Attaching axis meta information for user clarity and broadcasting

[NJManganelli]

This is a copy and paste of a discussion started during the IRIS-HEP Analysis Grand Challenge:
https://indico.cern.ch/event/1076231/

In particular, during Lindsey's presentation ('Columnar analysis with coffea', nanoevents.ipynb, https://github.com/CoffeaTeam/coffea/tree/master/binder)

Relevant example code:

# find distance between leading jet and all electrons in each event
dr = events.Jet[:, 0].delta_r(events.Electron)

# find minimum distance
ak.min(dr, axis=1)

Nick Manganelli:

Jim/Nick/Lindsey*, how horrifying would it be to attach an ‘axis label’ to the RecordArrays in some way? Looking at the dr, something that many analyses might use for object-object cross-cleaning, it would be ‘cute’ to be able to say axis=“jet” or axis=“lepton”, or in another call to some ak function, axis=“event”. The simple cases would be nothing more than an enumeration, but the ‘dr’ would of course have a different set of enumerations due to calls to combinatorial methods and order, etc.

Nick Smith:

This sounds like a great discussion to put in https://github.com/CoffeaTeam/coffea/discussions. I agree that in general we need to think about more convenient cross-cleaning and combinatorics syntax

Jim Pivarski:

@nick That would be like the jump from NumPy to xarray, which does put string-valued names on the axes. Not a bad idea, and maybe it could be done in Awkward itself (by attaching names to axes using "parameters").

Nick Smith:

Actually adding a name into the parameters for a given axis would help awkward know automatically how to broadcast combinatorics

Nick Manganelli:

Hmmm, possibly with some parse-ability and inheritance, e.g. ‘event’ vs ‘(intra-event)object’->’jet’->’jetAK4’

*corrections applied