v0.12.0

ForneyLab v0.12.0

Diff since v0.11.4

Closed issues:

• Use PDMats.jl to work with positive-definite matrices (#30)
• Cryptic error messages (#27)
• Write a "how to build a node" demo (#29)
• Improve error message for algo contruction on model with dangling edges (#70)
• How to implement boolean operations/factors for binary random variables, or more generally conditional probability tables? (#100)
• Probability density estimation for marginals parameterised with SampleList (#137)
• ERROR: LoadError: KeyError: key nothing not found (#146)
• Error: No applicable marginal update rule for GaussianMeanVariance node with inbound types: Message{GaussianWeightedMeanPrecision}, Message{GaussianMeanVariance} (#194)
• Multiplication of two categorical variables (#197)
• Incorrect outbound message update rules for GaussianMeanVariance (#198)
• Where are parameters stored in ForneyLab.Bernoulli? (#199)
• ForneyLab.step! must be explicitly imported (#200)

Merged pull requests:

• Refactor rule matching (#188) (@ThijsvdLaar)
• Natural parameter distributions (#189) (@ThijsvdLaar)
• Conjugate variational Inference (#192) (@ThijsvdLaar)
• Rename (#193) (@ThijsvdLaar)
• Update citation file (#195) (@ThijsvdLaar)
• Fix gaussian update (#203) (@ThijsvdLaar)
• Improve error messages (#204) (@ThijsvdLaar)
• Update compat (#205) (@ThijsvdLaar)
• Write And node demo (#206) (@ThijsvdLaar)
• Prepare for release 0.12.0 (#208) (@ThijsvdLaar)

v0.11.4

ForneyLab v0.11.4

Diff since v0.11.3

Closed issues:

• Algorithm construction bug (Nonlinear node) (#166)
• Question: Can ForneyLab do parameter learning and graph structure learning? (#176)
• code generation for repetitive graph structure (#179)
• Belief Propagation (#185)

Merged pull requests:

• Constraints (#178) (@ThijsvdLaar)
• Transition Mixture Node (#180) (@ThijsvdLaar)
• Nonlinear fix (#181) (@albertpod)
• Patch for numerical stability of transition mixture (#184) (@ThijsvdLaar)
• Refactor Nonlinear dimensionality specification (#187) (@ThijsvdLaar)
• Bump FL to version 0.11.4 (#190) (@ThijsvdLaar)
• Update compat for automated release (#191) (@ThijsvdLaar)

v0.11.3

ForneyLab v0.11.3

Diff since v0.11.2

Closed issues:

• Improve stability when inverting positive definite matrices (#86)
• Turing, SampledSignals and ForneyLab compatibility issues (#129)
• After last updates, ForneyLab cannot build message passing algorithm. (#132)
• Scheduler fails for posterior factors surrounding Addition node (#142)
• Confidence Intervals on parameters (#148)
• Free energy code generation using custom parameterized factor nodes (#149)
• Naive beginner's question - categorical variable input (#152)
• Stability of the log(det()) function (#153)
• Resulting Gaussians improper (#156)
• messagePassingAlgorithm not found (#159)
• MethodError: no method matching length(::Variable) (#161)
• KeyError: key :log_pdf not found (#162)

Merged pull requests:

• compat(): Fix for compatibility issue #129 (#130) (@bvdmitri)
• Make FactorNode and Function prod symmetric (#133) (@semihakbayrak)
• Introduce strategy for stable Cholesky factorizations (#134) (@ThijsvdLaar)
• CompatHelper: add new compat entry for "PositiveFactorizations" at version "0.2" (#138) (@github-actions[bot])
• Refactor user interface for EP update specification (#139) (@ThijsvdLaar)
• Multivariate nonlinear function of non Gaussian inputs (#141) (@semihakbayrak)
• fix bug - removePrefix() (#150) (@bartvanerp)
• A more universal entropy calculation method for sampling based inference (#151) (@semihakbayrak)
• Stability - improve log(det()) function (#154) (@bartvanerp)
• Update missed log(det()) functions (#158) (@bartvanerp)
• bump version (#160) (@ThijsvdLaar)
• fix(): Fix for #162 (#163) (@bvdmitri)
• update(): Bump compat entries to be compatible with ReactiveMP.jl (#168) (@bvdmitri)
• update(): Update Project.toml to satisfy Julia General registry auto-merge requirements (#169) (@bvdmitri)
• update(): Add TagBot github workflow for automatic releases and tags (#170) (@bvdmitri)
• Fix probit average energy (#171) (@bartvanerp)
• Migrate from TravisCI to GitHub Actions (#172) (@bvdmitri)
• Change LICENSE.md to MIT (#173) (@bvdmitri)

ForneyLab.jl v0.11.2

• Extended message passing rules around Equality node
• New demo with estimation of player skill ratings
• Bug fixes and improvements

ForneyLab.jl v0.11.1

• Unified message passing algorithm construction
• New approximation method for Nonlinear nodes
• New sampling-based rules for Gamma, Beta, and Dirichlet distributions
• Automated documentation builds

ForneyLab.jl v0.11.0

• Implement support for sequential Monte-Carlo
• Implement support for local variational methods
• Decouple algorithm assembly from code generation
• Extend Nonlinear node implementation to multiple arguments
• Refactor recognition factorization handling

ForneyLab.jl v0.10.0

• New implementation of Nonlinear node using unscented transform
• Changes in update rule naming scheme (Void -> V is substituted by Nothing -> N)
• Code style guide

ForneyLab.jl v0.9.2

• Poisson node implementation
• New documentation system
• Minor fixes (#14, #15, #32, #33)

ForneyLab.jl version 0.9.1

This version incorporates some minor changes that enable more accurate backward propagation through nonlinear nodes.

ForneyLab.jl v0.9.0

This release migrates ForneyLab.jl to Julia v0.7 and v1.0.