Use rnnip retraining in GhostTag VR jets
We should be using the new RNNIP training as an input for the Xbb tagger. It was a bit non-trivial to get this to work since in general FTAG5 is slimmed pretty aggressively.
I did have to remove the numConstituent
filter: we don't skim jets on numConstituents
when we make the derivations, but we only save tracks pointing at the leading three subjets in each large-R jet. This means that if we skip a jet because numConstituent < 2
, the third jet has missing track links. So now instead of filtering on numConstituents
I just write numConstituents
into the output file.