mean errors if distribution keys and the keys of the underlying array are inconsistent

[mzgubic]

I think the error in the following case is the right thing to do.

I wonder whether we should error earlier, when constructing the KeyedDistribution, and never end up with this situation in the first place?

julia> ka = KeyedArray(rand(3), down=["a", "b", "c"])
1-dimensional KeyedArray(NamedDimsArray(...)) with keys:
↓   down ∈ 3-element Vector{String}
And data, 3-element Vector{Float64}:
 ("a")  0.96797267150684
 ("b")  0.812350723246477
 ("c")  0.3433194732682169

julia> kd = KeyedDistribution(MvNormal(ka, [1, 1, 1]), [1, 2, 3])
KeyedMvNormal{MvNormal{Float64, PDMats.PDiagMat{Float64, Vector{Float64}}, KeyedArray{Float64, 1, NamedDimsArray{(:down,), Float64, 1, Vector{Float64}}, Base.RefValue{Vector{String}}}}}(
d: MvNormal{Float64, PDMats.PDiagMat{Float64, Vector{Float64}}, KeyedArray{Float64, 1, NamedDimsArray{(:down,), Float64, 1, Vector{Float64}}, Base.RefValue{Vector{String}}}}(
dim: 3
μ: [0.96797267150684, 0.812350723246477, 0.3433194732682169]
Σ: [1.0 0.0 0.0; 0.0 1.0 0.0; 0.0 0.0 1.0]
)

keys: ([1, 2, 3],)
)


julia> mean(kd)
ERROR: ArgumentError: key vectors must agree; got ["a", "b", "c"] != [1, 2, 3]
Stacktrace:
 [1] unify_one
   @ ~/.julia/packages/AxisKeys/rTIBR/src/broadcast.jl:81 [inlined]
 [2] map
   @ ./tuple.jl:246 [inlined]
 [3] unify_keys
   @ ~/.julia/packages/AxisKeys/rTIBR/src/broadcast.jl:73 [inlined]
 [4] KeyedArray(A::KeyedArray{Float64, 1, NamedDimsArray{(:down,), Float64, 1, Vector{Float64}}, Base.RefValue{Vector{String}}}, k2::Tuple{Vector{Int64}})
   @ AxisKeys ~/.julia/packages/AxisKeys/rTIBR/src/struct.jl:36
 [5] mean(d::KeyedMvNormal{MvNormal{Float64, PDMats.PDiagMat{Float64, Vector{Float64}}, KeyedArray{Float64, 1, NamedDimsArray{(:down,), Float64, 1, Vector{Float64}}, Base.RefValue{Vector{String}}}}})
   @ KeyedDistributions ~/JuliaEnvs/RiskModels/dev/KeyedDistributions/src/KeyedDistributions.jl:184
 [6] top-level scope
   @ REPL[92]:1

[glennmoy]

related to #17

[mzgubic]

This no longer breaks, it uses the keys from the KeyedDistribution.