Strip type information for CategoricalValues

[andreasnoack]

The type information is very noisy and can easily trigger limited output

julia> display(freqtable(categorical(["A", "B", "A", "B"]), categorical(["ONE", "ONE", "TWO", "TWO"])))
2×2 Named Matrix{Int64}
                         Dim1 ╲ Dim2 │   …
─────────────────────────────────────┼──────
CategoricalValue{String, UInt32} "A" │   …
CategoricalValue{String, UInt32} "B" │   …

[andreasnoack]

Hm. I guess this might actually be a NamedArrays or CategoricalArrays issue

[andreasnoack]

Well. I guess this could potentially happen here since these dicts could simply strip the type information

julia> tmp.dicts
(OrderedCollections.OrderedDict{CategoricalValue{String, UInt32}, Int64}("A" => 1, "B" => 2), OrderedCollections.OrderedDict{CategoricalValue{String, UInt32}, Int64}("ONE" => 1, "TWO" => 2))

so I'll reopen to discuss the options.

[nalimilan]

Yeah this should probably be fixed in CategoricalArrays. The only thing we could change here is to unwrap CategoricalValue variables, i.e. in the example convert them to String. IIRC we used to do that, but it's a bit weird to special-case CategoricalValue.

See https://github.com/JuliaData/CategoricalArrays.jl/pull/371.