I realized I needed a better way of testing the max likelihood code. So I put together a shower file of Chicos showers from 2006/2007 (actually two files, one with all data, and one with accidentals removed). Running this through the autoFit program makes it easy to test new versions of the code on a large number of showers at once. I did run into the problem that there are often many unhealthy sites. (The autoFit program did not need to account for this before because the chiquita data was only taken during times when all sites were up.) I added an option to the Rext() function in CTShowerUI.C that will remove unhealthy sites automatically. Right now it even removes unhealthy sites that were hit as part of the shower - this could be tinkered with later, but since the purpose right now is just to test the relative accuracy of different code versions, I didn't worry about it.
CTShower Code (May not be most recent.)For comparison purposes, here are the shower files for versions 2.42 and 3.01.
Possible tweaks to v3.01 that have been tested:
Number of showers that failed to reconstruct (out of 55 total):
v. 2.42: None
v. 3.01: 17 (all data), 16 (no accidentals)
v. 3.03a: 10 (all data), 5 (no accidentals)
v. 3.03b: 2 (all data), 1 (no accidentals)
Some plot of the energy distributions:
Energy distribution, all data:
Energy distribution, no accidentals:
Change in reconstructed energy when accidentals are removed.
Ideally this should be a small effect for the max likelihood methods.
Change in reconstructed energy between different versions of code. I have compared 301 to 242, and 303a/b to 301.
For completeness, here is the comparison between 303a and 303b.
Any differences reflect the ignoring of large hits.
