v2.5.1

ReactiveMP v2.5.1

Diff since v2.5.0

Closed issues:

• Macros @call_rule and @marginal_rule are broken for Delta node (#210)
• PyPlot.jl based examples crash in CI (#212)

Merged pull requests:

• Add average energy for uniform dist (#208) (@albertpod)
• add GPRegression by SSM demo (#209) (@HoangMHNguyen)
• Fix for the broken @call_rule for nodes that might have a variable number of inputs (#211) (@bvdmitri)
• Update INN demo with Plots.jl (#214) (@bartvanerp)

v2.5.0

ReactiveMP v2.5.0

Diff since v2.4.1

Closed issues:

• Revise mean(inv, ::PointMass) (#191)
• Error reporting for Progress in automatic inference API (#192)
• Wishart marginal not Hermitian in static model (#193)
• InverseWishart encounters a degrees-of-freedom argument error (#195)
• @node should not support interface names with underscores (#203)

Merged pull requests:

• Add MvNormalMeanScalePrecision node (#184) (@albertpod)
• Inverse Gamma Node (#194) (@abpolym)
• Fix mv-precision inference with known mean (#196) (@bvdmitri)
• Add mean(::cholinv) (#197) (@bvdmitri)
• Add Julia 1.8 in tests (#198) (@bvdmitri)
• Update vague for InverseWishart (#199) (@albertpod)
• add Base.size method for pointmass (#200) (@bvdmitri)
• Fix showprogress = true when iterations are not specified (#201) (@bvdmitri)
• fix: do not allow underscores in interfaces names (#204) (@bvdmitri)
• Change default meta spec from Any to Nothing (#206) (@bvdmitri)

v2.4.1

ReactiveMP v2.4.1

Diff since v2.4.0

Closed issues:

• make test testset="path" pass if path does not exist (#183)

Merged pull requests:

• Show the original stack trace for StackOverflowError in the inference function (#189) (@bvdmitri)
• Improve tests infrastructure (#190) (@bvdmitri)

v2.4.0

ReactiveMP v2.4.0

Diff since v2.3.1

Closed issues:

• distribution parameterization by keyword arguments (#1)
• Support of message passing for Boolean functions (#120)
• Sampling with size argument for Multivariate Gaussian Family is broken after new version of Distributions.jl (#174)
• Factorisation constraints do not play nicely with tuple-d variables (e.g. in Gaussian Mixture node) (#179)

Merged pull requests:

• Support of message passing for Boolean functions (#170) (@albertpod)
• Merge latest changes from master to dev_logic (#177) (@bvdmitri)
• Breaking: set returnvars = KeepLast() when iterations are not specified (#178) (@bvdmitri)
• fix constraints specification resolution for tupled variables (#180) (@bvdmitri)
• Add error for incompatible number of arguments in the model spec (#181) (@bvdmitri)
• Fix factorisation constraints for anonymous random variables from deterministic functions (#182) (@bvdmitri)
• Add tests for aliases (#185) (@bvdmitri)

v2.3.1

ReactiveMP v2.3.1

Diff since v2.3.0

Closed issues:

• Linear combination is inconsistent / numerically unstable (#164)
• FE computation for Categoricals fails when the parameter vector has 0 entries (#165)

Merged pull requests:

• chore: change error message for update! (#167) (@bvdmitri)
• Add custom rule layout for CVI approximation (#168) (@bvdmitri)
• fix: add clamplog (#169) (@bvdmitri)
• fix: automatically renormalize Contingency matrix (#173) (@bvdmitri)
• fix preoptimised factorisation constraint rules for nodes with 2 inputs (#175) (@bvdmitri)
• add extra rand methods for mv gaussians (#176) (@bvdmitri)

v2.3.0

ReactiveMP v2.3.0

Diff since v2.2.0

Closed issues:

• Implement RequireInboundMarginal (#151)

Merged pull requests:

• Add Inverse-Wishart (#161) (@albertpod)
• Wishart/InverseWishart messages (#163) (@bvdmitri)

v2.2.0

ReactiveMP v2.2.0

Diff since v2.1.0

Closed issues:

• Add Poisson node (#129)

Merged pull requests:

• feat: new functional dependencies pipeline RequireMarginal (#160) (@bvdmitri)

v2.1.0

ReactiveMP v2.1.0

Diff since v2.0.6

Closed issues:

• Docs should contain list of exported methods (#114)
• Display user-friendly error when indexing random variables within the @model macro (#134)
• Average energy fails for the multiplication node if scalar is the second argument (#135)
• Report where free energy computes NaN (#136)
• Linearize DistProduct for identical types of distributions (#137)
• Error if the same variable has been used both in LHS and RHS of tilde expression (#139)
• Question: Can you set returnvars to return KeepLast() from all RVs? (#142)
• average energy computation for the Wishart node (#148)
• Deprecate dim(d) with size(d, 1) (#157)

Merged pull requests:

• docs: add list of exported methods #114 (#125) (@bvdmitri)
• Add mode calculation for GammaShapeRate distribution (#126) (@wmkouw)
• Clean demo notebooks and docs (#127) (@pitmonticone)
• ci: do not run benchmarks during CI (#128) (@bvdmitri)
• feat: add Uniform distribution node (#130) (@bvdmitri)
• fix: error on unknown Distribution-like factor node (#131) (@bvdmitri)
• Add Poisson node (#132) (@albertpod)
• feat: check for uniqueness of node inputs, fixes #139 (#140) (@bvdmitri)
• Dev broadcasting syntax v2 (#147) (@bvdmitri)
• Fix average energy for Wishart (#149) (@albertpod)
• Clean docs (#150) (@pitmonticone)
• fix: add symmetric marginal rule for scalar multiplication (#152) (@bvdmitri)
• feat: error message when directly indexing variables in a model (#153) (@bvdmitri)
• feat: add diagnostic checks for Bethe Free Energy stream (#154) (@bvdmitri)
• feat: allow more options for returnvars (#155) (@bvdmitri)
• feat: linearise non-analytical product for objects of the same type (#156) (@bvdmitri)
• refactoring: replace dim with size (#158) (@bvdmitri)
• Merge 2.1.0 branch into master (#159) (@bvdmitri)

v2.0.6

ReactiveMP v2.0.6

Diff since v2.0.5

Merged pull requests:

• fix: minor fix for pointmass form constraint (#124) (@bvdmitri)

v2.0.5

ReactiveMP v2.0.5

Diff since v2.0.4

Merged pull requests:

• fix: simplify update! function (#123) (@bvdmitri)