Description
Description of the issue from
pisa/pisa/stages/discr_sys/ultrasurfaces.py
Line 159 in 6015551
At least in theory, this should always pick out the one exact event from the correct event category. If an event is not exactly matched, however, it's possible that the gradient for a numu_cc event might get picked from a nu_nc event, for example. We don't have any safeguards against that at this time, even though the information is in the DataFrame to do it.
Thich supposedly becomes particularly relevant when an event is encountered that wasn't part of the original training/classification problem.