LibBi might nominally stand for Library for Bayesian inference, although is not meant to be an abbreviation as such. Development of the software began in 2009, with a working title of just Bi (for Bayesian inference), which, at the time, was sufficiently generic for anything we might want to put in it. Lib was added closer to public release, when something more unique was required.
Whichever you prefer, and for as long as you like, until you settle on "Libby".
Development of LibBi began in 2009 under a CSIRO project that forms part of the Computational and Simulation Sciences platform. The aim of the project was to develop appropriate models and methodology for quantifying uncertainty in marine biogeochemical models. Work continues there and the project remains a major driver of the software’s development. Recognising potential interest in the broader scientific community, the software was released under an open source licence in June 2013.
The main developer is Lawrence Murray. Other suspects in the abovementioned project are John Parslow, Noel Cressie, Eddy Campbell, Nugzar Margvelashvili and Emlyn Jones (see this paper, preprint also on arXiv). Emlyn Jones in particular suffered through the earliest versions of the software. Pierre Jacob and Anthony Lee have both made significant contributions. Dan Pagendam has suffered through later versions of the software.
LibBi is licensed under the CSIRO Open Source Software License (GPL). This is the full text of the GPL version 2 with some additional provisions.
Please cite the following paper:
L. M. Murray, Bayesian state-space modelling on high-performance hardware using LibBi, 2013. [arXiv]
You can! But that may not be your best choice, depending on the problem you have at hand. LibBi differs from these packages in two ways:
The first point is reflected in the methods for inference that LibBi has
available. Its staple methods are from the family of sequential Monte Carlo
(SMC), not Gibbs (as in BUGS and JAGS) or Hamiltonian Monte Carlo (as in
Stan). LibBi can be used for non-SSMs by omitting the
initial blocks in a model specification, but its machinery for such models
is rudimentary at this stage. The potential is there to develop in such a
direction in future, however.
We are not aware of other software packages in this space that have the same high performance computing orientation as LibBi.