Mainz 2014 – wissenschaftliches Programm
T 37.7: Vortrag
Montag, 24. März 2014, 18:15–18:30, P103
Bayesian analysis toolkit — •Frederik Beaujean1, Allen Caldwell3, Daniel Greenwald2, Daniel Kollár3, and Kevin Kröninger4 — 1Excellenzcluster Universe, Ludwig-Maximilians-Universität München — 2Technische Universität München — 3Max Planck Institut für Physik, München — 4Georg-August-Universität, Göttingen
BAT, the Bayesian Analysis Toolkit (http://mpp.mpg.de/bat), is a software package developed in C++ designed to facilitate data analyses employing Bayes' theorem. The central task of drawing parameter samples from the posterior probability is accomplished with Markov chain Monte Carlo, and the output can be used for parameter estimation, limit setting, and uncertainty propagation. Additional algorithms, such as simulated annealing, allow extraction of the global mode of the posterior.
The only inputs required to start an analysis are the likelihood and the prior in the form of C++ code. BAT assists the user in offering a selection of widely used models common to high-energy physics problems. The package is interfaced to other software packages commonly used in high energy physics, such as ROOT, Minuit, RooStats and CUBA.
We present an overview of BAT, highlight example use cases, and discuss new features of the latest releases, including support for automatic parallelization. Last, we sketch improvements planned for a future massively parallel version of BAT.