Minutes for LArSoft coordination meeting on Jan 15, 2019
Present | Vito Di Benedetto, Giuseppe Cerati, Lynn Garren, Herb Greenlee, Kyle Knoepfel, Saba Sehrish, Erica Snider, Hans Wenzel |
Remote | Katherine Lato, Paul Russo, Tracy Usher |
Release and Project status report [Erica Snider]
- This week:
- Move lardata/Utilities/StatCollector.h to lardataalg to allow use in gallery.
- updating to art v3_01
- Need signoff for Genie v3, there is a test release v08_01_00_01.
- MicroBooNE: decision in 1-2 weeks
- [Lynn Garren] has there been any discussion about upgrading Geant4 to the current production version?
- [Erica] do not recall any.
Conclusion:
- Need to discuss about the Geant4 upgrade
- Genie v3 signoff follow up
Proposal for a LArTPC Overlay Workshop [Adi Ashkenazi, Wesley Ketchum, Erica Snider]
- See slides for the detailed agenda, the purpose of this talk was to get a feedback on the agenda
- suggestion from [Lynn]: The outline is very thorough, but if the topics could be reworded to avoid such that it does not sound as if MicroBooNE is instructing but rather it needs to be a collaboration.
- [Herb] Do the other experiments have overlay efforts?
- [Tracy] ICARUS will be interested (on a time scale of year form now)
- [Brett] in wirecell it is a matter of where to add the overlays
- [Kyle] suggested to talk to mu2e, they still doing event mixing and have requested new features as well
Conclusion:
- Will invite people broadly, plan to reach out to background dominated analysis groups within DUNE in particular to make sure they know about it
Proposed Updates to LArSoft: LArEventDisplay LArReco [Tracy Usher]
- The updates to the event display (in LArEventDisplay) are targeted to visualization of Cluster3D data (work in LArReco)
- LArSoft event display code is challenging to make any updates (people normally look for other tools)
- LArSoft v07 series appeared to present a path to simplification, art tools
- Though have rethought this given LArSoft plans
- 3D geometry drawers
- Experiments can implement custom 3D event display drawers in a separate tool
- Generic tool can draw the outline
- Custom tool can include custom fhicl parameters, .e.g, draw “bad†channels
- 4 drawing tools available
- Default, ICARUS, MicroBooNE, ProtoDUNE
- Simulation truth information drawers
- New 3D object available, SimEnergyDeposit
- Waveform drawers: use art tools for different hits
- DrawGausHits_tool
- Draws hits from multiple sources in diff colors
- Hit center plus width, allows a floating baseline, etc
- DrawSkewHits_tool
- Intended for DP
- ICARUS also uses raw waveform hit finder
- Cluster3D
- Events are sliced, detailed steps on slides
- A more detailed presentation at the calibration/reco workshop in Dec
- Major Cluster3D updates
- more space point generic, so accepts from other inputs, space point solver, WC, etc
- Space point quality metric
- Simplified quality metric
- GausHitFinder changes (specific to ICARUS)
- removed unused fhicl definitions
- Underlying tools: updated to the "morphological filter" which is used by ICARUS, but not by any other experiment
- SpacePointSolver
- Built a tool interface to read hits in order to handle the special case of the ICARUS geometry
- Standard tool which handles all the cases SpacePointSolver was originally build for
- ICARUS specific tool to hanbdle the horizontal induction wires
- Should be no differences in the operation of the module other than the tool usage
- In order to visualize the Cluster3D work, including new features
- Option to draw space points in a heat map based on uality metric
- PC axes
Conclusion
- Merge larreco feature branch: feature/usher_cluster3dspacepoints
- Merge lareventdisplay feature branch: feature/usher_edgedrawingoptions
- will be added in the next release
Restructuring anab::ParticleID [Kirsty Duffy, Adam Lister]
- anab::ParticleID is very restrictive
- This work is addressing the issue to update the LArSoft code with every new PID algorithm
- propose to use a vector of structs where all variables are replaced with it: sParticleIDAlgScores
- Details on each member in that struct is discussed in slides, the suggestions are as follows
- using bitset size of 32 will not increase the size, suggestion to call is mask "fPlaneMask" rather than "fPlaneID"
- Breaking changes
- feature branches for larsoft and uboone are being worked on,
- no other experiments code has been changed,
- detail instructions on how to update the code is provided)
- [Kyle] suggested:
- On slide 3: ROOT doesn’t recognize GCCXML ifndef guard, so get rid of that
- make sure when returning a vector, return a const ref
- [Marc Paterno] suggested to look at the total size of the new struct to determine how many bits can be used in the bitset
Conclusion
- Work in progress, some testing needs to be done, a bug needs to be fixed
Discussion of memory management in DUNE [Brett Viren]
- Based of the earlier issue with wirecell using too much memory, which was addressed
- A lot of memory is used by the ROOT IO buffers
- Currently wirecell readout is within 2GB, but downstreamm consumes more, details of the problem and proposals in the text on indico agenda
- [Kyle] suggested to present art multi-threading work in this context at one of the next LArSoft coordination meetings
- use of TBB, which tries to avoid wasting CPU
- will look at ROOT defaults after v2.12 is out
- [Brett Viren]: right on the edge of having a problem at ProtoDUNE
- This is a good test bed
- Try to fix the problems rather than respond to crises
- [Marc Paterno] What kind of data will use data from multiple APAs simultaneously
- [Brett Viren]:
- The generic LArTPC processing chain looks at single APAs when the data is dense, up to and including signal processing
- ROIs are at the channel level, so intrinsically sparse
- Then at imaging and PR phase, would want to merge sparse ROIs across APAs
- Downstream of that, some algorrithms respond in combinatoric ways, so would need memory. But don't know whether that will be a problem
- Before signal processing, subtracting coherent noise might be an issue if the noise spans APAs, but cannot be subtracted at the APA level
- There is cross APA noise in ProtoDUNE due to bad HV supply (which has been replaced), but believe that it can be addressed one APA at a time
- LArSoft:
- Noted LArSoft multi-threading effort
- Need to think about how to do per APA processing
- The concept of an "event" is complicated for neutrino detectors
- Can we treat APAs as events?
- Would it be better to have 150 files, one for each APA, or should we have one file with all 150 APAs and deal with the per APA internally
- [Lynn]: Can we get the email into a pdf file and post it?
- Yes. Brett will do this and add it to the indico agenda item (Already posted)
Conclusion