--
JaimeNorman - 2014-11-11
Introduction
The pt distribution of the decay products, in the decay Lc -> pKpi, is not a perfect description of the real data. Using MC for the background sample used to train the MVA methods could thus be biasing the MVA weights, leading to a non-optimal application. Using data as the background sample will describe the true background much better, and could thus be used as an alternative.
Integrated pT
A small random sample of data is removed from the data, of about 55000 Lc candidates, in order to have an independent background sample. 40,000 candidates are then taken from the sidebands of the invariant mass spectra, i.e. with an invariant mass outside the Lc peak, which make up the background. In this way, 99.6% of Lc candidates are still used in the application phase.
Above - invariant mass of the background taken from data
The pt distribution of the decay products (pion, kaon, proton) in data now follows the background pt distribution well.
The BDT response for data well matches that for background
Application - invariant mass spectra
Application of BDT weights, making tighter cuts, from 0 to 0.06, on the BDTA output distribution
Application of BDT weights, making tighter cuts, from -0.5 to 0.1, on the BDTA output distribution
It can be seen that progressively tighter cuts does not improve the significance, which remains constant.
BDT w/ adaptive boost response vs pt, for background (classID==1) and signal (classID==0) from the test sample
BDT w/ gradient boost response vs pt, for background (classID==1) and signal (classID==0) from the test sample
Responses vs pt in data.