wolffd@0
|
1 % Sigmoid Belief Hidden Markov Decision Tree (Jordan/Gharhamani 1996)
|
wolffd@0
|
2 %
|
wolffd@0
|
3 clear all;
|
wolffd@0
|
4 %clc;
|
wolffd@0
|
5 rand('state',0); randn('state',0);
|
wolffd@0
|
6 X = 1; Q1 = 2; Q2 = 3; Y = 4;
|
wolffd@0
|
7 % intra time-slice graph
|
wolffd@0
|
8 intra=zeros(4);
|
wolffd@0
|
9 intra(X,[Q1 Q2 Y])=1;
|
wolffd@0
|
10 intra(Q1,[Q2 Y])=1;
|
wolffd@0
|
11 intra(Q2, Y)=1;
|
wolffd@0
|
12 % inter time-slice graph
|
wolffd@0
|
13 inter=zeros(4);
|
wolffd@0
|
14 inter(Q1,Q1)=1;
|
wolffd@0
|
15 inter(Q2,Q2)=1;
|
wolffd@0
|
16
|
wolffd@0
|
17 ns = [1 2 3 1];
|
wolffd@0
|
18 dnodes = [2 3];
|
wolffd@0
|
19 eclass1 = [1 2 3 4];
|
wolffd@0
|
20 eclass2 = [1 5 6 4];
|
wolffd@0
|
21 bnet = mk_dbn(intra, inter, ns, dnodes, eclass1, eclass2);
|
wolffd@0
|
22
|
wolffd@0
|
23 bnet.CPD{1} = root_CPD(bnet, 1);
|
wolffd@0
|
24 % =========================================
|
wolffd@0
|
25 bnet.CPD{2} = softmax_CPD(bnet, 2);
|
wolffd@0
|
26 bnet.CPD{3} = softmax_CPD(bnet, 3, 'discrete', [2]);
|
wolffd@0
|
27 bnet.CPD{5} = softmax_CPD(bnet, 6);
|
wolffd@0
|
28 bnet.CPD{6} = softmax_CPD(bnet, 7, 'discrete', [3 6]);
|
wolffd@0
|
29 % =========================================
|
wolffd@0
|
30 bnet.CPD{4} = gaussian_CPD(bnet, 4);
|
wolffd@0
|
31
|
wolffd@0
|
32 % make some data
|
wolffd@0
|
33 T=20;
|
wolffd@0
|
34 cases = cell(4, T);
|
wolffd@0
|
35 cases(1,:)=num2cell(round(rand(1,T)*2)+1);
|
wolffd@0
|
36 %cases(2,:)=num2cell(round(rand(1,T))+1);
|
wolffd@0
|
37 %cases(3,:)=num2cell(round(rand(1,T)*2)+1);
|
wolffd@0
|
38 cases(4,:)=num2cell(rand(1,T));
|
wolffd@0
|
39
|
wolffd@0
|
40 engine = bk_inf_engine(bnet, 'exact', [1 2 3 4]);
|
wolffd@0
|
41
|
wolffd@0
|
42 % log lik before learning
|
wolffd@0
|
43 [engine, loglik] = enter_evidence(engine, cases);
|
wolffd@0
|
44
|
wolffd@0
|
45 % do learning
|
wolffd@0
|
46 ev=cell(1,1);
|
wolffd@0
|
47 ev{1}=cases;
|
wolffd@0
|
48 [bnet2, LL2] = learn_params_dbn_em(engine, ev, 10); |