v0.17.6

MLJModels v0.17.6

Diff since v0.17.5

Merged pull requests:

• Update registry 2024 12 19 (#579) (@ablaom)
• For a 0.17.6 release (#580) (@ablaom)

v0.17.5

MLJModels v0.17.5

Diff since v0.17.4

• (new models) Add AffinityPropagation from Clustering.jl to the MLJ model registry
• (new models) Add LaplaceClassifier, LaplaceRegressor from LaplaceRedux.jl

Merged pull requests:

• Add affinity propagation (#574) (@Yuan-Ru-Lin)
• The issue with stdlib versions being fixed to 0.0.0 has been fixed in new versions of Julia (#576) (@DilumAluthge)
• Update registry: Add LaplaceRedux models (#577) (@ablaom)
• For a 0.17.5 release (#578) (@ablaom)

Closed issues:

• allow users to specify float types from OneHotEncoder and ContinuousEncoder (#565)
• poorly maintained code, (#573)

v0.17.4

MLJModels v0.17.4

Diff since v0.17.3

Merged pull requests:

• Update the mode registry (#569) (@ablaom)
• For a 0.17.4 release (#570) (@ablaom)

v0.17.3

MLJModels v0.17.3

Diff since v0.17.2

Merged pull requests:

• Update the Model Registry (#567) (@ablaom)
• For a 0.17.3 release (#568) (@ablaom)

v0.17.2

MLJModels v0.17.2

Diff since v0.17.1

Merged pull requests:

• Remove export FeatureSelector, as moved to FeatureSelection.jl (#563) (@ablaom)
• For a 0.17.2 release (#564) (@ablaom)

Closed issues:

• Add FeatureSelection.jl models to the Model Registry (#557)
• FeatureSelector is still exported (#562)

v0.17.1

MLJModels v0.17.1

Diff since v0.17.0

Merged pull requests:

• For a 0.17.1 release (#560) (@ablaom)

v0.17.0

MLJModels v0.17.0

Diff since v0.16.17

• (mildly breaking) The FeatureSelector appearing in the Model Registry (as inspected using models()) is now provided by the FeatureSelection.jl package (pkg = FeatureSelection) instead of the MLJModels.jl package. This reflects a code migration, not a change in behaviour of the model. The only foreseeable breakage is any code with a call like @load FeatureSelection pkg=MLJModels; which was always redundant anyway, because the model never required explicit loading in MLJ.

• (new models) Add the RecursiveFeatureElimination model wrapper to the model registry (this will be a built-in model in next MLJ release).

• (new feature) Model wrappers, such TunedModel, BalancedModel and IteratedModel, can now be listed using models(...; wrappers=true).

• (internals) The name of a model in the Model Registry now reflects the name a model constructor instead of a model type (which are the same for all current models that aren't composite models aka, model wrappers). Model wrappers (or other models with contractor names different from the type) must overload a new trait MLJModelInterface.constructor to define the constructor for each type. For example, we now have constructor(::DeterministicTunedModel) = TunedModel. For more on this, see #558.

Merged pull requests:

• standardize AbstractFloats (#555) (@tiemvanderdeure)
• Base the Model Registry on constructors, rather than model types; and ancillary changes (#558) (@ablaom)
• For a 0.17 release (#559) (@ablaom)

Closed issues:

• Add PartitionedLS model to the registry (#552)

v0.16.17

MLJModels v0.16.17

Diff since v0.16.16

• (new model) Add partitioned least squares model, PartLS, from PartitionedLS.jl, to the model registry (#553)

Merged pull requests:

• Add PartionedLS model to registry (#553) (@ablaom)
• For a 0.16.17 release (#554) (@ablaom)

v0.16.16

MLJModels v0.16.16

Diff since v0.16.15

Merged pull requests:

• Overload save and restore to address serialization issue for BinaryThresholdPredictor (#550) (@ablaom)
• For a 0.16.16 release (#551) (@ablaom)

Closed issues:

• [compare] CategoricalArrays' cut vs MLJModels' discretizer (#129)
• Expose rng hyperparameter in DecsionTreeRegressor/Classifier (#223)

v0.16.15

MLJModels v0.16.15

Diff since v0.16.14

Merged pull requests:

• Update registry 2024/01/27 (#548) (@ablaom)
• For a 0.16.15 release (#549) (@ablaom)

Closed issues:

• Update model registry to reflect new BetaML model names (#547)