Clustering

From: Peter H. L. Christiansen (pchristi@nbi.dk)
Date: Mon Sep 18 2000 - 11:20:11 EDT

  • Next message: Christian Holm Christensen: "Re: Clustering"

    Hi Brahmsers,
    
    Some comments on clustering. None of the stuff has been commited and
    is in a test state here in Copenhagen.
    The testing has been done exclusively on TPM1.
    
    *
    Djam and I found that if there are no clusters the BrTPCClusterFinder
    code crashes. This is because the TList in BrTPCClusterTable is not
    created and so the code crashes in RemoveOnePadClusters().
    The solution implemented was to always create a TList. It might be empty. 
    This could be related to the problems with BrTPCPreProcess that we
    experienced this summer and I beleive that it should now work
    properly. If the BrTPCPreProcess removes all sequences then the code
    would crash.
    
    *
    The cut on max pixel was only implemented for deconvoluted
    clusters. Here I took the liberty of adding a whole lot more
    functionality to BrTPCCluster, so it now can calculate nPads, nBins,
    first pad, first time, and ADC max. I think that BrTPCCluster should
    be one of the keystones of BrTPCClusterFinder and will continue in
    that direction. This removes a lot of small "noise" clusters for TPM1.
    Effect : fewer clusters => more tracks in some events. Cleaner TPM1.
    TPM1 is not so saturated as before. I don't have a numeric value of
    the effect, sorry.
    
    *
    I found that many of the TPC sequences contains partly 0 and negative
    values. Adding a piece of code to BrTPCPreProcess that strips the
    sequences of these in the following way. (Default demand for a
    BrTPCSequence was set to 3 non zero ADC values - easy to change.
    
    ( -1, 0, 10, 12, 8, 6, 0, 5, 3, 0, 10, 20, 10, 5 )
    ==>
    ( 10, 12, 8, 6 )     &&  ( 10, 20, 10, 5 )
    ( 5, 3 ) is not kept since it is only 2 long.
    Effect : Smaller clusters.
    I have added the TPCPreProcess to the track follow finder and it works.
    
    *
    The problem with the with dpad == 0 and dtime == 0, seems to be
    related to a hole in the deconvolution. I can draw the clusters and they
    are clearly only 1 pad or 1 time bin clusters that should be removed. I am
    looking in the code.
    
    *
    Started looking at some good cuts for removing noise clusters and this
    involved being able to get to clusters that are part of tracks and
    clusters that are hits, so I would like to add a TObject pointer to
    BrDetectorHit that points to the BrTPCHitCluster and a BrTPCCluster
    pointer to BrTPCHitCluster that points to the cluster. In that way at
    some point also <dEdx> can easily be put in the track. Both pointers
    have the //! so that they are not streamed.
    
    I am open for good suggestion and critics.
    I will try to check in my changes and a testprogram for some of them
    before the weekend.  
    
    I will continue to work on the clustering, so if anybody else is
    working on that (Jens Ivar, perhaps ?), please contact me.
    
    Cheers,
       Peter 
    
    :-) --------------------------------- )-:
    |Peter H L Christiansen aka PAN @ NBI	|
    |EMAIL  : pchristi@nbi.dk		|
    |OFFICE : Tb1 @ NBI			|
    |PHONE  : 353 25269 <- New!!		|
    |SNAIL  : Sdr. Fasanvej 14 ST 2000 F	|
    |PHONE  : 38 872042 			|
    :-D --------------------------------- \-:
    



    This archive was generated by hypermail 2b29 : Mon Sep 18 2000 - 11:26:29 EDT