v0.3.0

MarkovKernels v0.3.0

Diff since v0.2.1

Features

• Support for scalars as input/output of affine maps
• Support for Univariate Dirac/Normal distributions
• Costum types can now opt into being a valid representation of a PSD matrix by implementing psdcheck(::MyType) = IsPSD() and the accompanying interface
• Support for Ǹumber``` and ÙniformScaling```as PSD parametrizations.
• New type for representing Categorical distributions
• New type for representing Stochastic matrices
• New type for representing likelihood functions over categories
• New type for canonical parametrization of log-quadratic likelihood functions
• New function htransform for implementing backward likelihood recursions

Breaking

• sample_type instead of typeof_sample
• sample_eltype instead of eltype_sample

docs

• Updated to showcase both forward and backward recursions for posterior inference.

Testing

• Now testing against current Julia release
• Aqua tests
• JET tests

Merged pull requests:

• Package extension for RecursiveArrayTools (#94) (@nathanaelbosch)
• Add Aqua.jl and JET.jl to the testing (#95) (@nathanaelbosch)
• Update CI.yml to test the current Julia release (#96) (@nathanaelbosch)
• fixes failign tests for complex normal distributions (#98) (@filtron)
• fix ambiguites and unbounded args (#100) (@filtron)
• Univariate Normals (#103) (@filtron)
• tidier cov parameters (#105) (@filtron)
• scalar2scalar, vector2scalar (#109) (@filtron)
• remove vecofvec diracs, add univariate dirac, tests... (#110) (@filtron)
• PSDTrait (#112) (@filtron)
• Categorical (#113) (@filtron)
• stochastic matrix and tests (#114) (@filtron)
• diagonal (#115) (@filtron)
• sample type as typeparam in AbstractDistribution (#116) (@filtron)
• logquadratic, uniformscaling, htransform for gaussmarkov (#118) (@filtron)
• Update docs (#119) (@filtron)
• remove outdated tutorial (#120) (@filtron)
• htransform cleanup (#121) (@filtron)
• version bump (#122) (@filtron)

Closed issues:

• condition vs making MarkovKernels callable (#79)
• Fix ambiguities, unbounded arguments, and piracies raised by Aqua.jl (#99)
• How to use PSDMatrices.PSDMatrix as a covariance parameter (#101)
• Scalar-valued MarkovKernels.Normal (#102)
• univariate Diracs (#104)
• Should AffineNormalKernel be its own type rather than a type parameter restriction? (#106)
• implement vector to scalar and scalar to scalar affine maps (#107)
• Allow more fine-grained control over choice of algorithms for stein and schur_reduce (#108)