Update and clean up calo PID variables going into global PID (for electrons)
Recent developments of low-level calo PID variables have to be ported to higher-level global PID variables (ProbNNs, DLLs)
-
Update global PID input to using new selective track-matching and brem variables from Rec!2116, Rec!2267 -
Remove duplicates (very correlated variables) like some variable + its likelihood version to shorten variable list for global PID. -
Normalize variables that depend on energy/momentum: e.g. EcalE
toEcalE/p
,HcalE
toHcalE/p
. Aside from better performance, this will also reduce sensitivity in ProbNN training to particle type source kinematics (we want to distinguish type, not source). -
The EcalE
variable can/should be replaced by the variables from Rec!2319- E/p with cell selection on minimum expected energy fraction
- Summed DLL of E/p per cell, parametrized on expected energy fraction
- MR is essentially ready, gives significant PID improvement (see Rec!2319 for more info)
-
Ensure global PID is not trained on electrons from pion and kaon decays. Check the difference in performance on training with or without electrons from converted photons. Ideally, electron PID should be optimized on electron coming directly from b/c-hadron decays. -
For DLLs -
For ProbNNs
-
-
Update variable to DLL conversion tables