v5.6.4

ReactiveMP v5.6.4

Diff since v5.6.3

Merged pull requests:

• Robust invS in outgoing Λ-rule of MvNormalMeanPrecision-node (#540) (@skoghoern)
• #549 Add MessageMapping docs (#550) (@abpolym)
• Fix bug average energy of Uninformative type (#553) (@mhidalgoaraya)

Closed issues:

• Add new node implementing Jaakkola & Jordan’s lower bound on sigmoid (log-sigmoid) in Bishop PRML (#507)
• Tutorial for the logistic sigmoid node (#530)
• What is message mapping? (#549)

v5.6.3

ReactiveMP v5.6.3

Diff since v5.6.2

Merged pull requests:

• #525 Add IntegrationTest.yml (#526) (@abpolym)
• implement missing marginal rule multiplication node and added tests (#531) (@RetrospectiveRotations)
• 🤖 Auto-format Julia code (#532) (@github-actions[bot])
• #534 Rename softdot marginal rules test file to include it in tests. (#535) (@abpolym)
• Change dispatch to AbstractVector to allow other vector implementations (#536) (@wouterwln)
• #524 Avoid Vararg deprecation warnings. (#537) (@abpolym)
• Improving codecov (#539) (@abpolym)
• #538 Allow for different numeric types for GammaShapeLikelihood (#544) (@abpolym)
• 541 missing average energy for mixture node (#546) (@abpolym)

Closed issues:

• BFE is Infinity when data connects to deterministic nodes (#317)
• The rules toward variance in Gaussian node produce improper InverseGamma distributions that is incompatible with projection like methods (#420)
• Missing marginal rules multiplication node (#505)
• Deprecation warning: Wrapping Vararg directly in UnionAll is deprecated (#524)
• Add IntegrationTest.yml (#525)
• Incorrectly named tests cause some test files to be ignored (#534)
• Missing Average Energy for Mixture node (#541)

v5.6.2

ReactiveMP v5.6.2

Diff since v5.6.1

Merged pull requests:

• 🤖 Auto-format Julia code (#503) (@github-actions[bot])
• use isapprox in the isonehot (#527) (@bvdmitri)
• 🤖 Auto-format Julia code (#528) (@github-actions[bot])

Closed issues:

• Dispatching issue around non-linear node (Unscented) (#501)

v5.6.1

ReactiveMP v5.6.1

Diff since v5.6.0

Merged pull requests:

• Issue 501 non linear node unscented (#508) (@Raphael-Tresor)
• #432: Check if probvec of Categorical q_out is one-hot encoded vector (#510) (@abpolym)
• Fix Docstring for as_marginal (#516) (@wouterwln)
• #515: Actually filter tests when doing make test test_args=... (#517) (@abpolym)
• Update ForwardDiff to 1. (#521) (@ofSingularMind)
• #522 Skip Aqua.jl checks during selective test runs (#523) (@abpolym)

Closed issues:

• Categorical node expects one-hot encoded vector, but doesn't check it (#432)
• Docstring is misleading for as_marginal (#506)
• Rules: Room for Computational Efficiency Improvements? (#513)
• Makefile: Running reduced testsets does not work (#515)
• Documentation for Makefile is outdated (#519)
• Allow skipping Aqua.jl checks during selective test runs (#522)

v5.6.0

ReactiveMP v5.6.0

Diff since v5.5.12

Merged pull requests:

• use MvNormalMeanScaleMatrixPrecision from ExpFamily package (#509) (@bvdmitri)

Closed issues:

• Missing rand procedure for MvNormalMeanScalePrecision (#387)

v5.5.12

ReactiveMP v5.5.12

Diff since v5.5.11

Merged pull requests:

• Fix q_t1 dimensionality bug (#504) (@wouterwln)

v5.5.11

ReactiveMP v5.5.11

Diff since v5.5.10

Merged pull requests:

• Mv normal mean scale matrix precision (#497) (@ofSingularMind)

v5.5.10

ReactiveMP v5.5.10

Diff since v5.5.9

Merged pull requests:

• 🤖 Auto-format Julia code (#498) (@github-actions[bot])
• Fix #499 missing linearization method (#500) (@bartvanerp)

Closed issues:

• Dispatching issue around non-linear node (linearization) (#499)

v5.5.9

ReactiveMP v5.5.9

Diff since v5.5.8

Merged pull requests:

• Revert "Dont check for proper in division of to accomodate inference" (#496) (@bvdmitri)

v5.5.8

ReactiveMP v5.5.8

Diff since v5.5.7

Merged pull requests:

• Dont check for proper in division of to accomodate inference (#486) (@Nimrais)
• Tighten dispatch for summation rules (#492) (@wouterwln)
• Tighten dispatch for other optimized rules (in place) (#493) (@wouterwln)
• update rule error printing, adds meta spec (#495) (@bvdmitri)