Difference between revisions of "Aod ntuple"
m |
|||
| Line 5: | Line 5: | ||
== Electron == | == Electron == | ||
| + | |||
| + | There are 3 types of quality cuts you can perform on the electron candidates: | ||
| + | |||
| + | <ol> | ||
| + | <li>Cuts based eID</li> | ||
| + | <li>Likelihood based eID</li> | ||
| + | <li>NeuralNet based eID</li> | ||
| + | </ol> | ||
| + | |||
| + | - alg 1 used both calorimeter and tracking information in addition to TRT | ||
| + | information. In 9.0.4 there is a problem with TRT simulation so one has to | ||
| + | mask TRT bit to recover the lost efficiency. The variable which combine | ||
| + | everything is isEM and it is used as follow: | ||
| + | |||
| + | elec*->isEM() | ||
| + | |||
| + | to mask TRT bit you should use: elec*->isEM()%16==0 | ||
| + | |||
| + | if you use isEM then you will select electrons with an overall eff of about | ||
| + | 80% in the barrel but much lower in the crack and endcup. | ||
| + | |||
| + | |||
| + | -alg 2 use likelihood ratio and use the following variable: | ||
| + | |||
| + | energy in different calorimeter sampling, shower shapes in both eta and phi | ||
| + | and E/P. No TRT information is used here. You need to access two variables | ||
| + | called emweight and pionweight then you can construct the likelihood ratio: | ||
| + | defined by: emweight/(emweight+pionweight). Emweight is the product of pdf's | ||
| + | for electrons and pionweight is the product of pdf's for pions. In AOD, you | ||
| + | use the following code: | ||
| + | |||
| + | ElecEMWeight = elec*->parameter(ElectronParameters::emWeight); | ||
| + | ElecPiWeight = elec*->parameter(ElectronParameters::pionWeight); | ||
| + | |||
| + | Then form the variable: | ||
| + | X = ElecEMWeight/(ElecEMWeight+ElecPiWeight) | ||
| + | |||
| + | Requiring X > 0.6 will give you more than 90% eff for electrons | ||
| + | |||
| + | |||
| + | -alg 3 use the NeuralNet and use as inputs the same variables used for | ||
| + | likelihood. To use it in AOD you should proceed as follow: | ||
| + | |||
| + | ElecepiNN = elec*->parameter(ElectronParameters::epiNN); | ||
| + | |||
| + | Requiring ElecepiNN > 0.6 will give you about 90% eff for electrons. | ||
| + | |||
| + | However, you should be aware that the ANN was trained in full eta range | ||
| + | while the likelihood was computed in 3 bins in eta: barrel, crack and | ||
| + | endcup. So I would suggest to use likelihood for now. I am working on | ||
| + | improving things but it will not be ready for Rome. For sure using | ||
| + | likelihood will give you 90% eff for electrons with a good jet rejection. | ||
== Muon == | == Muon == | ||
Revision as of 11:42, 19 May 2005
This page contains basic prescriptions to get physics objects from the AOD and the AOD-based Root ntuple.
Some comments on quality selection cuts will be added as work progresses.
Electron
There are 3 types of quality cuts you can perform on the electron candidates:
- Cuts based eID
- Likelihood based eID
- NeuralNet based eID
- alg 1 used both calorimeter and tracking information in addition to TRT information. In 9.0.4 there is a problem with TRT simulation so one has to mask TRT bit to recover the lost efficiency. The variable which combine everything is isEM and it is used as follow:
elec*->isEM()
to mask TRT bit you should use: elec*->isEM()%16==0
if you use isEM then you will select electrons with an overall eff of about 80% in the barrel but much lower in the crack and endcup.
-alg 2 use likelihood ratio and use the following variable:
energy in different calorimeter sampling, shower shapes in both eta and phi and E/P. No TRT information is used here. You need to access two variables called emweight and pionweight then you can construct the likelihood ratio: defined by: emweight/(emweight+pionweight). Emweight is the product of pdf's for electrons and pionweight is the product of pdf's for pions. In AOD, you use the following code:
ElecEMWeight = elec*->parameter(ElectronParameters::emWeight); ElecPiWeight = elec*->parameter(ElectronParameters::pionWeight);
Then form the variable: X = ElecEMWeight/(ElecEMWeight+ElecPiWeight)
Requiring X > 0.6 will give you more than 90% eff for electrons
-alg 3 use the NeuralNet and use as inputs the same variables used for
likelihood. To use it in AOD you should proceed as follow:
ElecepiNN = elec*->parameter(ElectronParameters::epiNN);
Requiring ElecepiNN > 0.6 will give you about 90% eff for electrons.
However, you should be aware that the ANN was trained in full eta range while the likelihood was computed in 3 bins in eta: barrel, crack and endcup. So I would suggest to use likelihood for now. I am working on improving things but it will not be ready for Rome. For sure using likelihood will give you 90% eff for electrons with a good jet rejection.