...
Variable | Importance |
AcdTotalTileEventEnergyRatio | 2034.17 |
Tkr1SSDVeto | 1201.47 |
CTBCORE | 812.28 |
AcdTkr1ActDistNorm | 575.06 |
AcdTkr1ActiveDistENorm | 476.57 |
AcdTotalEnergyNorm | 269.89 |
Tkr1ACDSideZ | 256.61 |
AcdActDistNorm | 118.93 |
Tkr1LATEdge | 107.11 |
AcdTileEnergyNorm | 106.51 |
AcdCornerDocaENorm | 81.52 |
AcdMaxTileEnergy | 74.32 |
AcdTileEventEnergyRatio | 73.48 |
AcdTkr1TileEnergyNorm | 25.67 |
AcdTileCount | 24 |
Class | Gamma Eff | Bkg Eff | Expression |
---|---|---|---|
transient | 83.5% | 0.45% | Prefilter && CPFGamProb > 0.2 |
source | 82.5% | 0.40% | transient && CPFGamProbCPFGamProb > max(.05, .55- .3* (CTBBestLogEnergy - 1.)) |
diffuse | 81.2% | 0.36% | source && AcdDiffuseVeto |
Using Using New Variables
AcdTkr1VetoSigmaHit and VetoHit1
...
For completeness here is a plot of AcdTkr1VetoSigmaHit v. AcdTkr1ActiveDist
VetoHit1 is the same variable, including the tracker SSDVeto.
VetoHit1 = sqrt(Acd2Tkr1VetoSigmaHit*Acd2Tkr1VetoSigmaHit+1.5*Tkr1SSDVeto*Tkr1SSDVeto)
Here is a plot of VetoHit1
...
Same plot, but the X-axis remapped with log10(VetoHit1+0.01)
Note the small peak in the signal distribution around 7-9 is from low energy deposits in the tile pointed to by the track.
...
For comparison, here is a plot of AcdTkr1VetoSigmaGap v. AcdTkr1RibbonDist:
In analogy with VetoHit1 above we can define VetoGap1 to take into account the Trk1SSDVeto:
VetoGap1 = sqrt(Acd2Tkr1VetoSigmaGap*Acd2Tkr1VetoSigmaGap+1.5*Tkr1SSDVeto*Tkr1SSDVeto)
Here is a plot of the low end of VetoGap1 for all events.
If we look at only those events that pass weren't vetoed because of a track tile association or b/c of excess energy in the ACD we see this:
For Comparison, here is a plot of VetoGap1 v. AcdTkr1RibbonDist for the same set of events.
AcdTkrVetoSigmaHit and VetoHit
We can also select the best association by looping over all tracks.
Here is a plot of that variable, AcdTkrVetoSigmaHit:
In analogy to VetoHit1 we can make VetoHit from AcdTkrVetoSigmaHit.
Here is a plot of that:
Then if we only look at events that weren't rejected because of VetoHit1 or excess energy:
So, we could consider cutting very loose on this and rejecting some event with a high probablity match.
Reproducing Existing Performance with simple cuts
Cut | Reason | Gamma Effic | Bkg Effic |
---|---|---|---|
VetoHit1 < 5 | Reject events with best rack pointing at struck tile/ribbon | 94.8% | 7.2% |
TotalTileEnergyCut | Reject events with excess ACD total energy | 89.3% | 1.2% |
VetoGap1 < 2 | Reject event in gaps of the ACD | 89.1% | 1.2% |
VetoHit < 2 | Reject events if other track points at hit tile | 89.0% | 1.1% |
Some other useful new Variables, (for Classification Tree)
Another type of variable we consider is the scaled energy in a cone around the track. We have three variables that reflect 15, 30 and 45 degreee cones.
Variable | AllEvents | Prefiltered Events |
---|---|---|
VetoHit1 |
|
|
AcdEngSc |
|
|
AcdEng45Sc |
|
|
AcdEng30Sc |
|
|
AcdEng15Sc |
|
|
Class | Gamma Effic | Bkg Effic | Expression |
---|---|---|---|
transient | 86.4% | 0.43% | prefilter && MVA_BDT > 0.38 |
source | 84.1% | 0.33% | prefilter && MVA_BDT > 0.43 |
diffuse | 81.2% | 0.27% | prefilter && MVA_BDT > 0.48 |