Class TrainBaumWelch

  • All Implemented Interfaces:

    public class TrainBaumWelchextends BaseBaumWelch
    Baum Welch Learning allows a HMM to be constructed from a series of sequence observations. This implementation of Baum Welch does not scale and is susceptible to underflows in long sequences of data. Baum Welch requires a starting point. You should create a HMM that has a reasonable guess as to the observation and transition probabilities. If you can make no such guess, you should consider using KMeans training. L. E. Baum, T. Petrie, G. Soules, and N. Weiss, "A maximization technique occurring in the statistical analysis of probabilistic functions of Markov chains" , Ann. Math. Statist., vol. 41, no. 1, pp. 164-171, 1970. Hidden Markov Models and the Baum-Welch Algorithm, IEEE Information Theory Society Newsletter, Dec. 2003.

SCaVis 1.8 ©