Hi all-
I have been running the code to reproduce Urs' BRBR result for the prl
with both anaQA and the new VubAnalysisApp and I noticed that there was
a difference in the IERR printout at the end of the job for most of the
chains, indicating something different was happening with the abc fit.
Debugging it led back to the splitOff study. The couts i put in
doSplitOffStudy showed some strange numbers for ecalTrk and phiAtEMCTrk
including numerous non-numbers (nan). Long story short...
anaQA/VubAnalysisApp runs with these:
< a.switchOffReading("ecalTrk");
< a.switchOffReading("phiAtEMCTrk");
But, ecalTrk and phiAtEMCTrk are used by doSplitOffStudy to compute
splitOffGam. splitOffGam is then used in selectPhotons to determine
goodPhoton so we can't switch off their reading, surely.
if (PHOTONSELECTION == 8) {
if (superric && acc && splitOffGam[i]==0 )
goodPhoton[i] = 1;
}
So, i took out the switchOffReading statements and now i don't see the
descrepancy between anaQA and VubAnalysisApp - and more importantly my
debug statements show reasonable values.
Urs said he'll run the whole thing again with these lines taken out and
we can see what the effect on the BRBR is (it'll likely be very small).
I'll continue looking tommorow to see exactly how often splitOffGam
gets improperly set to 1 and how much of a difference this makes on the
abc fit results, etc...
I didn't see this in my validation b/c i ran over a shortened vub-mix
chain where this effect is invisible.
Cheers,
Ed
|