GSI Forum
GSI Helmholtzzentrum für Schwerionenforschung

Home » PANDA » PandaRoot » Tracking » Changes in PndTpcRiemannTrackingTask
Changes in PndTpcRiemannTrackingTask [message #12336] Mon, 25 July 2011 15:21 Go to next message
Johannes Rauch is currently offline  Johannes Rauch
Messages: 41
Registered: September 2010
Location: TUM
continuous participant
From: *natpool.mwn.de
Dear Colleagues,

in the last days I made several changes and Improvements in the Pattern Recognition.
Many of the tracking parameters have changed their optimal settings, and new setters have been introduced.

In order to keep things simple, I set all parameters to reasonable default values in the PndTpcRiemannTrackingTask.

I also changed all the macros and took out the setter functions.
If you use your own macros, please use the PndTpcRiemannTrackingTask as follows, then all the default values will be used. They are already tuned and will be updated if necessary.


PndTpcRiemannTrackingTask* tpcSPR = new PndTpcRiemannTrackingTask();
tpcSPR->SetPersistence();
tpcSPR->useGeane(); // use RKTrackrep and GeaneTrackrep
tpcSPR->SetMCPid(); // use ideal particle identification
//tpcSPR->SetPDG(211);
fRun->AddTask(tpcSPR);


If you want you can use a specific pdg hypothesis, but then remember to comment out tpcSPR->SetMCPid();


If you know what you are doing and/or want to try different track finding parameters, please refer to PndTpcRiemannTrackingTask.h. I put comments there so that you understand what the parameters are doing.

Otherwise I recommend using the default values.

I hope recent changes will improve efficiency!

regards,

Johannes
Re: Changes in PndTpcRiemannTrackingTask [message #12338 is a reply to message #12336] Mon, 25 July 2011 21:53 Go to previous messageGo to next message
Dima Melnychuk is currently offline  Dima Melnychuk
Messages: 213
Registered: April 2004
Location: National Centre for Nucle...
first-grade participant
From: *play-internet.pl
Hi Johannes,

Unfortunately after your modification PndTpcRiemannTrackingTask still shows very low efficiency with ideal pid hypothesis.

For eta_c channel after your updates multiplicity of reconstructed charged particles looks like (with 4 generated kaons).
index.php?t=getfile&id=6549&private=0
Wheras before these modifications with pion particle hypotesis
index.php?t=getfile&id=6550&private=0

So ideal pid still does not work here properly.

Dima
Re: Changes in PndTpcRiemannTrackingTask [message #12339 is a reply to message #12338] Mon, 25 July 2011 23:06 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: 2.158.188*
Hi Dima,
Have you maybe tried to run kaon hypothesis with tpc, just to understand if this changes a lot the results (also vertex fitting) compared to pion hyp. If not, maybe the problem stays elsewhere.
Re: Changes in PndTpcRiemannTrackingTask [message #12340 is a reply to message #12336] Mon, 25 July 2011 23:24 Go to previous messageGo to next message
Dima Melnychuk is currently offline  Dima Melnychuk
Messages: 213
Registered: April 2004
Location: National Centre for Nucle...
first-grade participant
From: *play-internet.pl
Hi Stefano,

I tried kaon hypothesis and results (eta_c reconstruction efficiency and mass resolution) were very similar. I cannot quantitate the differnce between kaon and pion hypothesis since I tried them on different data sets, but the difference was within 10 %.

And I can try it once again to check the vertex resolution.

Dima
Re: Changes in PndTpcRiemannTrackingTask [message #12341 is a reply to message #12336] Tue, 26 July 2011 09:30 Go to previous messageGo to next message
Dima Melnychuk is currently offline  Dima Melnychuk
Messages: 213
Registered: April 2004
Location: National Centre for Nucle...
first-grade participant
From: *play-internet.pl
Hi Johannes,

despite the problem with ideal pid, running updated tpc code with pion hipotesis I see an improvement in eta_c reconstruction efficiency 8.6% vs 5.9% before the update. However it's still too low in comparison with STT.

And muliplicity of reconstructed charged particles with pion hipothesis and updated tpc code.

index.php?t=getfile&id=6551&private=0

Dima
Re: Changes in PndTpcRiemannTrackingTask [message #12343 is a reply to message #12336] Tue, 26 July 2011 11:33 Go to previous messageGo to next message
Johannes Rauch is currently offline  Johannes Rauch
Messages: 41
Registered: September 2010
Location: TUM
continuous participant
From: *natpool.mwn.de
Hi,

we are still working on some other problems:
- There is a bug when the Kalman tries to invert the charge
- We suffer from inefficiencies due to sectorization

I hope we can submit some fixes still today.

regards,

Johannes
Re: Changes in PndTpcRiemannTrackingTask [message #12344 is a reply to message #12343] Tue, 26 July 2011 11:53 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Johannes,
Quote:

There is a bug when the Kalman tries to invert the charge

just to understand, what do you mean by "the Kalman tries to invert the charge"? Is this some kind of feature which is inside the base Kalman classes and affects also the STT or it is only in the TPC reconstruction classes?
Cheers,
Lia.
Re: Changes in PndTpcRiemannTrackingTask [message #12345 is a reply to message #12344] Tue, 26 July 2011 12:12 Go to previous messageGo to next message
Johannes Rauch is currently offline  Johannes Rauch
Messages: 41
Registered: September 2010
Location: TUM
continuous participant
From: *natpool.mwn.de
Hi Lia,

the GFKalman treats q/p as a free parameter (which can also change sign).
But in the RKTrackRep, the charge was fixed. I fixed this bug, and now the sign of the charge is set according to fState[0][0] in setData().

I didn't check GeaneTrackRep yet, but maybe there is the same problem.

This problem resulted in a very bad fit if the charge was set wrong.

regards,

Johannes
Re: Changes in PndTpcRiemannTrackingTask [message #12346 is a reply to message #12345] Tue, 26 July 2011 13:15 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Johannes,
thanks for your reply.

Quote:

the GFKalman treats q/p as a free parameter (which can also change sign).
But in the RKTrackRep, the charge was fixed. I fixed this bug, and now the sign of the charge is set according to fState[0][0] in setData().

With your fix in the RKTrackRep, is the charge set to a fixed value along all the Kalman procedure, i.e. the Kalman is not allowed to change the sign of q/p anymore, or does the fix just set the final charge back to the input one when the Kalman output has the wrong sign (after all the fitting procedure)?

Quote:

I didn't check GeaneTrackRep yet, but maybe there is the same problem.

Are you looking also into GeaneTrackRep to check whether there is the problem or not? Just to know whether I have to investigate this by myself or not. Wink
Cheers,
Lia.
Re: Changes in PndTpcRiemannTrackingTask [message #12347 is a reply to message #12346] Tue, 26 July 2011 13:45 Go to previous messageGo to next message
Johannes Rauch is currently offline  Johannes Rauch
Messages: 41
Registered: September 2010
Location: TUM
continuous participant
From: *natpool.mwn.de
Quote:

With your fix in the RKTrackRep, is the charge set to a fixed value along all the Kalman procedure, i.e. the Kalman is not allowed to change the sign of q/p anymore, or does the fix just set the final charge back to the input one when the Kalman output has the wrong sign (after all the fitting procedure)?


The Kalman is fitting q/p, and the sign of q/p can change during the fitting process.
The problem was that a change in sign of q/p was not correctly passed to the RK propagator, which is now fixed.

That means: The absolute value of the charge is fixed, its sign can change.


I just looked into GeaneTrackRep and there everything seems to be fine.

cheers,

Johannes
Re: Changes in PndTpcRiemannTrackingTask [message #12348 is a reply to message #12347] Tue, 26 July 2011 14:01 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Quote:

That means: The absolute value of the charge is fixed, its sign can change.

Then you have a found track with a charge sign as input to the Kalman and AFTER the Kalman fit you can have that the charge has changed sign, right?

Since the Kalman changes the sign at runtime, could it happen that the charge associated to the last hit (last params I mean) has opposite sign with respect to the one of the first hit?
Cheers,
Lia.
Re: Changes in PndTpcRiemannTrackingTask [message #12349 is a reply to message #12348] Tue, 26 July 2011 14:07 Go to previous messageGo to next message
Johannes Rauch is currently offline  Johannes Rauch
Messages: 41
Registered: September 2010
Location: TUM
continuous participant
From: *natpool.mwn.de
Quote:

Then you have a found track with a charge sign as input to the Kalman and AFTER the Kalman fit you can have that the charge has changed sign, right?


Yes.

Quote:

Since the Kalman changes the sign at runtime, could it happen that the charge associated to the last hit (last params I mean) has opposite sign with respect to the one of the first hit?


When smoothing is not turned on, the track has only one state, and therefore only one charge.

I don't know what happens with smoothing turned on.
But what I saw in my display ist that usually in the second iteration the Kalman flips the charge if necessary. So with 3 iteration the charge should be consistent for all hits.

cheers,

Johannes
Re: Changes in PndTpcRiemannTrackingTask [message #12350 is a reply to message #12349] Tue, 26 July 2011 14:17 Go to previous messageGo to next message
Anonymous Poster From: *pools.arcor-ip.net
Hi,

even without smoothing we do forward and backward fits. So in principle we could have a different charge in the last point (from the forward fit) and the first point (from the backward fit).

However: Since we do repeated Kalman iterations this should never happen. A wrong charge hypothesis could be turned around in the first forward fit, but then it will stay. So with >=2 iterations this should never happen.

Cheers, Christian
Re: Changes in PndTpcRiemannTrackingTask [message #12351 is a reply to message #12350] Tue, 26 July 2011 15:47 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Christian

Quote:

However: Since we do repeated Kalman iterations this should never happen. A wrong charge hypothesis could be turned around in the first forward fit, but then it will stay. So with >=2 iterations this should never happen.

Actually for the STT we saw that more than 1 iteration does not improve the results, for some momenta (lower ones) it even worsen the efficiency.

For this reason we would like to run the reconstruction with only one iteration. This is the best option for us.

Anyway, we will check how many times, with one iteration, we get that the Kalman sets the first and last hit charges to a different value and how many times it changes the charge sign to the wrong one. Maybe it does not happen so often to make us to worry about.

Ciao,
Lia.
Re: Changes in PndTpcRiemannTrackingTask [message #12356 is a reply to message #12336] Tue, 26 July 2011 20:03 Go to previous messageGo to next message
Johannes Rauch is currently offline  Johannes Rauch
Messages: 41
Registered: September 2010
Location: TUM
continuous participant
From: *natpool.mwn.de
Quote:

we are still working on some other problems:
- There is a bug when the Kalman tries to invert the charge
- We suffer from inefficiencies due to sectorization

I hope we can submit some fixes still today.



I just commited a new version of the ClusterFinderSimple which gets rid of cluster splitting at the edges of the sectors.

This should improve efficiency a bit.


cheers,

Johannes
Re: Changes in PndTpcRiemannTrackingTask [message #12360 is a reply to message #12356] Wed, 27 July 2011 11:20 Go to previous messageGo to next message
Johannes Rauch is currently offline  Johannes Rauch
Messages: 41
Registered: September 2010
Location: TUM
continuous participant
From: *natpool.mwn.de
I forgot to commit the updated padplane file which is necessary.
It is now checked in.
Re: Changes in PndTpcRiemannTrackingTask [message #12362 is a reply to message #12336] Wed, 27 July 2011 12:00 Go to previous message
Dima Melnychuk is currently offline  Dima Melnychuk
Messages: 213
Registered: April 2004
Location: National Centre for Nucle...
first-grade participant
From: *fuw.edu.pl
Hi Johannes,

I've redone the eta_c reconstruction with updated code but still without updated padplane file.

And results are worse then before.
Using ideal pid in PndTpcRiemannTrackingTask I still have as before 0 reconstructed eta_c with 1000 events statistics.
But using kaon hypothesis efficiency decreased from 8.3% to 3.7%.

So I hope it is related to wrong padplane file and I will rerun digitiation/reconstruction now.

Dima
Previous Topic: PndSttMvdGemTracking with MC pdg.
Next Topic: bug in PidCorrelator
Goto Forum:
  


Current Time: Wed Dec 04 13:37:57 CET 2024

Total time taken to generate the page: 0.00922 seconds