v0.22.0

MLJ v0.22.0

Diff since v0.21.0

• (mildly breaking) The behaviour of levels and unique on CategoricalArrays has changed. (Such arrays are created in MLJ, for example, by coerce(array, Multiclass) or coerce(array, OrderedFactor). ) The levels and unique methods now return a CategoricalVector whereas previously they returned a vector of "raw" values. So, running levels(array) previously is equivalent to now running CategoricalArrays.unwrap.(levels(array)). The new behaviour is the result of breaking changes in CategoricalArrays.jl, on which MLJ.jl depends (#1172)

Merged pull requests:

• Bump actions/checkout from 4 to 5 (#1175) (@dependabot[bot])
• Regenerate documentation. No new release. (#1182) (@ablaom)
• Regenerate dev docs (#1183) (@ablaom)
• Restore some integration tests for BetaML. (#1184) (@ablaom)
• Bump CategoricalArrays, etc and major testing change 😱 (#1185) (@ablaom)
• For a 0.22 release (#1186) (@ablaom)

Closed issues:

• Reinstate CatBoost integraton test (#1092)
• Re-instate integration tests for scikit-learn models (#1119)
• Reinistate integration tests for SymbolicRegression? (#1152)
• Reinstate Outlier detection model s (#1153)
• Error when loading ContinuousEncoder (#1181)