|
wolffd@0
|
1 % Make an HMM with discrete observations
|
|
wolffd@0
|
2 % X1 -> X2
|
|
wolffd@0
|
3 % | |
|
|
wolffd@0
|
4 % v v
|
|
wolffd@0
|
5 % Y1 Y2
|
|
wolffd@0
|
6
|
|
wolffd@0
|
7 intra = zeros(2);
|
|
wolffd@0
|
8 intra(1,2) = 1;
|
|
wolffd@0
|
9 inter = zeros(2);
|
|
wolffd@0
|
10 inter(1,1) = 1;
|
|
wolffd@0
|
11 n = 2;
|
|
wolffd@0
|
12
|
|
wolffd@0
|
13 Q = 2; % num hidden states
|
|
wolffd@0
|
14 O = 2; % num observable symbols
|
|
wolffd@0
|
15
|
|
wolffd@0
|
16 ns = [Q O];
|
|
wolffd@0
|
17 dnodes = 1:2;
|
|
wolffd@0
|
18 onodes = [2];
|
|
wolffd@0
|
19 eclass1 = [1 2];
|
|
wolffd@0
|
20 eclass2 = [3 2];
|
|
wolffd@0
|
21 bnet = mk_dbn(intra, inter, ns, 'discrete', dnodes, 'eclass1', eclass1, 'eclass2', eclass2, ...
|
|
wolffd@0
|
22 'observed', onodes);
|
|
wolffd@0
|
23
|
|
wolffd@0
|
24 rand('state', 0);
|
|
wolffd@0
|
25 prior1 = normalise(rand(Q,1));
|
|
wolffd@0
|
26 transmat1 = mk_stochastic(rand(Q,Q));
|
|
wolffd@0
|
27 obsmat1 = mk_stochastic(rand(Q,O));
|
|
wolffd@0
|
28 bnet.CPD{1} = tabular_CPD(bnet, 1, prior1);
|
|
wolffd@0
|
29 bnet.CPD{2} = tabular_CPD(bnet, 2, obsmat1);
|
|
wolffd@0
|
30 bnet.CPD{3} = tabular_CPD(bnet, 3, transmat1);
|
|
wolffd@0
|
31
|
|
wolffd@0
|
32
|
|
wolffd@0
|
33 T = 5; % fixed length sequences
|
|
wolffd@0
|
34
|
|
wolffd@0
|
35 engine = {};
|
|
wolffd@0
|
36 engine{end+1} = jtree_unrolled_dbn_inf_engine(bnet, T);
|
|
wolffd@0
|
37 engine{end+1} = hmm_inf_engine(bnet);
|
|
wolffd@0
|
38 engine{end+1} = smoother_engine(hmm_2TBN_inf_engine(bnet));
|
|
wolffd@0
|
39 engine{end+1} = smoother_engine(jtree_2TBN_inf_engine(bnet));
|
|
wolffd@0
|
40 if 1
|
|
wolffd@0
|
41 %engine{end+1} = frontier_inf_engine(bnet); % broken
|
|
wolffd@0
|
42 engine{end+1} = bk_inf_engine(bnet, 'clusters', {[1]});
|
|
wolffd@0
|
43 engine{end+1} = jtree_dbn_inf_engine(bnet);
|
|
wolffd@0
|
44 end
|
|
wolffd@0
|
45
|
|
wolffd@0
|
46 inf_time = cmp_inference_dbn(bnet, engine, T);
|
|
wolffd@0
|
47
|
|
wolffd@0
|
48 ncases = 2;
|
|
wolffd@0
|
49 max_iter = 2;
|
|
wolffd@0
|
50 [learning_time, CPD, LL, cases] = cmp_learning_dbn(bnet, engine, T, 'ncases', ncases, 'max_iter', max_iter);
|
|
wolffd@0
|
51
|
|
wolffd@0
|
52 % Compare to HMM toolbox
|
|
wolffd@0
|
53
|
|
wolffd@0
|
54 data = zeros(ncases, T);
|
|
wolffd@0
|
55 for i=1:ncases
|
|
wolffd@0
|
56 %data(i,:) = cat(2, cases{i}{onodes,:});
|
|
wolffd@0
|
57 data(i,:) = cell2num(cases{i}(onodes,:));
|
|
wolffd@0
|
58 end
|
|
wolffd@0
|
59 [LL2, prior2, transmat2, obsmat2] = dhmm_em(data, prior1, transmat1, obsmat1, 'max_iter', max_iter);
|
|
wolffd@0
|
60
|
|
wolffd@0
|
61 e = 1;
|
|
wolffd@0
|
62 assert(approxeq(prior2, CPD{e,1}.CPT))
|
|
wolffd@0
|
63 assert(approxeq(obsmat2, CPD{e,2}.CPT))
|
|
wolffd@0
|
64 assert(approxeq(transmat2, CPD{e,3}.CPT))
|
|
wolffd@0
|
65 assert(approxeq(LL2, LL{e}))
|
|
wolffd@0
|
66
|
