Choose timezone
Your profile timezone:
B2G ttbar analysis meeting
→
Europe/Zurich
Description
Salvatore Rappoccio is inviting you to a scheduled Zoom meeting.
Topic: B2G ttbar analysis meeting
Time: Mar 31, 2020 04:00 PM Zurich
Join Zoom Meeting
https://cern.zoom.us/j/669690365
Meeting ID: 669 690 365
One tap mobile
+41432107042,,669690365# Switzerland
+41432107108,,669690365# Switzerland
Dial by your location
+41 43 210 70 42 Switzerland
+41 43 210 71 08 Switzerland
+41 31 528 09 88 Switzerland
+33 1 7037 9729 France
+33 7 5678 4048 France
+33 1 7037 2246 France
Meeting ID: 669 690 365
Find your local number: https://cern.zoom.us/u/asrJMsg2b
Join by SIP
669690365@zoomcrc.com
Join by H.323
162.255.37.11 (US West)
162.255.36.11 (US East)
221.122.88.195 (China)
115.114.131.7 (India Mumbai)
115.114.115.7 (India Hyderabad)
213.19.144.110 (EMEA)
103.122.166.55 (Australia)
209.9.211.110 (Hong Kong)
64.211.144.160 (Brazil)
69.174.57.160 (Canada)
207.226.132.110 (Japan)
Meeting ID: 669 690 365
- Sal, Cecilia, Robin, AC, Brendan Regnery, Christine, Garvita, Yuri, Henrick Jabusch, Hugo Becerril, Ia, Johan Sebastian Bonilla Castro, Juska, Kevin Nash, Ksenia, Reyer Band, Robin, Titas, Xuan Chen.
- Move to combine (definitely)
- Models (from Ia): several new ones (see Ia’s slides)
- Also talk to Bogdan to get some predictions for his new models: narrow same-sign ttbar (see Monday’s plenary talk)
- Christine: is the DM model the same as ATLAS? (https://indico.cern.ch/event/730562/contributions/3010701/attachments/1654440/2647868/CMSB2G_180523_KBehr.pdf)
- Ia: This is what they used for 13 TeV l+jets channel.
- Graviton, gluon, SSM Z’, and then reweighted existing models to others.
- Ia: Coupling does affect kinematics, so reweighting doesn’t necessarily do the right job. Should just produce.
- Christine: DM: Is this complicated politically?
- Ia: we still have only ttbar in final state, should go with us.
- Robin: If DM is one of several, then this shouldn’t be a problem. It’s a reinterpretation. We will probably have to present in EXO though.
- Christine: This should be appreciated in any case.
- Interference with SM ttbar. Heavy Higgs -> ttbar used it.
- Robin: This was in ttbar and failed to converge. Then it moved to Higgs.
- Sal: Should still think about spin-1 SM interference.
- Ia: ATLAS did spin zero in the past but I don’t know what they are doing now.
- Ia: Need to request MC.
- Cecilia: Before Anastasia left, she spent some time producing cards for various configurations (along with Roman). Did we check what signals are already requested?
- Ia: I think for UL nothing is requested. Pre-UL 2016 and 2017 were requested. A month ago asked Anastasia if she made progress, and she never replied.
- Ksenia: Not sure what status is.
- Ia: Also wanted to add HVT
- Cecilia : Found email from 2018
- Robin: Yes, wanted to combine HVT with other searches.
- Sal: May also be able to be added to “diboson” combination.
Titas:
- Sal: Measurements probably need to go to UL
- Christine : lots of experience with unfolding, also can check on work from Ashley
- Ia: Should check lepton ID
- Ia: New signal samples may not be doable with pre-UL MC. This will go into UL request.
- Cecilia: What is the deadline?
- Ia: Unclear to me.
- Sal: Signal samples are probably fine to have with pre-UL MC.
Tagger:
- Robin: Excited to see what ImageTop can do. Been updating the BEST tagger, which is a multi-object tagger not focused on ttbar. Added b-tagging efficiency. The confusion matrix is looking good to distinguish top from other things. Haven’t really looked at ttbar, been focusing on VLQs. Have a H->bb image tagger from Brendan for HH resonances. May be a good study to do apples-to-apples comparisons to put into the AN. We have a paper that Justin + company wrote last year, but stuff has evolved. Traditional one Christine optimized is still the baseline to compare to until we have a new baseline.
- Garvita: Has there been studies of DeepAK8 for top and W both? For b-tagging using DeepJet?
- Kevin: From Image side. There were a lot of cross-checks with DeepAK8. If we need to use mass decorrelation ImageTop is slightly better than others. Hopefully in next nanoaod but unsure. We have HOTVR image taggers too at lower pt. There’s also a Z-gamma tagger that could be a non-isolated top tagger. Not in nanoaod.
- Sal: What is the status for nanoaod?
- Kevin: Not in yet, but probably easiest to put into common JMAR-ntuple and maybe into the nanoaod.
- Sal:
- Yuri: Since there is a little momentum to add image top into nanoaod, can we get into
- Sal : Some things have large memory footprint, so there’s
- Christine: from JMAR we recommend taggers that have personpower for maintaining SFs as well. If there’s an argument that there’s support PLUS a large group of analyses that are using it, then it’s a stronger case than getting it in.
- Yuri: Then it’s a chicken-and-egg,
- Christine: Support for maintaining scale factors is important. Also need technical implementations.
- Kevin: If you’re in nanoaod, then it’s easier.
- Kevin: Should want the “baseline” ones in nanoaod for sure. But then there’s a thousand new ones, so we need the JMAR-tuple solution.
- Sal: Lots of technical
- Christine: Next week’s JME meeting is devoted to discussion between JME and BTV to get whatever we need into the JMAR-tuples.
- Sal: Still need head-to-head comparison
- Christine: So do some formal high-pt study?
- Sal: I think we should do an apples-to-apples comparison of the full event selections, at least in MC. Not necessarily background estimates.
- Robin: need to talk about who is doing what, we’re still in the VLQ BEST publication but may need to spend time on top tagging to see how it compares. BEST didn’t use mass decorrelation. Is it a good idea to go through the full analysis?
- Yuri: Background is also critical.
- Sal: Yes, not suggesting full background, but at least MC event level studies.
- Kevin: W’ VLQ search. Can produce basic selections.
- Framework:
- Sal: Propose to use nanoaod. Also propose to use a vectorized framework like RDataFrame or Coffea.
- Coffea uses uproot and numpy arrays
- Cecilia: We’re using nanoaod for sure. UHH did a lot of work in their framework. Using upper level code to have UHH inputs also.
- Titas: Using numpy arrays to convert as input to NNs
- Sal: Extremely good point, using NNs with uproot/etc is the way to go.
- Robin: BEST isn’t in the standard NANOAOD but can put it into JMAR tuples.
- Christine: Yes, we’ll put it into extended.
- Ia: There was a HN question from Julie about lepton-jet cleaning in nanoaod. Is there enough information to do that?
- Sal: Cannot do it in standard nanoaod but can in extended version.
- Christine: Coordinating with BTV but want to use same code base for this.
- Ia / Garv / Juska : Moving to coffea-based?
- Sal: Need storage and compute nodes to be the same. Need specific worker nodes that run coffea-dask.
- Christine: Corrections and things need to be there. Also people who are FNAL local can bug the coffea people.
- Juska: what about GPU clusters?
- Sal: Not necessary but I think very nicely incorporated.
- Garvita: LPC is fine. There will be a lot of github diving into coffea.
- Kevin: Can also do RDataFrame to make real-time skims. That could be a standard event loop.
- Ksenia + Henrich: Open to change but talking with Roman
- Cecilia: Before we had a lot of support for various things, so a bit hesitant to move along. An incentive would be to move something less heavy and more directed. We have a selection that takes some time, but ntuples after preselection is very fast.
- Sal: not needed to use the same framework but we should use nanoaod as the baseline.
- Titas: it has miniaod as input. Just got used to UHH framework. Tried and tested, can run all three years from same framework with different flags. Need similar input files to combine.
- Dilepton + hadronic plan to move to vectorize something, l+jets probably not.
- Ia: What are timescales of nanoaod production?
- Christine: V7 is coming out soon-ish. Takes a bit to make the extended nanoaod, we also need space to put these things. Probably still space at FNAL B2G area. Takes about a week to get a significant portion created. Monday will focused on JME and BTV needs, but we can get a good sense for what is overlapping. At least there are some people leaning toward producing samples with all PF Candidates because it is useful for MET studies. Want the code base to have different flags for different tiers of pf candidates in nanoaod (none, AK4+AK8, AK8, all). After Monday will have a sense for what samples can be done. Can utilize the QCD (and ttbar?) backgrounds, we perhaps can just do the signal samples.
- Ia: If all PF cands are there but for which datasets.
- Christine: If we have samples with all PF candidates we just have to make sure we have the
- Going forward:
- Slack channel (Sal will send invites)
-
AOB:
- Johan: Interested in looking into tagger-on-tagger studies and looking into coffea from my side.
There are minutes attached to this event.
Show them.
The agenda of this meeting is empty
