[an error occurred while processing this directive]

ICS-175A, Bayesian and Constraint Networks, Spring 2003
overview | announcements | home work | lecture notes

Requirements for the Bayesian Modeling Project

The project has 3 requirements:

  1. Building the Bayesian network and implementing in one of the tools
  2. Verifying andevaluating the generated Bayesian networks
  3. Experimenting and analyzing the Bayesian networks.

  1. Building the network: requires the generation of the graph (from an expert, or other source) and generating the tables (from an expert or from data). You should determine which variables are observed normally, which are hidden and which are likely to be queried.
  2. Verifying the network includes first verifying all the tests of conditional independencies: arguing convincingly that each parents makes a child node independent of the rest of the ancestors and making sure that the parent set is minimal. Then showing and justifying how the CPTs were created (from data? from expert? ) Basic evaluation include running queries and showing that the answers are reasonable and making sure that it all work properly.
  3. Network analysis and addition to Benchmark:
    Make a set of experiments with your network and analyze performance. First determine its tree-width. Second run various algorithms and compare their run time. report the results. This portion is best done with REES. So you may need to compile your network so it is recognized by REES. You report the results using REES.

    You make your network available on REES benchmarks.