/burn-in):

examples/benchmarks/benchmarks.jl

@@ -2,7 +2,7 @@ using BAT, ValueShapes, IntervalSets, Distributions, Plots, EmpiricalDistributio
 using BATTestCases
 using AHMI
 using StatsBase, ArraysOfArrays, LinearAlgebra, LaTeXStrings, QuadGK, PrettyTables, HypothesisTests, Statistics
-ENV["JULIA_DEBUG"] = "BAT"
+bat_logdebug()
 
 function setup_benchmark()
     if(!(("plots1D" in readdir()) && ("plots2D" in readdir()) && ("results" in readdir())))

examples/benchmarks/run_benchmark_ND.jl

@@ -249,7 +249,7 @@ function run_ND_benchmark(;
                     dis,
                     TransformedMCMC(
                         mcalg = algorithm,
-                        trafo = DoNotTransform(),
+                        pretransform = DoNotTransform(),
                         nchains = nchains,
                         nsteps = nsteps,
                         init = init,
@@ -262,7 +262,7 @@ function run_ND_benchmark(;
             elseif isa(algorithm,BAT.HamiltonianMC)
                 mcmc_sample = bat_sample(
                     dis,
-                    TransformedMCMC(mcalg = algorithm, trafo = DoNotTransform(), nchains = nchains, nsteps = nsteps)
+                    TransformedMCMC(mcalg = algorithm, pretransform = DoNotTransform(), nchains = nchains, nsteps = nsteps)
                 ).result
             end
             taf = time()
@@ -276,7 +276,7 @@ function run_ND_benchmark(;
                             dis,
                             TransformedMCMC(
                                 mcalg = algorithm,
-                                trafo = DoNotTransform(),
+                                pretransform = DoNotTransform(),
                                 nchains = nchains,
                                 nsteps = nsteps,
                                 init = init,
@@ -289,7 +289,7 @@ function run_ND_benchmark(;
                     elseif isa(algorithm,BAT.HamiltonianMC)
                         bat_sample(
                             dis,
-                            TransformedMCMC(mcalg = algorithm, trafo = DoNotTransform(), nchains = nchains, nsteps = nsteps)
+                            TransformedMCMC(mcalg = algorithm, pretransform = DoNotTransform(), nchains = nchains, nsteps = nsteps)
                         ).result
                     end
                     taf = time()

examples/benchmarks/utils.jl

@@ -207,10 +207,10 @@ function run1D(
 	)
 
     sample_stats_all = []
-    samples, chains = bat_sample(testfunctions[key].posterior, TransformedMCMC(mcalg = algorithm, trafo = DoNotTransform(), nchains = nchains, nsteps = nsteps))
+    samples, chains = bat_sample(testfunctions[key].posterior, TransformedMCMC(mcalg = algorithm, pretransform = DoNotTransform(), nchains = nchains, nsteps = nsteps))
     for i in 1:n_runs
         time_before = time()
-        samples, chains = bat_sample(testfunctions[key].posterior, TransformedMCMC(mcalg = algorithm, trafo = DoNotTransform(), nchains = nchains, nsteps = nsteps))
+        samples, chains = bat_sample(testfunctions[key].posterior, TransformedMCMC(mcalg = algorithm, pretransform = DoNotTransform(), nchains = nchains, nsteps = nsteps))
         time_after = time()
 
     	h = plot1D(samples,testfunctions,key,sample_stats)# posterior, key, analytical_stats,sample_stats)
@@ -438,10 +438,10 @@ function run2D(
 
     sample_stats_all = []
 
-    samples, stats = bat_sample(testfunctions[key].posterior, TransformedMCMC(mcalg = algorithm, trafo = DoNotTransform(), nchains = nchains, nsteps = nsteps))
+    samples, stats = bat_sample(testfunctions[key].posterior, TransformedMCMC(mcalg = algorithm, pretransform = DoNotTransform(), nchains = nchains, nsteps = nsteps))
     for i in 1:n_runs
         time_before = time()
-        samples, stats = bat_sample(testfunctions[key].posterior, TransformedMCMC(mcalg = algorithm, trafo = DoNotTransform(), nchains = nchains, nsteps = nsteps))
+        samples, stats = bat_sample(testfunctions[key].posterior, TransformedMCMC(mcalg = algorithm, pretransform = DoNotTransform(), nchains = nchains, nsteps = nsteps))
         time_after = time()
 
 		h = plot2D(samples, testfunctions, key, sample_stats)

examples/dev-internal/test_findmode.jl

@@ -59,9 +59,9 @@ my_function(; nt...)
 
 context = get_batcontext()
 target = posterior
-transformed_density, trafo = BAT.transform_and_unshape(PriorToGaussian(), target, context)
-inv_trafo = inverse(trafo)
-initalg = BAT.apply_trafo_to_init(trafo, InitFromTarget())
+transformed_density, f_transform = BAT.transform_and_unshape(PriorToNormal(), target, context)
+inv_trafo = inverse(f_transform)
+initalg = BAT.apply_trafo_to_init(f_transform, InitFromTarget())
 x_init = collect(bat_initval(transformed_density, initalg, context).result)
 
 f = fchain(inv_trafo, logdensityof(target), -)
@@ -83,8 +83,6 @@ rosenbrock(x, p) = (p[1] - x[1])^2 + p[2] * (x[2] - x[1]^2)^2
 f = rosenbrock
 
 
-using AutoDiffOperators
-
 b = Optimization.SciMLBase.NoAD()
 supertype(typeof(b))
 

examples/paper-example/paper_example.jl

@@ -9,10 +9,9 @@ using ArraysOfArrays
 using TypedTables
 using CSV
 import Cuba, AdvancedHMC, ForwardDiff
-using AutoDiffOperators
 #using AHMI
 
-BAT.set_batcontext(ad = ADSelector(ForwardDiff))
+BAT.set_batcontext(ad = ForwardDiff)
 
 
 function log_pdf_poisson(λ::T, k::U) where {T<:Real,U<:Real}

ext/BATCubaExt.jl

@@ -123,7 +123,7 @@ end
 
 function BAT.bat_integrate_impl(target::MeasureLike, algorithm::CubaIntegration, context::BATContext)
     measure = batmeasure(target)
-    transformed_measure, _ = transform_and_unshape(algorithm.trafo, measure, context)
+    transformed_measure, _ = transform_and_unshape(algorithm.pretransform, measure, context)
 
     if !BAT.has_uhc_support(transformed_measure)
         throw(ArgumentError("CUBA integration requires measures are supported only on the unit hypercube"))