issue_comments

I'll close this for the recent discussion about MultiIndex

@benbovy Sorry for my late reply.

I think I like your proposal, which bundles multiple concepts in xarray such as MultiIndex and multi-dimensional coordinates into one, which may result in simpler API. But actually I don't yet fully imagine how your proposal works with multi-dimensional coordinates. (maybe because I am not accustomed with multi-dimensional coordinates very well.)

Currently, 'rasm' example is like python In [1]: import xarray as xr In [2]: xr.tutorial.load_dataset('rasm', decode_times=False) Out[2]: <xarray.Dataset> Dimensions: (time: 36, x: 275, y: 205) Coordinates: * time (time) float64 7.226e+05 7.226e+05 7.227e+05 7.227e+05 ... xc (y, x) float64 189.2 189.4 189.6 189.7 189.9 190.1 190.2 190.4 ... yc (y, x) float64 16.53 16.78 17.02 17.27 17.51 17.76 18.0 18.25 ... Dimensions without coordinates: x, y Data variables: Tair (time, y, x) float64 nan nan nan nan nan nan nan nan nan nan ... Attributes: ...

Does your proposal (automatically) change this like python <xarray.Dataset> Dimensions: (time: 36, xy: 56375) Coordinates: * time (time) float64 7.226e+05 7.226e+05 7.227e+05 7.227e+05 ... xc (xy) float64 189.2 189.0 188.7 188.5 188.2 187.9 187.7 187.4 ... yc (xy) float64 16.53 16.69 16.85 17.01 17.17 17.32 17.48 17.63 ... * xy (xy) SuperIndex - x (xy) int64 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ... - y (xy) int64 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ... Data variables: Tair (time, xy) float64 nan nan nan nan nan nan nan nan nan nan nan ... Attributes: ... ?

@shoyer I personally think 2 is more intuitive for users, because it might be difficult to distinguish python <xarray.Dataset> Dimensions: (yx: 6) Coordinates: y (yx) object 'a' 'a' 'a' 'b' 'b' 'b' Data variables: foo (yx) int64 1 2 3 4 5 6 (which may be generated by indexing from x in your example) from python <xarray.Dataset> Dimensions: (y: 6) Coordinates: * y (y) object 'a' 'a' 'a' 'b' 'b' 'b' Data variables: foo (y) int64 1 2 3 4 5 6 What is the possible confusion if we adopt 2?

@shoyer Thanks for the comment.

It occurs to me that if we had full support for indexing on coordinate levels, we might not need a notion of a "MultiIndex" in the public API at all.

Actually I am not yet fully comfortable with my implementation, and I like your idea as this might be much cleaner and simpler than mine.

If my understanding is correct, does it mean that we will support ds.sel(x='a'), ds.isel(x=[0, 1]) and ds.mean(dim='x') with your example data? Will it raise an Error if Coordinate is more than 1 dimensional? How about ds.sel(x='a', y=[1, 2])?