ApplyingtheHiddenMarkovModelMethodologyfor培训课件.pptVIP

Bayesian Model Selection for HMM Model Size Selection Goal: To select the optimal number of states, K, for a HMM that maximizes S S S 1 2 K BIC and CS for HMM Model Size Selection (b) CS Approximation likelihood (a) BIC approximation CS prior penalty likelihood BIC 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10 HMM Model Size Selection Process Sequential search for HMM model size with a heuristic stopping criterion -- Bayesian model selection criterion approximated by BIC and CS Issue 2: State initialization by K-means clustering