Mercurial > hg > camir-aes2014
annotate toolboxes/FullBNT-1.0.7/HMM/dhmm_em_demo.m @ 0:e9a9cd732c1e tip
first hg version after svn
| author | wolffd |
|---|---|
| date | Tue, 10 Feb 2015 15:05:51 +0000 |
| parents | |
| children |
| rev | line source |
|---|---|
| wolffd@0 | 1 O = 3; |
| wolffd@0 | 2 Q = 2; |
| wolffd@0 | 3 |
| wolffd@0 | 4 % "true" parameters |
| wolffd@0 | 5 prior0 = normalise(rand(Q,1)); |
| wolffd@0 | 6 transmat0 = mk_stochastic(rand(Q,Q)); |
| wolffd@0 | 7 obsmat0 = mk_stochastic(rand(Q,O)); |
| wolffd@0 | 8 |
| wolffd@0 | 9 % training data |
| wolffd@0 | 10 T = 5; |
| wolffd@0 | 11 nex = 10; |
| wolffd@0 | 12 data = dhmm_sample(prior0, transmat0, obsmat0, T, nex); |
| wolffd@0 | 13 |
| wolffd@0 | 14 % initial guess of parameters |
| wolffd@0 | 15 prior1 = normalise(rand(Q,1)); |
| wolffd@0 | 16 transmat1 = mk_stochastic(rand(Q,Q)); |
| wolffd@0 | 17 obsmat1 = mk_stochastic(rand(Q,O)); |
| wolffd@0 | 18 |
| wolffd@0 | 19 % improve guess of parameters using EM |
| wolffd@0 | 20 [LL, prior2, transmat2, obsmat2] = dhmm_em(data, prior1, transmat1, obsmat1, 'max_iter', 5); |
| wolffd@0 | 21 LL |
| wolffd@0 | 22 |
| wolffd@0 | 23 % use model to compute log likelihood |
| wolffd@0 | 24 loglik = dhmm_logprob(data, prior2, transmat2, obsmat2) |
| wolffd@0 | 25 % log lik is slightly different than LL(end), since it is computed after the final M step |