MCMC samples selecting for online bayesian network structure learning
Shaozhong Zhang, Lu Liu · 2008
This paper presents an online learning algorithm for Bayesian network structure, which adopts Important Sampling method of Markov Chain Monte Carlo for online samples evaluation and proper model structure selecting combined with probability distribution of a former. It selects a set of optimized samples for online learning and adjusting based on an existing reliable model structure. And then it learns and adjusts structure online using an important samples set. At last it evaluates the obtained structure by model evaluation and select a reliable one as a new structure. The algorithm proposed in this paper reduces the calculating loads by important samples instead of all samples and implements structure learning online. The experiment shows that the algorithm in this paper can achieve online structure learning and it also has a preferable precision and convergence rapidly.