Unexpected dimension values after sum

[thirtysixbananas]

Not sure how important this is. I hit this scenario, but at least I can extract the numbers I need.

How are dimension values named after aggregation?
Here's a nice example:

A = rand(X(1.0:0.2:2.0), Y([:a, :b, :c]))
sum(A; dims=2)

┌ 6×1 DimArray{Float64, 2} ┐
├──────────────────────────┴───────────────────────────────── dims ┐
  ↓ X Sampled{Float64} 1.0:0.2:2.0 ForwardOrdered Regular Points,
  → Y Categorical{Symbol} [:combined] ForwardOrdered
└──────────────────────────────────────────────────────────────────┘
 ↓ →   :combined
 1.0  2.38466
 1.2  1.65471
 1.4  1.2786
 1.6  0.525329
 1.8  1.11983
 2.0  0.838832

The Y now has only one value which is :combined. Ok nice.

Here's an example where it's less nice:

@enumx MyThing Thing1 Thing2
A = rand(X(1.0:0.2:2.0), Y([(MyThing.Thing1, "rah"), (MyThing.Thing2, "lala"), (MyThing.Thing1, "hmm")]))
sum(A; dims=2)

┌ 6×1 DimArray{Float64, 2} ┐
├──────────────────────────┴──────────────────────────────────────────────────────────────────────────── dims ┐
  ↓ X Sampled{Float64} 1.0:0.2:2.0 ForwardOrdered Regular Points,
  → Y Sampled{Tuple{Main.MyThing.T, String}} [(Main.MyThing.Thing2, "lala")] ForwardOrdered Irregular Points
└─────────────────────────────────────────────────────────────────────────────────────────────────────────────┘
 ↓ →   (Thing2, "lala")
 1.0  1.22294
 1.2  1.76796
 1.4  1.25219
 1.6  2.16251
 1.8  0.974786
 2.0  1.41384

The Y now takes an arbitrary value of one of the pre-aggregation values.

[rafaqz]

Yeah, kinda sucks. But what do you want it to do is the question.

Seems its just taking the middle one. If it had length one that would be the expected answer but for everything else its arbitrary. I suspect any other decision might be equally arbitrary.

[thirtysixbananas]

Why not do the same as in the first case? Arbitrarily call it :combined.

If you wanted something more sophisticated, you could mimic what DataFrames does which is to allow to pass colname => func => newcolname which tells it "take func, apply it over colname, and put the result into a new col named newcolname".

[rafaqz]

That changes the type.