wolffd@0
|
1 function bnet = mk_hmm_bnet(T, Q, O, cts_obs, param_tying)
|
wolffd@0
|
2 % MK_HMM_BNET Make a (static) bnet to represent a hidden Markov model
|
wolffd@0
|
3 % bnet = mk_hmm_bnet(T, Q, O, cts_obs, param_tying)
|
wolffd@0
|
4 %
|
wolffd@0
|
5 % T = num time slices
|
wolffd@0
|
6 % Q = num hidden states
|
wolffd@0
|
7 % O = size of the observed node (num discrete values or length of vector)
|
wolffd@0
|
8 % cts_obs - 1 means the observed node is a continuous-valued vector, 0 means it's discrete
|
wolffd@0
|
9 % param_tying - 1 means we create 3 CPDs, 0 means we create 1 CPD per node
|
wolffd@0
|
10
|
wolffd@0
|
11 N = 2*T;
|
wolffd@0
|
12 dag = zeros(N);
|
wolffd@0
|
13 %hnodes = 1:2:2*T;
|
wolffd@0
|
14 hnodes = 1:T;
|
wolffd@0
|
15 for i=1:T-1
|
wolffd@0
|
16 dag(hnodes(i), hnodes(i+1))=1;
|
wolffd@0
|
17 end
|
wolffd@0
|
18 %onodes = 2:2:2*T;
|
wolffd@0
|
19 onodes = T+1:2*T;
|
wolffd@0
|
20 for i=1:T
|
wolffd@0
|
21 dag(hnodes(i), onodes(i)) = 1;
|
wolffd@0
|
22 end
|
wolffd@0
|
23
|
wolffd@0
|
24 if cts_obs
|
wolffd@0
|
25 dnodes = hnodes;
|
wolffd@0
|
26 else
|
wolffd@0
|
27 dnodes = 1:N;
|
wolffd@0
|
28 end
|
wolffd@0
|
29 ns = ones(1,N);
|
wolffd@0
|
30 ns(hnodes) = Q;
|
wolffd@0
|
31 ns(onodes) = O;
|
wolffd@0
|
32
|
wolffd@0
|
33 if param_tying
|
wolffd@0
|
34 H1class = 1; Hclass = 2; Oclass = 3;
|
wolffd@0
|
35 eclass = ones(1,N);
|
wolffd@0
|
36 eclass(hnodes(2:end)) = Hclass;
|
wolffd@0
|
37 eclass(hnodes(1)) = H1class;
|
wolffd@0
|
38 eclass(onodes) = Oclass;
|
wolffd@0
|
39 else
|
wolffd@0
|
40 eclass = 1:N;
|
wolffd@0
|
41 end
|
wolffd@0
|
42
|
wolffd@0
|
43 bnet = mk_bnet(dag, ns, 'observed', onodes, 'discrete', dnodes, 'equiv_class', eclass);
|
wolffd@0
|
44
|
wolffd@0
|
45 hnodes = mysetdiff(1:N, onodes);
|
wolffd@0
|
46 if ~param_tying
|
wolffd@0
|
47 for i=hnodes(:)'
|
wolffd@0
|
48 bnet.CPD{i} = tabular_CPD(bnet, i);
|
wolffd@0
|
49 end
|
wolffd@0
|
50 if cts_obs
|
wolffd@0
|
51 for i=onodes(:)'
|
wolffd@0
|
52 bnet.CPD{i} = gaussian_CPD(bnet, i);
|
wolffd@0
|
53 end
|
wolffd@0
|
54 else
|
wolffd@0
|
55 for i=onodes(:)'
|
wolffd@0
|
56 bnet.CPD{i} = tabular_CPD(bnet, i);
|
wolffd@0
|
57 end
|
wolffd@0
|
58 end
|
wolffd@0
|
59 else
|
wolffd@0
|
60 bnet.CPD{H1class} = tabular_CPD(bnet, hnodes(1)); % prior
|
wolffd@0
|
61 bnet.CPD{Hclass} = tabular_CPD(bnet, hnodes(2)); % transition matrix
|
wolffd@0
|
62 if cts_obs
|
wolffd@0
|
63 bnet.CPD{Oclass} = gaussian_CPD(bnet, onodes(1));
|
wolffd@0
|
64 else
|
wolffd@0
|
65 bnet.CPD{Oclass} = tabular_CPD(bnet, onodes(1));
|
wolffd@0
|
66 end
|
wolffd@0
|
67 end
|