|
wolffd@0
|
1 function bnet = mk_fhmm(N, Q, Y, discrete_obs)
|
|
wolffd@0
|
2 % MK_FHMM Make a factorial Hidden Markov Model
|
|
wolffd@0
|
3 %
|
|
wolffd@0
|
4 % There are N independent parallel hidden chains, each connected to the output
|
|
wolffd@0
|
5 %
|
|
wolffd@0
|
6 % e.g., N = 2 (vertical/diagonal edges point down)
|
|
wolffd@0
|
7 %
|
|
wolffd@0
|
8 % A1--->A2
|
|
wolffd@0
|
9 % | B1--|->B2
|
|
wolffd@0
|
10 % | / |/
|
|
wolffd@0
|
11 % Y1 Y2
|
|
wolffd@0
|
12 %
|
|
wolffd@0
|
13 % [bnet, onode] = mk_chmm(n, q, y, discrete_obs)
|
|
wolffd@0
|
14 %
|
|
wolffd@0
|
15 % Each hidden node is discrete and has Q values.
|
|
wolffd@0
|
16 % If discrete_obs = 1, each observed node is discrete and has values 1..Y.
|
|
wolffd@0
|
17 % If discrete_obs = 0, each observed node is a Gaussian vector of length Y.
|
|
wolffd@0
|
18
|
|
wolffd@0
|
19 if nargin < 2, Q = 2; end
|
|
wolffd@0
|
20 if nargin < 3, Y = 2; end
|
|
wolffd@0
|
21 if nargin < 4, discrete_obs = 1; end
|
|
wolffd@0
|
22
|
|
wolffd@0
|
23 ss = N+1;
|
|
wolffd@0
|
24 hnodes = 1:N;
|
|
wolffd@0
|
25 onode = N+1;
|
|
wolffd@0
|
26
|
|
wolffd@0
|
27 intra = zeros(ss);
|
|
wolffd@0
|
28 intra(hnodes, onode) = 1;
|
|
wolffd@0
|
29
|
|
wolffd@0
|
30 inter = eye(ss);
|
|
wolffd@0
|
31 inter(onode,onode) = 0;
|
|
wolffd@0
|
32
|
|
wolffd@0
|
33 ns = [Q*ones(1,N) Y];
|
|
wolffd@0
|
34
|
|
wolffd@0
|
35 eclass1 = [hnodes onode];
|
|
wolffd@0
|
36 eclass2 = [hnodes+ss onode];
|
|
wolffd@0
|
37 if discrete_obs
|
|
wolffd@0
|
38 dnodes = 1:ss;
|
|
wolffd@0
|
39 else
|
|
wolffd@0
|
40 dnodes = hnodes;
|
|
wolffd@0
|
41 end
|
|
wolffd@0
|
42 bnet = mk_dbn(intra, inter, ns, 'discrete', dnodes, 'eclass1', eclass1, 'eclass2', eclass2, ...
|
|
wolffd@0
|
43 'observed', onode);
|
|
wolffd@0
|
44
|
|
wolffd@0
|
45 for i=hnodes(:)'
|
|
wolffd@0
|
46 bnet.CPD{i} = tabular_CPD(bnet, i);
|
|
wolffd@0
|
47 end
|
|
wolffd@0
|
48 i = onode;
|
|
wolffd@0
|
49 if discrete_obs
|
|
wolffd@0
|
50 bnet.CPD{i} = tabular_CPD(bnet, i);
|
|
wolffd@0
|
51 else
|
|
wolffd@0
|
52 bnet.CPD{i} = gaussian_CPD(bnet, i);
|
|
wolffd@0
|
53 end
|
|
wolffd@0
|
54 for i=hnodes(:)'+ss
|
|
wolffd@0
|
55 bnet.CPD{i} = tabular_CPD(bnet, i);
|
|
wolffd@0
|
56 end
|
|
wolffd@0
|
57
|
|
wolffd@0
|
58
|
