v1.8.1

MLJBase v1.8.1

Diff since v1.8.0

Merged pull requests:

• Fix deprecated Vararg expression (#1000) (@devmotion)
• For a 1.8.1 release (#1001) (@ablaom)

v1.8.0

MLJBase v1.8.0

Diff since v1.7.0

• Drop support for Julia versions < 1.10.

Merged pull requests:

• Allow selectcols to have tuples as "indices" argument (#992) (@ablaom)
• Update some compats and dump Julia 1.6 support in favor of new LTS release (#997) (@ablaom)
• For a 1.7.1 (#998) (@ablaom)
• Bump 0.8.0 (#999) (@ablaom)

Closed issues:

• Document range scales better (#946)
• selectcols should handle tuple input (#991)
• Re-enable CompatHelper workflow (#993)

v1.7.0

MLJBase v1.7.0

Diff since v1.6.0

Merged pull requests:

• Relax restrictions on model type in resampling (evaluate!) (#985) (@ablaom)
• For a 1.7 release (#988) (@ablaom)

v1.6.0

MLJBase v1.6.0

Diff since v1.5.0

• (enhancment) Arrange that pipelines support transformers that need a target variable for training (#984)

Merged pull requests:

• Docstring patches (#983) (@abhro)
• Add support in pipelines for Unsupervised models for which target_in_fit is true (#984) (@ablaom)
• Fix default_logger docstring (#986) (@ablaom)
• For a 1.6 release (#987) (@ablaom)

v1.5.0

MLJBase v1.5.0

Diff since v1.4.0

• (enhancement) Add facility to specify a logger globally, with a setter default_logger(logger), and a getter default_logger() (#979)

Merged pull requests:

• Add a default_logger (#979) (@ablaom)
• For a 1.5.0 release (#982) (@ablaom)

v1.4.0

MLJBase v1.4.0

Diff since v1.3.0

Merged pull requests:

• Have wrappers overload constructor trait (#980) (@ablaom)
• For a 1.4 release (#981) (@ablaom)

v1.3.0

MLJBase v1.3.0

Diff since v1.2.1

• (Performance enhancement) Remove type instability for predict(mach::Machine, ...) in the easy and typical case that mach does not wrap a Symbol model (#969)

• (New feature) Give evaluate and evaluate! the option compact=true, to return a CompactPerformanceEvaluation object with minimal memory footprint (#973)

• (New feature) Add an InSample() resampling strategy that trains and tests on the same data (whatever is specified by rows, or all supplied data) (#975)

• (Display improvement) Split the table displayed as part of an PerformanceEvaluation object over two tables, if needed, to deal with overly wide tables (#973)

Merged pull requests:

• Add prompt to docstring REPL example (#968) (@abhro)
• Address some predict/transform type instabilities (#969) (@ablaom)
• Update docstring examples and code (#970) (@abhro)
• Make test of iterator(...) more robust (#972) (@ablaom)
• Add CompactPerformanceEvaluation objects and the option in evaluate! to construct them (#973) (@ablaom)
• Add InSample resampling strategy (#975) (@ablaom)
• For a 1.3 release (#977) (@ablaom)

Closed issues:

• Add InSample resampling strategy (#967)
• Tests failing on dev (#971)
• selectrows struggling with views of a table (#974)

v1.2.1

MLJBase v1.2.1

Diff since v1.2.0

Merged pull requests:

• Fix problem with training losses for pipelines (#965) (@ablaom)
• For a 1.2.1 release (#966) (@ablaom)

v1.2.0

MLJBase v1.2.0

Diff since v1.1.2

• (enhancement) Expose feature_importances in pipelines with a supporting supervised component, and in TransformedTargetModels with supporting atomic model (#963)

Merged pull requests:

• Fix some error message diagnostics (#962) (@ablaom)
• Allow Pipeline and TransformedTargetModel to support feature importances (#963) (@ablaom)
• For a 1.2.0 release (#964) (@ablaom)

v1.1.2

MLJBase v1.1.2

Diff since v1.1.1

Merged pull requests:

• Reduce sigdigits in parameter range display #948 (#958) (@adarshpalaskar1)
• Fix problem with serialization of nested models when component model overload save/restore (#960) (@ablaom)
• For a 1.1.2 release (#961) (@ablaom)

Closed issues:

• Serialized Composite Model Fails with XGBoost (#927)
• Reduce sigdigits in parameter range display (#948)