Zoom link: https://cern.zoom.us/j/61532761187?pwd=bnd0aFIrM3JHV05oYnpOZm11cFh6dz09
Meeting ID: 615 3276 1187
Passcode: 770451
Aliya: so what is the result?
Louis: sth generic that can be plugged in many analyses
Alexander: in principle, you can use it for sth very boring in DQM :)
Alexander: and what would be the requirements?
Louis: we may want to put the intro to ML before
Alexander: will you be available to do it?
Louis: depends on the other exercises....
Patrick: when shall we know?
Louis: (...)
Alexander: about the time, should we upgrade this to a 4h-long exercise?
Patrick: I'd rather keep the 4h slots for objects themselves. We could put part of the material in lectures.
Alexander: you also subcribed for CAT, BTW?
Aliya: if it happens, it should also cover other technologies
Patrick: I thought MArcel wanted to cover workflow management here. We could also consider having only one afternoon of topical exercises and 4 afternoons of "POG/4h" exercises
Aliya: BTW, I can talk show cases tackled later on in PAG exercises
Andrea: also for POG exercise, combine may be used. An introduction to combine would be useful for everybody.
Daniel: some of these topics are rather interesting for a wider audience, perhaps? lecture?
Patrick: the programme of the lectures is not yet exactly fixed, but indeed, we could move part of this to the lectures, and make sure that the exercise is more hands-on
Alexander: how long do you think you would need? 2h are sufficient?
Juliette: not yet quite clear, should check with Artur
Matthias: what prerequisites? any coding involved?
Juliette: no strong coding exercise for sure. Question still open. Can check other exercises.
Alexander: 1) will that also cover what enters the TauID?
Andrea: I was thinking to start a 30 min talk about reco, then move to the exercise. But it's flexible, can change a bit.
Alexander: 2) who would help you?
Andrea: currently alone, need to get people on board. From DESY, anyone who may have worked on TauID. Otherwise I'd talk to people in TAU POG.
Alexander: on the full chain starting from MC to NanoAOD, does that involve sth more complicated like Gridpacks? or just showing the chain starting from a trivial P8 fragment?
Louis: not yet exactly clear. Probably fine to start from P8. But if we merge this with some generators, we could consider sth more general.
Alexander: it looks to me that there is a lot of useful information even with P8 only. Probably easier to separate from GEN exercise. Your PPD exercise would be more focused on the user experience, and leave the matching, merging, etc. to the GEN exercise.
Simone: in principle, if the GEN exercise comes before, we could coordinate?
Patrick: exercises should be self-sufficient, bc it's not guaranteed that students will even take both exercises
Alexander: have you found people to help you with the PPD exercise?
Louis: haven't found people from Hamburg so far. Could find people in Hamburg. Now, the exercise from PPD is calibrated for 2h, while here we would need 4h in Hamburg. That's why I was considering merging GEN and PPD.
Ankita: I could help for PPD.
Patrick: merging GEN and PPD exercises, we could consider it, but let's wait to see how much material we have, and what your experience with the school that will take place at CERN next week. We can continue the discussion offline.
Ankita: and in the GEN exercise, do we only want one generator or several? (e.g. we're trying to advertise Sherpa in CMS)
Alexander: I'd rather have several generators, and people should get an idea of the available parameters.
Simone: issue is that it will be the first time. It's impossible to learn how to run 5 different codes. I'd rather try 2 or 3.
Alexander: as you are 4 people, should we pick two topics? like 2 flavours of the same exercise
Daniel & Philip: that's definitely an option, also depends on how much we have
Patrick: note that we are also thinking to have topical exercises on ColumnFlow, uproot, etc., which could be considered as prerequisites to your long exercise (+ CAT exercise)
Philip: makes sense
Marcel: should be a prerequisite then, yes. Also agree to split into 2 exercises, with focus on different technicalities (H->4l, and HH)
Laurids: in the context of a TOP PAG exercise, I'd also like to have columnar analysis, there is potential synergy.
Patrick: any benefit from having a topical exercise on RDataFrame?
Andreas: the exercise already includes a tutorial on RDataFrame, so not a must
Aliya: there was an exercise on RDataFrame at DESY, right? Andrei from Aachen should also know about it
Alexander: how do we proceed to find facilitators?
Andreas: not obvious, will keep asking around
Alexander: we have a little bit of money for external helpers, but not so much
Laurids: physics case would be ttbar xsec measurement with Run 3 with columnar python-based framework. Still have to discuss the exact details.
Patrick: do you use kinematic fits?
Laurids: no kinematic fits
Patrick: prerequisites?
Laurids: sth columnar analysis?
Patrick: topical ex must be specific, one has to choose a particular technology
Laurids: so probably none
Dominic: it's one thing to ask people to use NumPy
Matthias: then my topical exercise would be a good prerequisite
Patrick: about the prerequisites, we had discussed together about a topical exercise on ML, are you still considering it?
Lisa: could be done, TBD. It depends if we decide to use ML.
Alexander: can you clarify on the low and high level analyses? will you take the decision? it's hard for me to judge on the quantity of work.
Lisa: we'd appreciate feedback
Patrick: I would be very pragmatic and avoid relying on heavy HTC jobs
Lisa & Karim: we do need statistical but we can possibly prepare the material, and just let them run some small samples, just to get a feeling of how it runs
Patrick: would make sense to talk to the Muon DPG facilitators
Lisa: sure, although we don't plan to work on reconstruction
Alexander: on the low-level analysis, what would be the physics result that they would present
Lisa: compare the clustering of the algos (...). Still have to think of a good figure of merit.
Alexander: that's my only worry with the low-level analysis. At the end, the students have to present a physics result.