• HOME
• Research
• Teaching
• Software
• Personal
• SMfSB

• SBML-shorthand
• GDAGsim
• BayesLin

Software / BayesLin

BAYES-LIN

An object-oriented environment for Bayes linear local computation

BAYES-LIN is intended for people who already know about Bayes linear methods, graphical modelling and local computation, and want a collection of object-oriented programming tools for carrying out computations. Local computation in BAYES-LIN is achieved via message-passing between objects representing clique-tree nodes. If you are relatively new to Bayes linear methods, then I would suggest trying out [B/D] before experimenting with this package.

BAYES-LIN is a set of modules for the XLISP-STAT statistical programming environment, and hence assumes some familiarity with that system, and the basic concepts of object-oriented programming. It also pre-supposes that you have a working LISP-STAT system already installed.

• An outline of future developments is also available (please read this!).

Code and Documentation:

BAYES-LIN 0.1

• Code (13/10/97)
• Documentation:
• Gzipped PDF (13/10/97)

BAYES-LIN 0.1 uses an obsolete syntax, and is not very O-O - it is mainly available for historical reasons. However, it is the only publicly available version supporting graphics windows, so if you are interested in diagnostic graphics, this is the version you need to download.

BAYES-LIN 0.3

• Code (7/4/2000)
• Documentation (7/4/2000)

BAYES-LIN 0.3 is properly O-O, and uses DAG nodes as the basis for model specification. The software now automatically constructs an appropriate junction tree for computation. The propagation algorithm currently used is essentially Lauritzen's CG propagation algorithm for the pure Gaussian case.

Other documentation

• The underlying theory is presented in my paper with Micheal Goldstein, Bayes linear analysis for graphical models: The geometric approach to local computation and interpretive - precise details from my list of publications.
• Poster presentation on Bayes linear local computation and BAYES-LIN, available as PostScript.
• I also have a paper in the COMP-STAT'98 proceedings (Physica Verlag) describing the object-oriented approach to Bayes linear local computation used in BAYES-LIN. The slides for the talk presented at COMP-STAT are available as PostScript

The code was developed under Linux, but should run on any platform with a fully functional LISP-STAT system, including MS Windows PCs, Macs, and any X-Windows UNIX machine.

BAYES-LIN is Copyrighted, but open-source, and free to all.

BAYES-LIN is NOT a friendly general purpose package for Bayesian inference! If you are seeking such a thing for free on the WWW, try BUGS (MCMC based). If you are interested in graphical belief modelling software generally, try consulting Russell Almond's list of Software for Manipulating Belief Networks.

Copyright 1997-2000 Darren Wilkinson