Non-fiducial ECal ParticleNet Study

[DuncanWilmot]

Issue description

Train ECal ParticleNet on non-fiducial events to determine whether this subset is problematic for an ECal-only ParticleNet model. We are planning to have two people work on this project and compare results to cross-validate

Tasks

• Skim events with a cut that removes all fiducial events (my current thinking is to use a shared directory of processed files and make the additional fiducial cut in dataset.py instead of file_processor.py)
• Train and Evaluate ParticleNet on these events
• Plot results (ROC curves and pT bias)

Existing material to start off

• A description of fiducial events and some example code can be found at Issue#27
• You can take a look at some v14 ParticleNet code but keep in mind this is in development and some of it includes HCal information
• I will organize a more succinct directory with stable v14 ECal ParticleNet code. Once that is done I will merge it with IncandelaLab LDMX-scripts and update this issue with a link to the new directory

[aether-2023]

Missing cellmodule.txt for v14 (LDMX-scripts/GraphNet/data/v14), which is needed when comparing ECal face projection of incoming particle and the ECal cells position. Did cellmodule.txt changed for v13 and v14? May I use the v12 cell geometry for non-fiducial study?

[tvami]

@DuncanWilmot did you see this?

[DuncanWilmot]

@aether-2023 Answered on Mattermost but the file is at /home/duncansw/LDMX-scripts/GraphNet/data/v13/cellmodule.txt. Same for v14 as for v13