GUNDAM, for Generalized and Unified Neutrino Data Analysis Methods, is a suite of applications which aims at performing various statistical analysis with different purposes and setups.
GNU Lesser General Public License v2.1
13
stars
10
forks
source link
LTS 1.8.x and main : Control over fit starting point #536
The fit starting point can be changed by a few options from the command line, but, by default, will be the starting point is taken as the prior value. This has the effect that any Asimov study will start at the maximum likelihood, and not be representative of the actual fit.
This is a proposal for two independent things:
Change the --kick-mc option to accept the size of the kick as an argument (default when triggered is currently 0.1)
Apply the kick by default. If no kick is wanted, then use --kick-mc 0.0
The effect will be that the MINOS minimization will take more likelihood calls to converge, but during an Asimov run, almost all of the calls are spent calculating the Hessian. This will play nicely with the MCMC which already controls the starting point as part of the burn-in.
ClarkMcGrew
commented
2 weeks ago
The fit starting point can be changed by a few options from the command line, but, by default, will be the starting point is taken as the prior value. This has the effect that any Asimov study will start at the maximum likelihood, and not be representative of the actual fit.
This is a proposal for two independent things:
--kick-mcoption to accept the size of the kick as an argument (default when triggered is currently 0.1)--kick-mc 0.0The effect will be that the MINOS minimization will take more likelihood calls to converge, but during an Asimov run, almost all of the calls are spent calculating the Hessian. This will play nicely with the MCMC which already controls the starting point as part of the burn-in.