|
wolffd@0
|
1 if 1
|
|
wolffd@0
|
2 O = 4;
|
|
wolffd@0
|
3 T = 10;
|
|
wolffd@0
|
4 nex = 50;
|
|
wolffd@0
|
5 M = 2;
|
|
wolffd@0
|
6 Q = 3;
|
|
wolffd@0
|
7 else
|
|
wolffd@0
|
8 O = 8; %Number of coefficients in a vector
|
|
wolffd@0
|
9 T = 420; %Number of vectors in a sequence
|
|
wolffd@0
|
10 nex = 1; %Number of sequences
|
|
wolffd@0
|
11 M = 1; %Number of mixtures
|
|
wolffd@0
|
12 Q = 6; %Number of states
|
|
wolffd@0
|
13 end
|
|
wolffd@0
|
14 cov_type = 'full';
|
|
wolffd@0
|
15
|
|
wolffd@0
|
16 data = randn(O,T,nex);
|
|
wolffd@0
|
17
|
|
wolffd@0
|
18 % initial guess of parameters
|
|
wolffd@0
|
19 prior0 = normalise(rand(Q,1));
|
|
wolffd@0
|
20 transmat0 = mk_stochastic(rand(Q,Q));
|
|
wolffd@0
|
21
|
|
wolffd@0
|
22 if 0
|
|
wolffd@0
|
23 Sigma0 = repmat(eye(O), [1 1 Q M]);
|
|
wolffd@0
|
24 % Initialize each mean to a random data point
|
|
wolffd@0
|
25 indices = randperm(T*nex);
|
|
wolffd@0
|
26 mu0 = reshape(data(:,indices(1:(Q*M))), [O Q M]);
|
|
wolffd@0
|
27 mixmat0 = mk_stochastic(rand(Q,M));
|
|
wolffd@0
|
28 else
|
|
wolffd@0
|
29 [mu0, Sigma0] = mixgauss_init(Q*M, data, cov_type);
|
|
wolffd@0
|
30 mu0 = reshape(mu0, [O Q M]);
|
|
wolffd@0
|
31 Sigma0 = reshape(Sigma0, [O O Q M]);
|
|
wolffd@0
|
32 mixmat0 = mk_stochastic(rand(Q,M));
|
|
wolffd@0
|
33 end
|
|
wolffd@0
|
34
|
|
wolffd@0
|
35 [LL, prior1, transmat1, mu1, Sigma1, mixmat1] = ...
|
|
wolffd@0
|
36 mhmm_em(data, prior0, transmat0, mu0, Sigma0, mixmat0, 'max_iter', 5);
|
|
wolffd@0
|
37
|
|
wolffd@0
|
38
|
|
wolffd@0
|
39 loglik = mhmm_logprob(data, prior1, transmat1, mu1, Sigma1, mixmat1);
|
|
wolffd@0
|
40
|
