You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Do analytical Barlow Beeston before likelihood evaluation
Doing the analytical Barlow Beeston before evaluating the likelihood has
two big advantages, instead of doing it while evaluating the
CMSHistSum/CMSHistErrorPropagator:
1. We don't have to hack the dirty state propagation. This was
necessary before, because it is not allowed to change a RooRealVar
server during the evaluation of the client. Also, the hack relies on
the implementation details of the CachingSimNLL (evaluating main pdf
before constraints), so it's also a problem for using RooFit regular
NLL code path.
2. If the analytical minimization and the likelihood evaluation
factorize, we don't need to worry how to differentiate through the
analytical minimization and the likelihood evaluation together.
0 commit comments