- Write MC to SRawEvent files
- Do on Rivanna
- Compare Precision and Recall for Filter
- Compare Precision and Accuracy for momentum reconstruction
- Compare Precision and Accuracy for physics-based kinematic variable reconstruction.
- Compare full tracks vs incomplete tracks
- Heat map of momentum error
- Delta pz and a function of pz
- Look at actual events flagged to be complete
- What is the efficiency for those events?
- What is the error in reconstruction?
- Check if actually 10% full tracks
- Look at incomplete events
- Out of total number, how many are reconstructable with QTracker/KTracker
- What is the error in reconstruction for those?
Compare masking to training with hodoscope data or with trigger part of hodoscopes
Get generator and QTracker on Shannon
Check hodoscope occupancy
Cosmic ray mode?
Boer Mulders
- Evaluate E906 Data using QTracker
- Find \lambda, \mu, \nu based on Drell-Yan from the target and regression
- Produce and extract artificial BM function
- Optimizing the analysis and extraction to have smallest error
Questions to be answered:
- Trigger for simulated events
- J/Psi peak reconstruction
- Run a test with J/Psi, Drell-Yan, and random, try to reconstruct the peak.
- Parallelization of QTracker process
- How much is GPU based
- How much is CPU based
- IO Test
- Libraries to create CODA files
- Read them with the Decoder
- From Srawevent to reconstruction
- Reading and writing SRawEvents
- Reading done
- Writing need to know structure and use root
Later: Meet with Kenichi about DAQ/Encoder timing
- Introduction
- -Overview and why Q-tracker is needed
- -Goals
- -How we will achieve these goals
- -outline of what's to come (order of how things are broken down)
- Training Data production
- - MC production (details on quality/matching), event simulation quality/matching, trigger dependence, detector resolution dependence
- - Quantifying variational equivalence in hit matrix (hit matrix covariance) and error propagation
- - Detector efficiency map as a function of time/run (bad DC or hodo regions)
- - Detector Covariance
- - Energy loss and momentum corrections (difference needed for data to MC comparison)
- - Change in momentum accuracy and precision as a function of change in covariance
- Limitations in the MC (reducing knowns and noting unknowns)
- - quantifying the difference between ELOSS from MC and ELOSS from Experiment
- - detailed field map and alignment in MC used
- - J/psi peak position and width as a function of field configuration, resolution, eloss, momentum corrections
- Experimental data matching
- - biases in positive and negative tracks and/or geometry
- - Proof that deviations do not lead to bias or larger systematic errors
- - Quantified robustness of accuracy and precision of 4-momenta, vertex position, and J/psi and psi' peak
- - Assumptions about distributions of processes
- - True occupancy limits
- - Ensuring correct drift time implementation and impact
- Q-tracker Algorithm
- Overview with choice of DNN type for each part and why
- Describe each portion's architecture, and what it does
- Describe all configuration options for each portion
- Describe the integration of all parts together
- Training
- - Training process and philosophy for best results
- - Training evolution based on trial and error (lessons learned)
- - Folding in degrees of noise in phases of training and testing at each level
- - Validation studies with 906 data in training and in predictions
- Performance and Quality checks
- - Best px,py,pz all being better or equal to Ktracker
- - Best vx, vy, vz, and studies on improved iterative reconstruction
- - Fiducial cuts
- - Fiducial settings (requirements for optimal performance)
- - Integrating timing cuts
- - basic analysis using J/psi+psi/ width measurements after corrections to the spectrum imposed
- - Probability cut variation
- - Systematic variation of all Qtracker configuration settings
- - Detector effects not taken into account
- - Reliability of incomplete tracks and threshold conditions for cut-off
- - Combinatoric analysis after final quality and vertex cuts
- - Analysis of remaining dump and non-target produced tracks
- - Analysis of acceptance and occupancy dependence on beam intensity (rate effects)
- - Final comprehensive quantitative comparison to Ktracker