Importance sampling - Githubissues

ROOTPWA-Maintainers / ROOTPWA

ROOTPWA is a toolkit for partial-wave analysis of multi-particle final states produced in high-energy particle reactions. It is used to determine hadron spectra from experimental data.

Other

8 stars 14 forks source link

Importance sampling #176

Closed suhlatwork closed 8 years ago

suhlatwork commented 8 years ago

update of #139:

use a new method in fitResult class to get the intensity of the model, this adds supports for more than one rank
clear separation of generator stuff in importanceSampler class and the weight calculation in modelIntensity
use the new modelIntensity class also in the standard genPseudoData
in case the fit result contains the phase-space integrals those can be used, this changes the required command line arguments for genPseudoData and genPseudoDataImportanceSampler, integrals can still be overwritten with the new -i option
use the nBodyPhaseSpaceKinematics class to calculate phase-space weights from the importanceSampler
more detailed call statistic for functions called from BAT
fix if BAT was compiled with OpenMP
fix a couple of segmentation violations

suhlatwork commented 8 years ago

@fkrinner: maybe you have an idea on the following points

final state mass dependence is still not flat
the CalculateObservables method is called about 10 times more often than LogLikelihood, for my pi eta eta test case with 10000 events the prior is called 21809 times, the likelihood 20352, and the calculation of observables 184000.

suhlatwork commented 8 years ago

I reimplemented the possibility to set a mass prior. This could be used to create a flat final state mass distribution.

suhlatwork commented 8 years ago

Unfortunately it turned out that the simple way of enabling multi-rank fits was rather slow as it relies on fitResult internals which need to parse the reflectivities from wave names for each call to the spin-density matrix elements. So for the moment again a simplified calculation only correct for rank-1 fits is added, the multi-rank stuff should be added again once the fitResult is updated.

suhlatwork commented 8 years ago

Rebased to current master to get rid of merge conflicts.

suhlatwork commented 8 years ago

I hope I have answered to all the comments either by changing the code or by commenting.

suhlatwork commented 8 years ago

Improved the description of the two genPseudoData... scripts.

suhlatwork commented 8 years ago

Rebased to current master to fix a merge conflict.

suhlatwork commented 8 years ago

Another rebase to fix merge conflicts

suhlatwork commented 8 years ago

fixed the failing test

bgrube commented 8 years ago

Merged, thank you.