Recode does not maintain ordered type

[EssamWisam]

See

using ScientificTypes

julia> x = categorical([1, 2, 3, 4, 5], ordered = true)
5-element CategoricalArray{Int64,1,UInt32}:
 1
 2
 3
 4
 5

julia> scitype(x)
AbstractVector{OrderedFactor{5}} (alias for AbstractArray{OrderedFactor{5}, 1})

julia> y = recode(x, 5 => 9)
5-element CategoricalArray{Int64,1,UInt32}:
 1
 2
 3
 4
 9

julia> scitype(y)
AbstractVector{Multiclass{5}} (alias for AbstractArray{Multiclass{5}, 1})

It's as if ordered = true is no longer the case after recode. Why is that the case?

[nalimilan]

Ordering issues are tricky. Currently whether the result is ordered is determined by the fact that no new levels are added, following the same logic as when concatenating two ordered categorical arrays.

In your example it seems logical to keep the result ordered, as a level is just renamed to another one. But it's not so easy as it seems: if you called recode(x, 5 => 9, 1 => 9) or recode(x, 5 => 2), the result should probably not be ordered. However, recode(x, 5 => 4) is OK, and recode(x, 5 => 3, 4 => 3) is OK too.

At least I guess we could check whether the RHS of pairs are distinct and do not match existing levels. Other cases are probably too complex, users would maybe not expect orderedness to be preserved anyway.