[Tracking] Updating dependencies to CategoricalArrays 1.0

[ablaom]

CategoricalArrays 1.0 breaks a few things in MLJ. See here for some details. Here's the rough sequence for rolling these out:

• Update CategoricalDistributions.jl and tag as breaking, v0.2.0
• Update ScientificTypes.jl
• Update StatisticalMeasures.jl
• Update StatisticalMeasuresBase.jl
• Update MLJBase.jl Bump the compat requirements for CategoricalArrays and CategoricalDistributions MLJBase.jl#1028
• MLJ Bump CategoricalArrays, etc and major testing change 😱 #1185

I will do these.

Then, in no particular order, we should review the compat requirements for the following core packages. I do not expect breakages but in a few cases we might start getting dep warnings which would be good to address these early.

• MLJTuning.jl Does not appear to need updating (tested with a draft MLJBase update)
• MLJIteration.jl ditto
• MLJEnsembles.jl Bump compats: CategoricalArrays="1.0", CategoricalDistributions="0.2" MLJEnsembles.jl#42
• MLJFlux.jl (needed for MLJTransforms doc generation) Bump compat for CategoricalArrays to 1.0 FluxML/MLJFlux.jl#311 passes tests, docs failing
• MLJTransforms (needed for testing in MLJModels) Bump [compat] entry: CategoricalArrays="1.0" MLJTransforms.jl#61 local tests: ✔️
• MLJMultivariateStatistics (needed for MLJModels testing) Bump compats for StatsBase and CategoricalDistributions MLJMultivariateStatsInterface.jl#63 local tests: ✔️
• MLJDecisionTreeInterface.jl (needed for tests in FeatureSelection.jl and MLJModels.jl) Bump CategoricalArrays compat to 1.0 MLJDecisionTreeInterface.jl#60 local tests: ✔️
• MLJModels.jl (needed for tests in MLJBalancing) Bump compat CategoricalArrays="1" MLJModels.jl#597 local tests: ✔️
• Imbalance.jl (needed for testing MLJBalancing) Extend the compats for CategoricalArrays and CategoricalDistributions Imbalance.jl#105
• MLJBalancing.jl (doesn't appear to need code changes, but depends on MLJBase)
• FeatureSelection.jl Extend compat for StatisticalMeasures to include v0.3 FeatureSelection.jl#30 local tests: ✔️
• ARFFFiles.jl (needed by OpenML, on which MLJ depends) Bump CategoricalArrays = "1" JuliaData/ARFFFiles.jl#26 local tests: ✔️
• Maxnet.jl Bump or extend compat for CategoricalArrays to include 1.0 tiemvanderdeure/Maxnet.jl#28

Here are some other packages I've checked and don't appear to need attention:

• MLJScikitLearnInterface

Additional action required after MLJ itself is updated:

• Regenerate docs after MLJ update MLJTransforms.jl#62 (MLJTransforms)
• Regenerate documentation after MLJ update FluxML/MLJFlux.jl#312 (MLJFlux)

cc @DilumAluthge @devmotion

[ablaom]

And then, some packages providing implementations of the MLJ API will likely need updating to CategoricalArrays 1.0 and/or CategoricalDistributions 0.2:

• BetaML
• CatBoost (requested)
• EvoTrees
• LaplaceRedux (requested)
• MLJLIBSVMInterface
• OneRule (requested)([issue posted]
• OutlierDetectionInterface (requested)
• SIRUS (requested)

The steps for updating are:

1. update the compats
2. change in /src and /test any reference to CategoricalArrays.levels(x) to unwrap.(CategoricalArrays.levels(x)) if necessary (or if you are not sure). Look out for postfix applications like x |> levels. Possibly one has to do something similar for any references to CategoricalArrays.unique but I haven't encountered a problem with that one yet.
3. change any reference to MLJBase.classes or MLJModelInterface.classes to CategoricaArrays.levels (or CategoricalDistributions.levels)
4. check tests

[juliohm]

Bumped into this issue today in one of our test suites. MLJ.jl is holding CategoricalArrays.jl < v1