Bring CANYON-MED to argopy

[gmaze]

Contemplating to bring CANYON-B to argopy

There is a nice Python code written by @MarineFou and @fricour see fork here: https://github.com/euroargodev/CANYON-MED

Few comments:

• no NN library is used, it's all encoded brut force !
• I don't see any big difficulties in bringing this to argopy:
  • a new class/function to make predictions could be added as a argopy/related sub-module
  • text files with NN weights can easily be added to the argopy/static/assets folder
  • a new entry to the argo accessor could make this easy to use, the API could look like this:

from argopy import DataFetcher

ds = DataFetcher(ds='bgc', params='DOXY').float(<WMO>).load().data

ds = ds.argo.canyon_med.predict()  # Shall predict all variables: NO3, PO4, SiOH4, PAT, CT, pHT
ds = ds.argo.canyon_med.predict('NO3')
ds = ds.argo.canyon_med.predict(['NO3', 'PO4'])

After this, the xarray dataset would have the predicted variables added !

This could be a first step toward bringing CANYON-B to argopy as well.

poke: @Sauzede @catsch @HCBScienceProducts

[gmaze]

poke: @laurcopp !

[HCBScienceProducts]

I think this would be a nice gimmick, but I wouldn't see it as a priority. First, I'd think, the BGC data access and interactions need to be(come) robust. E.g. I'd prioritize BGC-traj data access (-> think of PPOX_DOXY in air data, or parking drift phase BBP time series) over this.

[gmaze]

I think this would be a nice gimmick, but I wouldn't see it as a priority.

surely not a priority, but a very good use-case to proof-test how we can bring BGC regression methods to argopy. CANYON-B on target.

I'd prioritize BGC-traj data access

This was raised recently here (#332 ) and we talked about this in LOV with @Sauzede I would need to first understand how these data are being used

[gmaze]

@HCBScienceProducts you can make the case for traj files in here: https://github.com/euroargodev/argopy/discussions/214#discussioncomment-2659566

[laurcopp]

Good job @gmaze ! Thanks a lot