v0.5.4

MLJMultivariateStatsInterface v0.5.4

Diff since v0.5.3

• Bump the compatibility requirements, CategoricalDistributions="1"

Merged pull requests:

• Extend StatsBase="0.34" (#61) (@ablaom)
• Bump compats for StatsBase and CategoricalDistributions (#63) (@ablaom)
• For a 0.6 release (#64) (@ablaom)
• Ammend version number to make it non-breaking (#65) (@ablaom)

v0.5.3

MLJMultivariateStatsInterface v0.5.3

Diff since v0.5.2

Merged pull requests:

• CompatHelper: bump compat for StatsBase to 0.34, (keep existing compat) (#59) (@github-actions[bot])

v0.5.2

MLJMultivariateStatsInterface v0.5.2

Diff since v0.5.1

Closed issues:

• Documentation tracking issue (#38)
• Doc strings have "unknown" as source package (#56)

Merged pull requests:

• Fix pkg metadata (#57) (@ablaom)

v0.5.1

MLJMultivariateStatsInterface v0.5.1

Diff since v0.5.0

• add loadings to report for PCA model.

Closed issues:

• Replace transpose and permutedims with adjoints to avoid copying data every time (#27)
• How to access loadings() or reconstruct()? (#53)

Merged pull requests:

• report loadings for PCA models (#54) (@OkonSamuel)
• For a 0.5.1 release (#55) (@OkonSamuel)

v0.5.0

MLJMultivariateStatsInterface v0.5.0

Diff since v0.4.0

Bumps the MultivariateStats requirement to 0.10, which forces some breaking changes (#51; @OkonSamuel):

• (breaking) The output of fitted_params and report for LDA models has changed. See the doc-strings for details.
• (breaking) Interpreting some output (now and previously) needs an ordering of target classes, and this ordering may no longer agree with levels(y). However the class label order adopted is now explicitly output by fitted_params.
• (breaking) In BayesianLDA and BayesianSubspaceLDA target class priors may no longer be specified as vectors. They must either be dictionaries or UnivariateFinite objects from CategoricalDistributions.jl.

Closed issues:

• Allow support in LDA, BayesianLDA, SubspaceLDA, and BayesianSubspaceLDA for unique target class in training. (#41)

Merged pull requests:

• Allow for a lone target class in default LDA models and cousins (#48) (@ablaom)
• upgrade to multivariatestats v0.10 (#51) (@OkonSamuel)
• For a 0.5.0 release (#52) (@ablaom)

v0.4.0

MLJMultivariateStatsInterface v0.4.0

Diff since v0.3.2

• (breaking) Rename the class_means key in reports to projected_class_means, for consistency with key name in fitted_params(model, ...). Affects LDA and its relatives (#47).
• (breaking) Rename out_dim in hyper-parameters and reports with outdim for consistency with use in other models. Effects ICA, and LDA and relatives of LDA (#47).
• (breaking) Rename the pratio hyperparameter in PCA to variance_ratio as more informative (#47)
• (breaking) Rename nc in reports to nclasses as more informative. Affects LDA and its relatives (#47)
• (breaking) Rename hyper-parameter k in ICA to outdim for consistency with other models and report (#47)
• (docs) Add detailed MLJ-compliant document strings to all models (#39) @josephsdavid, @ablaom

Closed issues:

• Inconsistencies in field names (#46)

Merged pull requests:

• Make docstrings (#39) (@josephsdavid)
• Fix inconsitencies in names of hyperparameters, report keys, and fitted_params() keys. (#47) (@ablaom)
• For a 0.4 release (#50) (@ablaom)

v0.3.2

MLJMultivariateStatsInterface v0.3.2

Diff since v0.3.1

Closed issues:

• Issue to trigger new releases (#4)
• LDA and cousins are not tracking all levels of the target. (#40)
• fitted_params for ICA throws an error (#42)

Merged pull requests:

• Re-instate projection for ICA (#43) (@ablaom)
• For a 0.3.2 release (#44) (@ablaom)

v0.3.1

MLJMultivariateStatsInterface v0.3.1

Diff since v0.3.0

Merged pull requests:

• add test for MS.Gaus (#36) (@testercwt)
• For a 0.3.1 release (#37) (@ablaom)

v0.3.0

MLJMultivariateStatsInterface v0.3.0

Diff since v0.2.2

• Bump Julia requirement to 1.6
• (mildly breaking) Change the default value of inverse in kernelPCA to true (#28)
• Internally use adjoints instead of transposes on matrixified table input, except in LDA models where MultivariateStats.jl lacks adjoint support (partially addresses #27)
• Add support for MultivariateStats 0.9 (#29) @testercwt

Merged pull requests:

• upgrade to MS 0.9 (#29) (@testercwt)
• Use adjoints not transposes (#32) (@ablaom)
• Change default hyperparameter inverse in KernelPCA (#33) (@ablaom)
• For a 0.3 release (#35) (@ablaom)

v0.2.2

MLJMultivariateStatsInterface v0.2.2

Diff since v0.2.1

• extend [compat] for MLJModelInterface to include 1.0

Closed issues:

• set default branch as dev (#24)

Merged pull requests:

• merge master into dev (#23) (@OkonSamuel)
• CompatHelper: bump compat for "MLJModelInterface" to "1.0" (#25) (@github-actions[bot])
• For a 0.2.2 release (#26) (@ablaom)