Test RNN locally

I'm opening this ticket to keep track of some validation that I'll be doing with hartman and mguth. The starting point is going to be getting the json file that Nicole produces into the ntuple dumper config.

I suspect that some things will crash, since not everything Nicole is doing is going to be supported right away. To modify this I would check out FlavorTagDiscriminants using git-fatlas. There are a few places that may require edits:

• The track sort order might need a new enum, a new sort function and a new regex to define the sort order.
• The track selection might also need some tweaks. Feel free to edit IP3D_2018 however needed.

After all this is working someone needs to read in the output file and recompute the RNNIP scores between Keras and the output, to validate that everything is correct.

I'll be able to help out here, and plan to do a lot of this work myself, but I wanted to put everything down here so that anyone who wants to help can.