Title: UW classification: background rejection
1UW classificationbackground rejection
- (An update more details at future meetings)
2Review, tree status
- Training samples
- 5M allgamma from v7r0p2 (SLAC farm)
- DC2 background 4 of a day from v7r0p2 (UW
farm) - Packages classifier and GlastClassify perform
the training, save the resulting decision trees
as data in GlastClassify - Current GR v7r1 tag has all the above trees.
- Package merit applies above trees to other
tuple variables to define the CT tuple variables
CTgoodCal, CTvertex, CTgoodPsf, CTgamma
3Bills DC1 Background Analysis
Pr(GAM) cut Acceptance(m2 sr) Diffuse rate (Hz) Background Background
Pr(GAM) cut Acceptance(m2 sr) Diffuse rate (Hz) events rate (Hz)
0.50 2.40 Â 344 0.070
0.90 2.37 0.36 138 0.028
0.95 2.07 0.31 31 0.01
- Issues
- What is the residual rate? Can one satisfy the SR
signal/noise of 1051 and the Aeff requirement of
0.8 m2 ? - How does it compare with the extragalactic
diffuse rate? - Can we identify characteristics of the remaining
events allowing better rejection? - The background rate is now X2 more than for DC1.
Â
4My requirementBe able to measure the spectrum
of the extragalactic diffuse to from 30 Mev to
gt100 Gev.
- Science issues
- is there an excess above 1 GeV that can only be
explained by WIMP annihilation? - To what energy are we really sensitive? When does
background overwhelm the extragalactic diffuse? - Note I use the EGRET measurement as a reference
it is surely an overestimate. - We need to understand the background energy
distribution! - The following is based on 0.01 day DC2
background generated by v7r1 at UW, and the
Extragalactic_diffuse source
5Current, very preliminary status
CTgamma cut 0.9 0.96 0.99
6Current estimate current background dominates
above 10 GeV
- So what are the characteristics of those events?
Bill noticed this clue
one suggestion was that the MIP finding might
help separate the residual background.
7I confirm Bills observation
- Clearly see downward and upward events
0.90
0.96
0.99
McZdir
Tkr1Zdir
8So, what about the MIP finder?
- First attempt to use it in the classification
tree was not a success. But this is a small
fraction of the original population. Will look at
it for these remaining events. - First, for all background
- CalMipZDir shows the fraction for which there is
a measurment - And for these, there is clear correlation with
the actual incoming direction, both up- and
down-going
9And now, the residual
McZdir
- Preliminary result disappointing no apparent
signal.
Tkr1Zdir
10Next steps
- Import Bills new cal tree definitions
- Use v7r1 for training background better ACD
should help - 0.04 day already done at UW, we can do 0.01 day
in a day. - Try MIP finder again, and review (with Bill, all
the variables).