Breaking: Materialize DimArray or DimStack From a Table

[JoshuaBillson]

Description:

resolves #335

This PR aims to let users construct either a DimStack or DimArray from a table with one or more coordinate columns.

Unlike the existing contructor, rows may be out of order or even missing altogether.

Performance:

The algorithm is O(n), requiring two forward passes for each dimension to determine the correct order of rows.

1000x1000: 0.005 Seconds

2000x2000: 0.025 Seconds

4000x4000: 0.108 Seconds

8000x8000: 0.376 Seconds

Example:

julia> r = DimArray(rand(UInt8, 1000, 1000), (X, Y));

julia> t = r |> DataFrame |> Random.shuffle!;

julia> restored = DimArray(t, dims(r));

julia> all(r .== restored)
true

Next Steps:

1. Finalize method signatures.
2. Decide what to do with missing rows. Currently, users may choose a missing value to fill in (missing by default).
3. Add support for a :geometry column.
4. Write test cases.
5. Update docs.

[JoshuaBillson]

There's still a few questions that need resolving:

1. What do we do with missing rows? This can be handled explicitly in Rasters.jl, but DimArrays have no concept of missing values. We could let users choose a missing value to write, or we could encode it explicitly with missing. We could also disallow the existence of missing rows, but I don't think this is an ideal solution.
2. How should we handle a :geometry column? I know that several packages in the Julia ecosystem are using this convention, but I don't have much experience with them. Can we assume :geometry will contain tuples of coordinates, or could they also be some sort of geometry like Meshes.Point?

[asinghvi17]

I think it's best to test that the geometry column's elements have GI.PointTrait - then you can always extract e.g. GI.x(point) and the same for y, z, and m. You can also interrogate the dimension via GI.is3d, GI.ismeasured.

Going forward to get the geometry column it's probably best to have first(GI.geometrycolumns(table)) - the fallback implementation will give you (:geometry,), but to handle tables with other geometry columns which may be indicated by e.g. metadata, it's better to use this.

[rafaqz]

DimensionalData.jl does not depend on GeoInterface

[JoshuaBillson]

DimensionalData.jl does not depend on GeoInterface

Being able to interact with geometries in a generic fashion would help us interop with other packages. GeoInterface's only dependency is Extents, which we already depend on. After importing DimensionalData, GeoInterface loads in 0.008 seconds.

I also see that Rasters already depends on GeoInterface. We could simply ignore the :geometry column in DimensionalData, then implement the additional functionality in Rasters. However, I think the problem is general enough to be implemented here.

[rafaqz]

It's not about deps or timing, it's about clean feature scope.

The promise here is that Rasters (and YAX) have all the geo deps and features, so non-geo people don't have to worry about them. Some of the biggest contributors here use DD for unrelated fields.

I would ignore point columns here entirely and instead write the code so it's easy for Rasters to handle them. Taking the underscores off a few functions and documenting them as an real interface will help that.

[JoshuaBillson]

I've updated the docstrings for both the DimArray and DimStack constructors. By default, materializers will now use the Contains selector for irregular and non-numeric coordinates, with the option to specify an alternative when desired. We're exporting the restore_array and coords_to_index methods to be used by downstream packages like Rasters.jl. I've also written several test cases to show that we can handle tables with out of order and missing rows. If everything looks good, I think we can go ahead and merge this PR.

[rafaqz]

This is good, but a few minor changes are needed for it to be correct. Constructed selectors should be allowed so At can have atol. We can just construct selector types at the outer level (they're just filled with nothing.

Then there is a bit more dispatch needed to make the fast paths correct for At/Near/Contains with their standard behaviour.

[codecov]

Codecov Report

❌ Patch coverage is 84.76821% with 23 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.18%. Comparing base (886a2b2) to head (55f8017).
⚠️ Report is 1 commits behind head on breaking.

Files with missing lines	Patch %	Lines
src/table_ops.jl	81.44%	18 Missing ⚠️
src/stack/stack.jl	84.61%	4 Missing ⚠️
src/tree/tree.jl	75.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           breaking     #739      +/-   ##
============================================
- Coverage     84.62%   84.18%   -0.45%     
============================================
  Files            53       54       +1     
  Lines          5307     5456     +149     
============================================
+ Hits           4491     4593     +102     
- Misses          816      863      +47     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[tiemvanderdeure]

Okay docs fails because of this:

 Error: 5 docstrings not included in the manual:

    DimensionalData.restore_array :: Tuple{AbstractVector, AbstractVector, Tuple, Any}
    DimensionalData.coords_to_indices :: Tuple{Any, Tuple}
    DimensionalData.guess_dims :: Tuple{Any}
    DimensionalData.data_col_names :: Tuple{Any, Tuple}
    DimensionalData.get_column :: Tuple{Any, Type{<:Dimension}}

But anyway I think all of these functions should be internal. Should I just remove the docstrings - or make them into comments or something?

(Never seen this before so I'm not 100% about what this is testing)

[rafaqz]

Yeah we force docstrings to be in the docs. I think these should just be comments not docstrings

[tiemvanderdeure]

All right let's see if this passes now. I think before merging we should add some documentation, though.

Another thing is that this will only work with exactly DimArray now, not other abstract dimarrays or abstractdimstacks. We probably want this functionality with rasters and maybe yaxarrays as well, without creating dispatch mayhem.

I've tried to solve this now by adding a type argument to dimarray_from_table, so you could call dimarray_from_table(Raster, table, dims; kw...) and the only lines you would have to copy are these:

https://github.com/JoshuaBillson/DimensionalData.jl/blob/aec86a9600d6ceb3af0aaa7dadc0f051c638a87d/src/array/array.jl#L519-L544

[rafaqz]

Yes perfect that's why I removed the underscores, those a_from_b methods can become dev api

[rafaqz]

@tiemvanderdeure this looks good to merge ?

[tiemvanderdeure]

Yeah I think it is! I don't have more, if you've reviewed it and found nothing then do merge :)

[tiemvanderdeure]

I think this got merged into main and released. Was that on purpose? It is very slightly breaking.

[rafaqz]

It got merged into breaking! which I do want to merge now but it will get a breaking bump

[tiemvanderdeure]

Ah I got confused because it is listed in the tag https://github.com/rafaqz/DimensionalData.jl/releases/tag/v0.29.21

Didn't know that it also listed PRs into other branches than main

[rafaqz]

Oh I didn't know that either.. thats actually bad