Index aliasing for xarray dataarrays

[deltamarnix]

Instead of getting data with ncol=5, nrow=10, we want to use j=5, i=10.
We should be able to set aliases on the xarray labels.

[wpbonelli]

Eventually I think we can use custom xarray indices not just for i/j/k indexing but more generally. There is some work in progress to bundle e.g. RangeIndex and RasterIndex with xarray. We can keep an eye on it. The current advice for range indices with integer step is to use an xarray.PandasIndex wrapping a pandas.RangeIndex. That we can begin to prototype immediately, I think.

[wpbonelli]

from the xarray person doing the majority of index-relating development

I think that using a combination of indexes (custom data structures) and accessors (custom API possibly consuming index data) is a great way to extend Xarray.

UXARRAY/uxarray#185 (reply in thread)

[wpbonelli]

After a bit more reading, it seems like the recommended pattern here is to store grid/time in a custom xarray index. (The index is the place for extra state that should be preserved with dataset operations). The index then adds some extra "discretization-awareness" on top of what the data tree already knows about its shape. Specific functionality (e.g. plotting) can be built on the index and exposed with accessors if needed.

[wpbonelli]

I sketched out how we might attach custom indices to xattree components in modflowpy/xattree@fdfa082. Rough and dead simple but I think it might work? Will soon try it here..

[wpbonelli]

think it's working with modflowpy/xattree@14e2531.

grid.data
<xarray.DataTree 'grid'>
Group: /
    Dimensions:  (rows: 3, cols: 3, nodes: 9)
    Coordinates:
      * nodes    (nodes) int64 72B 0 1 2 3 4 5 6 7 8
      * i        (rows) int64 24B 0 1 2
      * j        (cols) int64 24B 0 1 2
    Dimensions without coordinates: rows, cols
    Data variables:
        a        (rows, cols) float64 72B 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
        aa       (nodes) float64 72B 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
    Indexes:
      ┌ i        GridIndex
      └ j
    Attributes:
        rows:     3
        cols:     3