New API design for indexing Argo profiles from xarray accessor

[gmaze]

Given a standard xarray dataset loaded with argopy as a collection of samples, i.e. with a single dimension N_POINTS.

Would it be nice to have an easy indexing and selecting API for Argo profiles using our argo accessor ?

This may look like this:

ds.argo.sel(PLATFORM_NUMBER=1902605, CYCLE_NUMBER=4)
ds.argo.sel(wmo=1902605, cyc=4)

ds.argo.isel(PLATFORM_NUMBER=2, CYCLE_NUMBER=1)

ds.argo.isel(n_prof=12)
ds.argo.isel(wmo=2, cyc=1)

ds.argo[12]  # n-eme profile
ds.argo[3, 123]  # i-eme float, j-eme profile

This should require an efficient way to naviguate the ragged array structure of Argo samples.
May be Akward Array could be used here.

[gmaze]

In fact, we may even implement this for the DataFetcher, that would make it easily lazy !

ArgoSet = argopy.DataFetcher(ds='bgc', src='erddap', mode='expert').region([-55, -47, 55, 57, 0, 10])

ArgoSet.sel(PLATFORM_NUMBER=1902605, CYCLE_NUMBER=4)
ArgoSet.sel(wmo=1902605, cyc=4)

ArgoSet.isel(PLATFORM_NUMBER=2, CYCLE_NUMBER=1)

ArgoSet.isel(n_prof=12)
ArgoSet.isel(wmo=2, cyc=1)

ArgoSet[12]  # n-th profile
ArgoSet[3, 123]  # i-th float, j-th profile

Although this raises the question of what is the n-th profile for a data selection with an increasing number of profiles over time ...

[gmaze]

Seems like xarray provides custom indexing methods see here:

pydata/xarray#7041

https://docs.google.com/presentation/d/1YwZA80nt3TfG-wLPNAFPod7GeZxu1z_kW1HnzctomLc/edit#slide=id.g314077217e3_1_47