v1.0.0

MLJBase v1.0.0

Diff since v0.21.14

• (breaking) Remove the definition of all measures (mae, log_loss, etc), which are instead to be be loaded from StatisticalMeasures.jl. Remove completely measures previously corresponding to loss functions in LossFunctions.jl, which must now be explicitly imported and wrapped. For Julia < 1.9, make StatisticalMeasures a hard dependency and re-export it's names. For Julia 1.9 provide default measures for models using a Pkg extension, including StatisticalMeasures.jl as a weak dependency. See this migration guide for dealing with these breaking changes.

• (breaking) Remove the LossFunctions.jl dependency to close #570.

• Adapt the resampling (evaluate!) code (and a semi-duplication for model Stacks) to the new measure API set out in StatisticalMeasuresBase.jl.

• (breaking) Remove deprecated @pipeline and @from_network macros (#932)

• (breaking) Remove target as alias for transformer in TransformedTargetModel (#933)

Merged pull requests:

• Towards a 1.0.0 release (#908) (@ablaom)
• Replace built-in measures with measures in StatisticalMeasures.jl (#909) (@ablaom)
• Patch to fix StaticNetworkComposite issue (#931) (@ablaom)
• Remove deprecated code supporting @from_network and @pipeline. (#932) (@ablaom)
• Remove target as alias for transformer in TransformedTargetModel (#933) (@ablaom)
• Improve test coverage (#934) (@ablaom)
• For a 1.0.0 release (#935) (@ablaom)
• Add test for display of hyper-parameter ranges and models (#936) (@ablaom)

Closed issues:

• Add measures for multi-targets (#502)
• 1.0 checklist (#638)
• Remove target as alias for transformer in TransformedTarget(transformer=...) (#856)
• Multiclass measures with dictionary output throw errors in cross-validation using evaluate! (#915)
• Problem with StaticNetworkComposite models that have multiple inputs (#930)