v4.0.0

RxInfer v4.0.0

Diff since v3.10.1

The new major release of RxInfer brings several enhancements and updates to improve your experience.

Breaking Changes

• The Transition factor node function has been renamed to DiscreteTransition in line with updates from ExponentialFamily.jl 2.0.
• The MatrixDirichlet is now called DirichletCollection, reflecting changes in ExponentialFamily.jl 2.0.

These changes are part of our ongoing efforts to align with the latest developments in the ExponentialFamily.jl package and to provide a more intuitive and performant implementation of POMDP models.

New Features

• We’ve introduced RxInfer inference sessions, offering a new approach to analyze performance of RxInfer inference routines as well as optional sharing of such sessions with core developer to improve debugging and support.
• A built-in callback hook is now available for performance tracking
• Error hints can now be disabled permanently with Preferences.jl

Improvements

• We’ve improved the wording, naming, and implementation of Hidden Markov and POMDP examples to make them faster and more user-friendly.
• As usual, this release includes various bug fixes and minor improvements to enhance overall stability and performance.
• Dependencies have been updated, for example, in addition to ExponentialFamily.jl the ExponentialFamilyProjection.jl package has also been updated to its latest version, ensuring compatibility and access to new features.

We encourage all users to update to this latest version to take advantage of these improvements. As always, we welcome your feedback and contributions to continue enhancing RxInfer.

Merged pull requests:

• Session storing and sharing (#420) (@bvdmitri)
• Bump projection deps (#424) (@bvdmitri)
• Implement BenchmarkCallback (#425) (@wouterwln)
• Update version to 4.0.0 (#426) (@bvdmitri)
• Split show for benchmarkcallbacks (#430) (@wouterwln)
• use circular buffer instead of vectors for benchmark stats (#431) (@bvdmitri)

Closed issues:

• Suggest MeanField() when constraints are needed (#400)
• Document that Linearization works only with differentiable functions (#422)
• CI is failing with CVI and projection (#423)