wolffd@0: O = 3;
wolffd@0: Q = 2;
wolffd@0: 
wolffd@0: % "true" parameters
wolffd@0: prior0 = normalise(rand(Q,1));
wolffd@0: transmat0 = mk_stochastic(rand(Q,Q));
wolffd@0: obsmat0 = mk_stochastic(rand(Q,O));
wolffd@0: 
wolffd@0: % training data
wolffd@0: T = 5;
wolffd@0: nex = 10;
wolffd@0: data = dhmm_sample(prior0, transmat0, obsmat0, T, nex);
wolffd@0: 
wolffd@0: % initial guess of parameters
wolffd@0: prior1 = normalise(rand(Q,1));
wolffd@0: transmat1 = mk_stochastic(rand(Q,Q));
wolffd@0: obsmat1 = mk_stochastic(rand(Q,O));
wolffd@0: 
wolffd@0: % improve guess of parameters using EM
wolffd@0: [LL, prior2, transmat2, obsmat2] = dhmm_em(data, prior1, transmat1, obsmat1, 'max_iter', 5);
wolffd@0: LL
wolffd@0: 
wolffd@0: % use model to compute log likelihood
wolffd@0: loglik = dhmm_logprob(data, prior2, transmat2, obsmat2)
wolffd@0: % log lik is slightly different than LL(end), since it is computed after the final M step