wolffd@0: % Sigmoid Belief Hierarchical Mixtures of Experts
wolffd@0: 
wolffd@0: clear all
wolffd@0: clc
wolffd@0: X = 1;
wolffd@0: Q1 = 2;
wolffd@0: Q2 = 3;
wolffd@0: Y = 4;
wolffd@0: dag = zeros(4,4);
wolffd@0: dag(X,[Q1 Q2 Y]) = 1;
wolffd@0: dag(Q1, [Q2 Y]) = 1;
wolffd@0: dag(Q2,Y)=1;
wolffd@0: ns = [1 3 4 3];
wolffd@0: dnodes = [2 3 4];
wolffd@0: onodes=[1 2 3 4];
wolffd@0: bnet = mk_bnet(dag,ns, dnodes);
wolffd@0: 
wolffd@0: rand('state',0); randn('state',0);
wolffd@0: 
wolffd@0: bnet.CPD{1} = root_CPD(bnet, 1);
wolffd@0: bnet.CPD{2} = softmax_CPD(bnet, 2, 'max_iter', 3);
wolffd@0: bnet.CPD{3} = softmax_CPD(bnet, 3, 'discrete', [2], 'max_iter', 3);
wolffd@0: bnet.CPD{4} = softmax_CPD(bnet, 4, 'discrete', [2 3], 'max_iter', 3);
wolffd@0: 
wolffd@0: T=5;
wolffd@0: cases = cell(4, T);
wolffd@0: cases(1,:)=num2cell(rand(1,T));
wolffd@0: %cases(2,:)=num2cell(round(rand(1,T)*2)+1);
wolffd@0: %cases(3,:)=num2cell(round(rand(1,T)*3)+1);
wolffd@0: cases(4,:)=num2cell(round(rand(1,T)*2)+1);
wolffd@0: 
wolffd@0: engine = jtree_inf_engine(bnet, onodes);
wolffd@0: 
wolffd@0: [engine, loglik] = enter_evidence(engine, cases);
wolffd@0: 
wolffd@0: disp('learning-------------------------------------------')
wolffd@0: [bnet2, LL2] = learn_params_em(engine, cases, 4);