Important note: I'm NOT proposing to do this now, or in a short timescale. It is just a random thought that would be interesting to do in the future, probably after publishing the paper we have discussed several times.
Something someone could do in the future is to take a look at the parameters within corsika of the showers labeled with different event types. That would allow us to understand which are the kind of showers providing a better performance and which are the showers providing a worse performance.
Example: In La Palma, with a high geo-magnetic field in certain pointing directions, those events in which the first interaction occurs high up in the atmosphere get heavily affected by the GMF, and therefore the reconstruction will be terrible. Those events will be very easily identified by the ML and labeled as "bad events".
Finding the different populations of events providing good/bad performance would give information to the simulation and low-level analysis teams to either improve the analysis of those events, or help us identify them farther through new parameters.
Important note: I'm NOT proposing to do this now, or in a short timescale. It is just a random thought that would be interesting to do in the future, probably after publishing the paper we have discussed several times.
Something someone could do in the future is to take a look at the parameters within corsika of the showers labeled with different event types. That would allow us to understand which are the kind of showers providing a better performance and which are the showers providing a worse performance.
Example: In La Palma, with a high geo-magnetic field in certain pointing directions, those events in which the first interaction occurs high up in the atmosphere get heavily affected by the GMF, and therefore the reconstruction will be terrible. Those events will be very easily identified by the ML and labeled as "bad events".
Finding the different populations of events providing good/bad performance would give information to the simulation and low-level analysis teams to either improve the analysis of those events, or help us identify them farther through new parameters.