non-dimension coordinates

[Beforerr]

In Python Xarray, we can have non-dimension coordinates that contain coordinate data, but are not a dimension coordinate (see https://docs.xarray.dev/en/stable/user-guide/data-structures.html#coordinates) .

I'm just wondering if there is any similar support for this in the ecosystem. Actually this idea is kinda of useful for working with multidimensional coordinates.

[rafaqz]

Have been thinking about something like this in Rasters as CF standards allow having multiple lookup coordinates for a single dimension. Is that what you mean?

My current idea is to allow dimensions/lookup nesting, so a dimension can have alternative names and associated lookups. Then e.g. X=At(100) may be in metres but you can use Lon=At(100) and that's degrees. The Lon dimensions would be a field of the X dimensions lookup array, and used as a fallback when Lon is not found in the normal top level Tuple of dims.

[Beforerr]

I am more thinking about multidimensional coordinates. You can check the draft PR here #1086 . I am not sure it's a good idea to put the type in this package or just split it into a new package.

[asinghvi17]

Ah so you want to mix the dimensions of the dimensions and coordinates? So X and Y can be 1D regular but there is also a 2D lon and 2D lat associated.

[Beforerr]

Right. Something Like that, I think it would be beneficial to differentiate between coordinate and dimension.

[rafaqz]

Try to avoid starting a new package

This functionality is planned for our overhaul of CF standards in Rasters.jl. DD lookups will be able to represent any CF object which includes single and multidimensional co-ordinate variables. Without changing the dims interface ;)

But it will be a few months.

[Beforerr]

Thanks for clarifying! That sounds great. A few months is no problem—I can work around it in the meantime. Looking forward to the overhaul and the DD lookups supporting full CF objects.

[rafaqz]

Great! can I ping you for reviews?

[Beforerr]

Absolutely, one is glad to be of service

[JamesWrigley]

I would also like this 👀 Happy to test it out when ready.

[JamesWrigley]

I/Claude implemented non-dimension coordinates in a branch: JamesWrigley@d080267

Which allows you to do e.g.:

julia> x = DimArray(randn(2, 3), (X(1:2; lookups=(foo=[10, 11],)), Y))
┌ 2×3 DimArray{Float64, 2} ┐
├──────────────────────────┴─────────────────────── dims ┐
  ↓ X Sampled{Int64} 1:2 ForwardOrdered Regular Points,
  → Y
├────────────────────────────────────────────────────────┴───────────────────────────────────────────────── coordinates ┐
  foo (X) Sampled{Int64} [10, 11] ForwardOrdered Irregular Points
└───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘
 1   0.588798  0.751046  0.348582
 2  -0.51124   0.402246  0.897739

julia> x[foo=At(11)]
┌ 3-element DimArray{Float64, 1} ┐
├────────────────────────── dims ┤
  ↓ Y
└────────────────────────────────┘
 -0.51124
  0.402246
  0.897739

julia> x[X=At(2)]
┌ 3-element DimArray{Float64, 1} ┐
├────────────────────────── dims ┤
  ↓ Y
└────────────────────────────────┘
 -0.51124
  0.402246
  0.897739

@rafaqz, is this along the lines of what you had in mind, and if so is it worth making a PR? The downside of this approach is that it only supports 1D lookups, not sure what you have planned for N-dimensional lookups.

[JamesWrigley]

(bump)

[rafaqz]

Yeah, I implemented this for Rasters CF standards too. Problem is having any time to finish things

[JamesWrigley]

Big mood 😅 Is it worth me making a PR out of the branch? Or I could have a go at finishing off your branch if it's pushed somewhere.

[JamesWrigley]

(gentle bump, happy to help out if I can)