GSI Forum
GSI Helmholtzzentrum für Schwerionenforschung

Home » PANDA » PandaRoot » Analysis » combinations of pid algorithms
combinations of pid algorithms [message #13489] Wed, 16 May 2012 15:31 Go to next message
Dmitry Khaneft is currently offline  Dmitry Khaneft
Messages: 75
Registered: January 2009
continuous participant
From: *kph.uni-mainz.de
Dear all,

I have another issue. I tried different combinations of PID algorithms and found out a strange thing.

For example if I use following code with "PidAlgoEmcBayes" enabled only

theAnalysis->FillList(eplus, "ElectronVeryTightPlus","PidAlgoEmcBayes");


or another one with many others included

theAnalysis->FillList(eplus, "ElectronVeryTightPlus","PidAlgoEmcBayes;PidAlgoStt;PidAlgoMvd;PidAlgoDrc;PidAlgoDisc;PidAlgoMvd");


then the first case gives me less events compare to the second one. It seems very strange for me. I would expect that by adding new algorithms I should be getting less rather then more events.

Cheers,
Dmitry
Re: combinations of pid algorithms [message #13490 is a reply to message #13489] Wed, 16 May 2012 15:36 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *to.infn.it
Dmitry Khaneft wrote on Wed, 16 May 2012 15:31


the first case gives me less events compare to the second one. It seems very strange for me. I would expect that by adding new algorithms I should be getting less rather then more events.



Why? If you multiply your initial pdf with another pdf, you will obtain a different probability distribution. I cannot see a direct correlation of drop in efficiency with increasing number of algorithms.

Re: combinations of pid algorithms [message #13492 is a reply to message #13490] Wed, 16 May 2012 15:42 Go to previous messageGo to next message
Dmitry Khaneft is currently offline  Dmitry Khaneft
Messages: 75
Registered: January 2009
continuous participant
From: *kph.uni-mainz.de
Stefano Spataro wrote on Wed, 16 May 2012 15:36


Why? If you multiply your initial pdf with another pdf, you will obtain a different probability distribution. I cannot see a direct correlation of drop in efficiency with increasing number of algorithms.


Sorry may be I was not clear. It is exactly what I would expect but for some reason my results are opposite. Efficiency goes up with increasing number of algorithms.

[Updated on: Wed, 16 May 2012 15:43]

Report message to a moderator

Re: combinations of pid algorithms [message #13493 is a reply to message #13492] Wed, 16 May 2012 15:43 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *to.infn.it
I am not saying that this is correct. I am just saying that the two things are uncorrelated, I would say.
Re: combinations of pid algorithms [message #13494 is a reply to message #13489] Wed, 16 May 2012 15:53 Go to previous messageGo to next message
Dmitry Khaneft is currently offline  Dmitry Khaneft
Messages: 75
Registered: January 2009
continuous participant
From: *kph.uni-mainz.de
Hm. I was following tutorial and found these lines

Quote:


...

theAnalysis->FillList( looseElectrons, "ElectronLoose", "PidAlgoEmcBayes;PidAlgoDrc;PidAlgoMvd");

In this case the probabilities for PID selection are achieved by multiplying the probability values of the chosen algorithms, i.e. Pe = Pe,1 × Pe,2 × Pe,3, etc.



The only case one can get increase in efficiency is if one of the probabilities >1 what is impossible. Am I right?

[Updated on: Wed, 16 May 2012 15:53]

Report message to a moderator

Re: combinations of pid algorithms [message #13496 is a reply to message #13494] Wed, 16 May 2012 15:57 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *to.infn.it
I understand the sentence can be misleading.
The algorithm multiply PDFs (probability density functions), but the final probability is the pdf divided by the sum or all the pdf. -> pdf(e) / (pdf(e)+pdf(mu)+pdf(pi)+pdf(k)+pdf(p)).

If the denominator goes down, the probability can increase.
Re: combinations of pid algorithms [message #13498 is a reply to message #13489] Thu, 17 May 2012 10:28 Go to previous message
Ralf Kliemt is currently offline  Ralf Kliemt
Messages: 507
Registered: May 2007
Location: GSI, Darmstadt
first-grade participant

From: *pool.mediaWays.net
Hi.

In other words: when you increase the probability of a particle being a proton the other probabilities will drop accordingly.

You should also be really careful which PID you want to include for which particle species. Soem detectors might introduce unnecessary mix-ups. Always crosscheck with MC information to be safe.

Cheers, Ralf.
Previous Topic: editing pidana.par
Next Topic: access to different variables
Goto Forum:
  


Current Time: Sat Dec 14 08:21:52 CET 2024

Total time taken to generate the page: 0.00712 seconds