Bayes Net Interchange Format Working Group
To establish agreed on interchange format for Bayesian Networks.
General comments: While I hope that we can eventually agree on a comprehensive
format for all aspects of knowledge representation (!), I believe we will make
most progress by keeping our initial aims modest. A simple format for
"classical" Bayesian nets is, I believe, a good place to start. The quicker
we get agreement on a base format, the quicker we can move on to interesting
issues like continuous variables, decision models, local structure, query
results, chain graphs, templates, etc etc etc.
Comments on the procedure always welcome!
- Study initial Microsoft Proposal
Microsoft Proposal - Note - this URL is somewhat out of date. Microsoft
promises a revised page shortly.)
- Some initial issues have been raised by Fabio Cozman of
CMU, as a result of attempting to construct a Java parser which conforms to the
(revised and unpublished) microsoft proposal. As a result,
slightly simplified format is at
CMU modified format. See changes for a
short list of differences between this format and the Microsoft format. Most
changes are to keep the parser small, since Java parsers get shipped over the
net when used in applets.
This site includes a complete parser in Java, a test program that reads nets
and confirms conformation to the format, and graphical editors which can read
and write the format.
- Raises issues with the proposals, suggest alternatives for resolving the
issues, and discuss.
I will digest discussion and generate web pages for the issues raised.
Upon timeout or quiesence of discussion, vote on alternatives for each issue.
- that creates V1.0, we then iterate as needed.
Bruce D'Ambrosio Home Page / OSU / firstname.lastname@example.org