Closed AndrewLevin closed 5 years ago
One possible solution would be to only fit the region THR_DAC < 100, which would reduce the number of wiggles that the polynomial must model. From Marek's plots
https://mattermost.web.cern.ch/api/v4/files/izr38nebhbbbfpj7r86msaj8ao?download=1
it seems we are always trimming at THR_DAC < 60, so probably we do not need to fit the region THR_DAC > 100.
One possible solution would be to only fit the region THR_DAC < 100, which would reduce the number of wiggles that the polynomial must model. From Marek's plots
https://mattermost.web.cern.ch/api/v4/files/izr38nebhbbbfpj7r86msaj8ao?download=1
it seems we are always trimming at THR_DAC < 60, so probably we do not need to fit the region THR_DAC > 100.
Did you get the data and try to apply this proposal to the plot above? Any updates?
potentially addressed by #239; will re-open should this persist afterward.
Poorly fitted scurveMean vs thrDac data found by the QC8 team indicate further changes to the fit algorithm are necessary.
Brief summary of issue
In slides 7 and 16 of https://indico.cern.ch/event/822091/contributions/3436929/attachments/1847158/3031036/fsimone_QC8PFA_20052019.pdf multiple examples can been seen where the fit is not properly modeling the non-linear behavior at low thrDac.
Types of issue
Expected Behavior
Fit should be very sensitive to low thrDac points. Even individual one-off points should pull the fit down.
Current Behavior
Fit often looks linear in the low thrDac region when the data does not.
Context (for feature requests)
Trimming is useless when the low thrDac region is modeled this badly.
Your Environment