From lefebvre@uvic.ca Mon Apr 26 18:55:01 2004
Date: Mon, 26 Apr 2004 18:35:33 -0700 (PDT)
From: Michel Lefebvre <lefebvre@uvic.ca>
To: atlas Victoria group - hec-emec analysis -- Ashok Agarwal
    <numod@uvvm.uvic.ca>, Alan Astbury <astbury@uvic.ca>,
     Ian Gable <igable@uvic.ca>, Margret Fincke-Keeler <mgf@uvic.ca>,
     Michel Lefebvre <lefebvre@uvic.ca>, Paul Poffenberger <pof@uvic.ca>,
     Randy Sobie <rsobie@uvic.ca>, Richard Keeler <rkeeler@uvic.ca>,
     Rob McPherson <Robert.McPherson@cern.ch>, Warren Shaw <shaw@uvic.ca>,
     Tamara Hughes <tmhughes@uvic.ca>, Tayfun Ince <incet@uvic.ca>,
     Monika Wielers <Monika.Wielers@cern.ch>
Subject: clustering algorithm in TBRootAna

Greetings,
I have recently implemented the ATLAS Athena LAr clustering algorithm in 
TBRootAna.  Discussions with Alan and Sven Menke led Sven to refine the 
Athena implementation (yesterday).  ClusteringAlg will be particularly 
useful to Tamara's work, but could be useful to Tayfun's also.

I have put some information on the TBRootAna example web page
http://particle.phys.uvic.ca/~web-atlas/atlas/hec-emec/tbrootana/examples/examples.html
which can also be accessed from the TBRootAna page
http://particle.phys.uvic.ca/~web-atlas/atlas/hec-emec/tbrootana/

I have committed the ClusteringAlg systemAlg to the repository, as well as 
the CanvasManagerAlg.  The latter greatly helps managing your canvases in 
your UserAlg, and basically encapsulates all the required goodies to store 
your histos in a postscript file as well.  The example on the TBRootAna 
web site is a rather complete analysis using both CanvasManagerAlg and 
ClusteringAlg.

As you can read from the example web site, clustering is based on three 
thresholds (separately for hec and emec):
seed threshold (applied to the energy)
neighbor threshold (applied to the absolute value of the energy)
cell threshold (applied to the absolute value of the energy)
The result (separately for the hec and emec) is a cluster ID keyed map of 
clusters (vector<int> with the febnos) which is very easy to use in the 
TBRootAna framework.

The original Athena implementation did not order the seed cells in any
way.  I pointed out to Sven that this means that the final makeup of the
clusters (more precisely their cells above cell threshold) depends on how
the seed cell list is traversed at each iteration of the algorithm.  Sven
and I agreed that this was not very satisfactory, but that it may not
produce large effects, since it does not affect in which cluster the cells
above seed threshold and the cells above neighbor threshold end up.

A first refinement, first mentioned by Alan, consists in ordering the seed
cell list in order of energy.  To reduce the cpu requirement, Sven is 
proposing to only order the initial seed cell list (first iteration) in 
order of signal/noise.

Since cells are not of the same size, this would give an unfair advantage
to large cells.  The TBRootAna implementation orders the seed cell list in
descending order of energy density, at each iteration.  Hence seed cells
with higher energy density have first grab at the cells above cell
threshold.

Please contact me if you have any questions.

Cheers
Michel