Reconstruction WG Meeting Minutes
September 16th, 2024
11:00 - 11:40 AM EDT
==================================
Introduction
Derek Anderson, Shujie Li
=========================
- New meeting time!
- 11 am EDT on Mondays
- Planning several topics
- UGM follow-up with TOF
- Muon ID w/ cluster shapes
- Timeslice-based reco
- IRT V2 follow-up
- Revisiting algorithm interface Discussion
- Q&A sessions
- Good to start soon
- Will discuss w/ Users Learning WG
Proposal for Cleaning Up Reco Algorithms
Derek Anderson
========================================
- Today is a proposal for feedback
- This is not a high-priority item
- The "reco" category of algorithms has a lot
of disparate things sitting in there
- They span multiple processes (particle reco
vs. jet reco)
- And span multiple stages of reconstruction
(early and late synthesis)
- Could clean up algorithms and corresponding
plugin (reco.cc) by divvying things up
- 3 possible categories:
(a) Particle reconstruction (electron/general
particle reco + transformations)
(b) Jet reconstruction
(c) Kinematic calculation
- Remainder of things (Charged particle selector
and MC Smeared Particle) would go into meta
category
Discussion
----------
Stephen Maple
- Have been working on electron reconstruction, trying
to addi in ioslation cut to the DIS electron identification
- But have found that I need to duplicate the E - Pz algorithm
to apply the isolation cut
- That seems wasteful, so is it possible to evolve the E - Pz
algorithm somehow?
Derek Anderson
- Plan is to expand the DIS electron indentification to use
more criteria, so long term more criteria like an isolation
cut will likely get folded into that algorithm as it gets
improved into a general DIS electron indentifier
- Tyler better able to speak on topic than I am
Tyler Kutz
- Would suggest holding off implementing isolation cut in
EICrecon
- We're planning pretty dramatic changes in this areas
of there construction, so don't want to have to rip
out new work in a few weeks
- Unless these changes have already been made?
Derek Anderson
- Not yet
- For context: is this isolation cut for pTDR plots?
Stephen Maple
- No, but I think it would be useful in general
Derek Anderson
- Got it: wanted to check if we missed a need for the
pTDR from the PWGs
Follow-Up on Track-Cluster Type Proposal
Derek Anderson
========================================
- Previously discussed a proposal for a new datatype,
the PseudoParticle, which would bundle detector
signals together for candidate particles
- Today is following up on a couple requests
from that meeting
- One question was on how the wiring around
the type would Look
- See diagram on slide 2
- Question mark line indicates transport of
PID signals into algorithm somehow (process is
still a WIP)
- Note that PFA is still a WIP
- Slide 3 shows diagram updated to split PseudoParticle
functionality across multiple types
- The 3 types suggested during previous discussion:
(1) Track-Cluster Associations (w/ goodness of
association recorded)
(2) Charged Particle Candidate (track + clusters
+ PID hypothesis)
(3) Neutral Particle Candidate (clusters + PID
hypothesis)
- The question mark lines indicate that PID signals
are somehow combined into PIDHypothesis objects
prior to being ingested by PFA1 and eID1 algorithms
Discussion
----------
Tyler Kutz
- Multiple types approach looks good!
- For clarification:
- When you say "PID is combined upstream" in notes
from previous meeting, does that mean RICH + DIRC
or RICH + DIRC + TOF
Derek Anderson
- RICH + DIRC + TOF! All 3 types of systems will be combined
somehow into a PIDHypothesis (so a likelihood) which is
then what gets fed into algorithms and updated
Tyler Kutz
- Good! TOF will be important for low momentum PID
- Also for clarification:
- What does "track-cluster subtraction" mean? Is that
removing clusters associated with tracks to isolate
the neutrals?
Derek Anderson
- Exactly! This is subtracting off the expected energy
deposited by a track from a cluster
- If the leftover energy is functionally 0, cluster
is removed
- Otherwise, leftover energy and unmatched clusters
fed into 2nd step to form neutral particles
Tyler Kutz
- Got it
- Lastly, want to caution against combining EMCal and
HCal clusters upstream
- Knowing the relative depositions in the EMCal vs.
HCal is extremely useful for electron ID (for
example)
Derek Anderson
- Good point!
- I can't remember the chain of reasoning that led
to the suggestion of combining clusters upstream
Rosi Reed
- It's also useful for even pi0's!
- One thing that could be useful would be cluster-
cluster associations
- That way users don't have to loop over everything
themselves and make themselves
- Could store the goodness of the association like
the track-cluster assocations
- This could happen in advance of making topoclusters
Tyler Kutz
- So would this be for clusters matched to tracks?
Rosi Reed
- This would be for clusters in general, not necessarily
those matched to tracks
- Then topoclusters could be made downstream of
associations for users who need them
Derek Anderson
- I like that idea, and Tristan is working on the
algorithm which would make Topoclsuters
- Also I can see how this might fit into the updated
scheme
- A leg would be added to make cluster-cluster
assocations which would then feed into PFA2
AOB
===
Derek Anderson
- I'll contact Pavel about which KFParticle repo
to use for the container
Wouter Deconick
- Sounds good. A centralized repository for
external packages is really crucial for A
couple reasons
(a) Sustainability: we're not going to be
able to support an ePIC-only KFParticle
package
- Other packages, like ACTS, have models
where experiments contribute to a single
repository
(b) Reviews: we need to be able to answer
questions like "How does ePIC depend
on extrenal pakages like KFParticle?"
- For other packages, this is clear. But
KFParticle right now, not so much...
Derek Anderson
- Agreed!