readme

是实现关系型贝叶斯网络一中机器学习算法

  This code supplies a Java implementation for Relational Markov Networks with discrete potentials. It is being released for educational and research purposes only under the GNU General Public License (see http://www.gnu.org/copyleft/gpl.html).

  Useful references:
  - For an introduction to RMNs and their application to collective classification of web-pages, see "B. Taskar, P. Abbeel, and D. Koller - Discriminative probabilistic models for relational data (UAI 2002)".
  - For an application of RMNs to collective named entity recognition, see "R. Bunescu and R. Mooney - Collective Information Extraction with Relational Markov Networks -  http://www.cs.utexas.edu/users/ml/papers/cie-submitted-04.pdf".
  - The RMNs representation is based on Factor Graphs. An excellent tutorial on Factor Graphs is "Factor Graphs and the Sum-Product Algorithm - Kschischang, Frey, Loeliger (1998)".
  - Inference is based on the sum/max-product algorithm in factor graphs (see tutorial above).
  - Learning is based on Collins' voted perceptron from "Ranking Algorithms for Named-Entity Extraction: Boosting and the Voted Perceptron (2002)".

  Observations:
  - To compile the package: 
    > javac -source 1.4 *.java
  - To test the package on a sample RMN: 
    > java -classpath ../ rmn.FactorGraph
    (the inference on this sample RMN should give the same results as, for example, Kevin Murphy's Bayes Net Toolbox for Matlab code)

  

  Copyleft: Razvan C. Bunescu, 2004
            [e-mail]