Missing inputs for DeepDoubleX and DeepAK8

cms-jet / PFNano

NanoAOD framework for testing jet algorithms.

7 stars 54 forks source link

Missing inputs for DeepDoubleX and DeepAK8 #12

Open jmduarte opened 4 years ago

jmduarte commented 4 years ago

In general, I'm interested in using this format as a standard for training/re-training deep jet taggers. So it'd be nice to have a complete set of inputs (enabled with some flag).

For DeepDoubleX, we're missing the 27 "jet-level" inputs that are the same as the ones used for the original double-b tagger. For example, they're listed here: https://github.com/cms-sw/cmssw/blob/CMSSW_10_6_X/DataFormats/BTauReco/interface/BoostedDoubleSVTagInfoFeatures.h (except one of them is not used).
For DeepAK8, we're missing the track covariance matrix entries (possibly among a few others): https://github.com/cms-sw/cmssw/blob/CMSSW_10_6_X/RecoBTag/FeatureTools/plugins/DeepBoostedJetTagInfoProducer.cc#L76-L88)

I can try to help add these in a pull request, although I think I'll need some guidance. In particular, how can we add variables to the jet table?

@andrzejnovak

andrzejnovak commented 4 years ago

I think for the "jet-level" variables there would have to be a new module, similar to the JetConstituents table.

Adding additional per particle inputs should be more straightforward, I think it would just require replicating some of this logic somewhere around here.

rappoccio commented 4 years ago

We can probably do all of it within the JetConstituentTableProducer, no?

nsmith- commented 4 years ago

For jet-level ones I think I got a solution in https://github.com/cms-jet/NanoAODJMAR/compare/master...DAZSLE:DeepDoubleX Javier is going to fill in the details (since I'm not so familiar with the variables)

alefisico commented 4 years ago

Just merged #14 Is there something else missing?

andrzejnovak commented 4 years ago

Yeah the pf cand related info

On Wed, Jun 3, 2020, 10:05 Alejandro Gomez Espinosa < notifications@github.com> wrote:

Just merged #14 https://github.com/cms-jet/NanoAODJMAR/pull/14 Is there something else missing?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/cms-jet/NanoAODJMAR/issues/12#issuecomment-638033777, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADE5EBBX4OWXSQW3HPQ2TTDRUX76JANCNFSM4NIXGGBA .

alefisico commented 4 years ago

Yeah the pf cand related info … On Wed, Jun 3, 2020, 10:05 Alejandro Gomez Espinosa < @.***> wrote: Just merged #14 <#14> Is there something else missing? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#12 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADE5EBBX4OWXSQW3HPQ2TTDRUX76JANCNFSM4NIXGGBA .

is there something still missing here?

andrzejnovak commented 4 years ago

I browsed through this yesterday, all the per jet and SV information is in. It's a bit trickier with the PF cands. Ideally, we'd be able to get the info from the FeatureProducers for the taggers themselves, but I didn't figure it out yet.