Learning Multi-order Transition Networks Method Using Nodes Ranking and Local Search and Scoring
Hui Wang · Journal of Yantai University · 2010
At present, learning dynamic Bayesian network are mainly used in building prior network and one-order transition network. Problems such as inefficiency and unreliability exist in learning transition network with multi variables and complex structure. In this paper,a method of learning multi-order transition network by means of vector time series data is presented. First the method of setting up multi-order data set is given. The perfect directed acyclic graph is built through calculating the conditional relative average entropy. Then a transition network can be obtained by nodes ranking based on perfect directed acyclic graph and local search and scoring. The method will be more efficient, reliable and practical.