Change ReturnDataView attributes to xarray.DataArray?

[dweindl]

I think changing from plain numpy arrays to xarray.DataArray for all data in ReturnDataView would be quite nice:

1. labeled dimensions make it much easier for newcomers to figure out how to slice / plot / ... the data, this should be possible at barely any cost
2. adding coordinates on top would be even more convenient, this would add number of identifier lists, but the cost would probably also be negligible

This should be fully backwards-compatible.

# to happen inside ReturnDataView
import xarray as xr
import numpy as np
sx = xr.DataArray(
    np.random.randn(2, 3, 2),
    dims=("time", "observable", "parameter"),
    coords=dict(time=[0, 10], observable=["obs1", "obs2", "obs3"], parameter=["p1", "p2"])
)

# suggestion:
# sx = rdata.sx
# allows for much more convenient slicing, e.g.
sx.sel(time=10, observable="obs2", parameter="p1")

print(sx)

<xarray.DataArray (time: 2, observable: 3, parameter: 2)>
array([[[ 1.17411809, -1.06761371],
        [-0.20386803, -0.11627882],
        [-1.61609636,  0.15455949]],
       [[ 0.90712023, -0.4521838 ],
        [ 1.40426045, -0.35439355],
        [-0.36240905,  2.25663359]]])
Coordinates:
  * time        (time) int64 0 10
  * observable  (observable) <U4 'obs1' 'obs2' 'obs3'
  * parameter   (parameter) <U2 'p1' 'p2'

[FFroehlich]

Oh sounds nice, didn't know about xarrays!