Bayesian Network Module Specification
for OpenaAI Project (a first draft)
Abstract
This document is a draft for a specification of a Bayesian Network Module for the OpenAI project. This document was written with the intention of providing a beginning for an evolution to a more precise specification.
Status of this document
This document is a first draft for review by OpenAI members and other interested parties. All the comments, critiques (constructive) and proposals are welcome.
Proposals Specifications
The Bayesian Network module must execute the described requisites in OpenAI Specifications Level 1.
In order to improve code re-using and to provide familiarity to future programmers the name and structure of the packages and class should be so similar as possible to the Neural Net packages.
The module have to implements:
-
Creation of new Bayesian Networks and possibility of load and save it from specific XML formats (initially XBN format or XMLBIF format).
-
Easy modifications in the networks and facility for save the changes.
-
Inference in the networks from concrete evidences.
-
Facility for to add new inference algorithms.
-
Training network facilities in different cases ( known network structure, unknown network structure, all the CPT values known, unknown CPT values...)
-
Facility for integration with another OpenaAI modules (GA, NN,..) for training the network.
-
Easy extension of new training methods and algoritms.
A conceptual differentiation in submodules (network structure, inference engine, learning facilities.. ¿?) should exist.
Last modified: 2002/04/11
Versión en español