Merge branch 'main' into 238-enhance-workflow-clarity-of-the-project

Author: bvdmitri

Project.toml

@@ -1,7 +1,7 @@
 name = "RxInfer"
 uuid = "86711068-29c9-4ff7-b620-ae75d7495b3d"
 authors = ["Bagaev Dmitry <d.v.bagaev@tue.nl> and contributors"]
-version = "2.16.2"
+version = "2.17.0"
 
 [deps]
 BayesBase = "b4ee3484-f114-42fe-b91c-797d54a0c67e"
@@ -34,7 +34,7 @@ MacroTools = "0.5.6"
 Optim = "1.0.0"
 ProgressMeter = "1.0.0"
 Random = "1.9"
-ReactiveMP = "~3.13.0"
+ReactiveMP = "~3.14.0"
 Reexport = "1.2.0"
 Rocket = "1.7.0"
 TupleTools = "1.2.0"

codemeta.json

@@ -9,12 +9,12 @@
     "downloadUrl": "https://github.com/reactivebayes/RxInfer.jl/releases",
     "issueTracker": "https://github.com/reactivebayes/RxInfer.jl/issues",
     "name": "RxInfer.jl",
-    "version": "2.16.2",
+    "version": "2.17.0",
     "description": "Julia package for automated, scalable and efficient Bayesian inference on factor graphs with reactive message passing. ",
     "applicationCategory": "Statistics",
     "developmentStatus": "active",
     "readme": "https://reactivebayes.github.io/RxInfer.jl/stable/",
-    "softwareVersion": "2.16.2",
+    "softwareVersion": "2.17.0",
     "keywords": [
         "Bayesian inference",
         "message passing",

docs/src/manuals/inference/manual.md

@@ -85,7 +85,7 @@ m_n_updates = getmarginals(m_n)
 
 ## [Feeding observations](@id user-guide-manual-inference-observations)
 
-By default (without any extra factorisation constraints) model specification implies Belief Propagation message passing update rules. In case of BP algorithm `RxInfer` package computes an exact Bayesian posteriors with a single message passing iteration. To enforce Belief Propagation message passing update rule for some specific factor node user may use `where { q = FullFactorisation() }` option. Read more in [Model Specification](@ref user-guide-model-specification) section. To perform a message passing iteration we need to pass some data to all our data inputs that were created with [`datavar` function](@ref user-guide-model-specification-data-variables) during model specification.
+By default (without any extra factorisation constraints) model specification implies Belief Propagation message passing update rules. In case of BP algorithm `RxInfer` package computes an exact Bayesian posteriors with a single message passing iteration. To enforce Belief Propagation message passing update rule for some specific factor node user may use `where { q = BetheFactorisation() }` option. Read more in [Model Specification](@ref user-guide-model-specification) section. To perform a message passing iteration we need to pass some data to all our data inputs that were created with [`datavar` function](@ref user-guide-model-specification-data-variables) during model specification.
 
 To feed an observation for a specific data input we use `update!` function:
 

docs/src/manuals/model-specification.md

@@ -334,22 +334,22 @@ where `RecognitionFactorisationConstraint` can be the following
 
 1. `MeanField()`
 
-Automatically specifies a mean-field factorisation
+Automatically specifies a naive mean-field factorisation
 
 Example:
 
 ```julia
 y ~ NormalMeanVariance(y_mean, y_var) where { q = MeanField() }
 ```
 
-2. `FullFactorisation()`
+2. `BetheFactorisation()`
 
-Automatically specifies a full factorisation
+Automatically specifies the Bethe factorisation. For more information on Bethe factorisation, please the section on [Bethe approximation](/docs/src/library/bethe-free-energy.md#bethe-approximationid-lib-bethe-approximation).
 
 Example:
 
 ```julia
-y ~ NormalMeanVariance(y_mean, y_var) where { q = FullFactorisation() }
+y ~ NormalMeanVariance(y_mean, y_var) where { q = BetheFactorisation() }
 ```
 
 3. `q(μ)q(v)q(out)` or `q(μ) * q(v) * q(out)`

src/graphppl.jl

@@ -222,7 +222,9 @@ function write_fconstraint_option(form, variables, fconstraint)
     elseif @capture(fconstraint, MeanField())
         return :(ReactiveMP.MeanField())
     elseif @capture(fconstraint, FullFactorisation())
-        return :(ReactiveMP.FullFactorisation())
+        return :(ReactiveMP.BetheFactorisation())
+    elseif @capture(fconstraint, BetheFactorisation())
+        return :(ReactiveMP.BetheFactorisation())
     else
         error("Invalid factorisation constraint: $fconstraint")
     end

test/constraints/factorisation_constraints_tests.jl

@@ -465,7 +465,7 @@
         # But it would be nice to support this case too
         @test_throws ErrorException ReactiveMP.resolve_factorisation(cs, getvariables(model), TestFactorisationStochastic, (d, z))
         # Deterministic node should ignore `resolve_factorisation` and multiple proxy vars
-        @test ReactiveMP.resolve_factorisation(cs, getvariables(model), TestFactorisationDeterministic, (d, z)) == FullFactorisation()
+        @test ReactiveMP.resolve_factorisation(cs, getvariables(model), TestFactorisationDeterministic, (d, z)) == BetheFactorisation()
     end
 
     ## Warning testing below

test/node_tests.jl

@@ -68,9 +68,9 @@
         @testset "Default functional dependencies" begin
             import ReactiveMP: DefaultFunctionalDependencies
 
-            @testset "Default functional dependencies: FullFactorisation" begin
-                # We test `FullFactorisation` case here
-                m, x, y, z, node = make_dummy_model(FullFactorisation(), DefaultFunctionalDependencies())
+            @testset "Default functional dependencies: BetheFactorisation" begin
+                # We test `BetheFactorisation` case here
+                m, x, y, z, node = make_dummy_model(BetheFactorisation(), DefaultFunctionalDependencies())
 
                 # Test that pipeline dependencies have been set properly
                 @test ReactiveMP.get_pipeline_dependencies(ReactiveMP.getpipeline(node)) === DefaultFunctionalDependencies()
@@ -187,12 +187,12 @@
         @testset "Require inbound message functional dependencies" begin
             import ReactiveMP: RequireMessageFunctionalDependencies
 
-            @testset "Require inbound message functional dependencies: FullFactorisation" begin
+            @testset "Require inbound message functional dependencies: BetheFactorisation" begin
                 # Require inbound message on `x`
                 pipeline = RequireMessageFunctionalDependencies((1,), (NormalMeanVariance(0.123, 0.123),))
 
-                # We test `FullFactorisation` case here
-                m, x, y, z, node = make_dummy_model(FullFactorisation(), pipeline)
+                # We test `BetheFactorisation` case here
+                m, x, y, z, node = make_dummy_model(BetheFactorisation(), pipeline)
 
                 # Test that pipeline dependencies have been set properly
                 @test ReactiveMP.get_pipeline_dependencies(ReactiveMP.getpipeline(node)) === pipeline
@@ -218,8 +218,8 @@
                 # Require inbound message on `y` and `z`
                 pipeline = RequireMessageFunctionalDependencies((2, 3), (NormalMeanVariance(0.123, 0.123), nothing))
 
-                # We test `FullFactorisation` case here
-                m, x, y, z, node = make_dummy_model(FullFactorisation(), pipeline)
+                # We test `BetheFactorisation` case here
+                m, x, y, z, node = make_dummy_model(BetheFactorisation(), pipeline)
 
                 # Test that pipeline dependencies have been set properly
                 @test ReactiveMP.get_pipeline_dependencies(ReactiveMP.getpipeline(node)) === pipeline
@@ -385,12 +385,12 @@
         @testset "Require marginal functional dependencies" begin
             import ReactiveMP: RequireMarginalFunctionalDependencies
 
-            @testset "Require marginal functional dependencies: FullFactorisation" begin
+            @testset "Require marginal functional dependencies: BetheFactorisation" begin
                 # Require marginal on `x`
                 pipeline = RequireMarginalFunctionalDependencies((1,), (NormalMeanVariance(0.123, 0.123),))
 
-                # We test `FullFactorisation` case here
-                m, x, y, z, node = make_dummy_model(FullFactorisation(), pipeline)
+                # We test `BetheFactorisation` case here
+                m, x, y, z, node = make_dummy_model(BetheFactorisation(), pipeline)
 
                 # Test that pipeline dependencies have been set properly
                 @test ReactiveMP.get_pipeline_dependencies(ReactiveMP.getpipeline(node)) === pipeline
@@ -416,8 +416,8 @@
                 # Require marginals on `y` and `z`
                 pipeline = RequireMarginalFunctionalDependencies((2, 3), (NormalMeanVariance(0.123, 0.123), nothing))
 
-                # We test `FullFactorisation` case here
-                m, x, y, z, node = make_dummy_model(FullFactorisation(), pipeline)
+                # We test `BetheFactorisation` case here
+                m, x, y, z, node = make_dummy_model(BetheFactorisation(), pipeline)
 
                 # Test that pipeline dependencies have been set properly
                 @test ReactiveMP.get_pipeline_dependencies(ReactiveMP.getpipeline(node)) === pipeline
@@ -583,11 +583,11 @@
         @testset "Require everything functional dependencies" begin
             import ReactiveMP: RequireEverythingFunctionalDependencies
 
-            @testset "Require everything functional dependencies: FullFactorisation" begin
+            @testset "Require everything functional dependencies: BetheFactorisation" begin
                 pipeline = RequireEverythingFunctionalDependencies()
 
-                # We test `FullFactorisation` case here
-                m, x, y, z, node = make_dummy_model(FullFactorisation(), pipeline)
+                # We test `BetheFactorisation` case here
+                m, x, y, z, node = make_dummy_model(BetheFactorisation(), pipeline)
 
                 # Test that pipeline dependencies have been set properly
                 @test ReactiveMP.get_pipeline_dependencies(ReactiveMP.getpipeline(node)) === pipeline