How to do Flexible Binning

[luketpickering]

DUNE analysis needs 2+ dimensional binning

ND280 analysis needs 2D arbitrary rectangular binning (TH2Poly-like)

Drop other constraints below:

I've implemented hyper-rectangular NDim binning in a minimally intrusive way (MaCh3 Core just sees a 1D axis in 'global bin number'). However, the current implementation works as a product of 1D axes and so cannot solve the ND280 problem. I have implementations for an arbitrary hyper-rectangular version, but didn't commit to implementing it yet as it wasn't needed for our DUNE use case (but might be in the future).

Apparently @KSkwarczynski is also working on a separate new binning technique, Add details/discussion here and lets discuss new binnings!

[EdAtkin]

Personally I would b happy for us to drop the TH2Poly support and move to hyper-rectangular binning. Whilst the TH2Poly has been interesting to study for T2K ND280 I think the overhead it has added, in terms of supporting what is quite an annoying ROOT class, is not worth it.

I will also say that this makes the bin migration much easier to support in any dimension (rather than just the x axis at the moment).

It also should mean that we can loop over global bin number rather than X-bins and Y-bins because it's trivial to calculate the x, y, z (etc.) bin number if we have each 1D axis defined somewhere. This should lead to a small speed increase in the event reweighting.

[KSkwarczynski]

Apparently @KSkwarczynski is also working on a separate new binning technique, Add details/discussion here and lets discuss new binnings!

I am not working on this right now just cherry picked some features for Abi so she can use 1.5.0

I think as long as we can do non-uniform bin edges I am happy in abandoning TH2Poly.
Even though I conceptually love idea of having banana, cat etc. binning that TH2Poly support, it caused me much more issues than worth.

[github-actions]

Stale issue message