search

SNEWS2 / snewpdag

SNEWS2 alert calculations
BSD 3-Clause "New" or "Revised" License
2 stars 13 forks source link

top-down time series comparison #87

Closed tsengj10 closed 1 year ago

tsengj10 commented 1 year ago

This PR implements a simplified version of the top-down algorithm, which at this stage only considers signal-only neutrino lightcurves as time series, i.e., not really useful except as a proof of concept.

Some changes were also included which allow configuration files to set times with strings rather than numeric timestamps.

The csv files can be run in the usual way. topdown.csv is for individual calculations. topdown-series.csv is set up to run the calculation many times, with randomly generated time series based on a single true supernova direction.

tsengj10 commented 1 year ago

The log-likelihood function has been modified in TopDownSeries so that the maximum Poisson likelihood for any detector/time bin is normalized to 1, i.e., its contribution to log(L) is 0. This is like the chi2 figure of merit, which does the same for a Gaussian likelihood. The confidence regions look a bit more contiguous after the change, and don't appear to be as much affected by including the low-occupancy bins at late times (in other words, including 10s of a time series doesn't look very different from 1s).