Should there be a MissingVector like SparseVector Type?

[Tokazama]

Sometimes I want to have something that is essentially SparseVector but instead of returning zeros it would return missing. Would this be a welcome addition here?

[quinnj]

We have this already over at SentinelVectors.jl: https://github.com/JuliaData/SentinelArrays.jl/blob/main/src/missingvector.jl. We could move it here if people would prefer.

[Tokazama]

Sorry, I should be more specific. I mean something that is sparsely populated. SentinalArray still has to allocate memory for values that are NaN in the data field right?

julia> x = SentinelVector{Float64}(undef, 1000);

julia> x[[1, 2, 4]] = [1, 2, 3]
3-element Vector{Int64}:
 1
 2
 3

julia> y = sparsevec(zeros(1000));

julia> y[[1, 2, 4]] = [1, 2, 3]
3-element Vector{Int64}:
 1
 2
 3

julia> Base.summarysize(x)
8056

julia> Base.summarysize(y)
152

[nalimilan]

I imagine that could be useful, but I'd rather put this in a dedicated package, and keep only simple convenience functions in Missings.jl. It's a standard convention to have packages called XXXArrays.jl defining the XXXArray type.

[quinnj]

@Tokazama, sorry, I linked specifically to the "missingvector.jl" file in the SentinelArrays.jl package, which actually is the home to 3 different array types: SentinelArray, MissingVector, and ChainedVector. The MissingVector type is the one I was referring to, which is an efficient representation of an array of missing values, and also supports all the expected mutable operations.

[nalimilan]

Let's close this then as this should probably not live in Missings.jl anyway.