ttH killer development

[tcoulvert]

URGENT:

• message Nitish about BSM samples to be processed in v2 parquets
• fix input comparison plots for train/val/test as train/val don't look like test
• compute rough k-factors for MC-Data agreement -> OR simple data-driven bkg estimate via linear/exponential fit in the sidebands
• write a template JSON config for the run script -> cover the main current use cases e.g. process ttH vars, move to ttH sideband, etc

Non urgent:

• refactor plotting scripts to use ROOT rather than hist+mplhep
• make EOS website to host plots for easy sharing and saving
• compare different hyperparams (extra variables, BCE vs. NLL Losses, etc)
• explore different standardization/normalization methods
• add electron1/2, muon1/2 variables
• make MR into HiggsDNA HHbbgg_Updates to include jet idx of each bjet, to make jet work easier
• explore different architectures -> RNN currently not necessary