Daniel@0: % Make the following network (from Jensen (1996) p84 fig 4.17)
Daniel@0: %    1
Daniel@0: %  / | \
Daniel@0: % 2  3  4
Daniel@0: % |  |  |
Daniel@0: % 5  6  7
Daniel@0: %  \/ \/
Daniel@0: %  8   9
Daniel@0: % where all arcs point downwards
Daniel@0: 
Daniel@0: 
Daniel@0: N = 9;
Daniel@0: dag = zeros(N,N);
Daniel@0: dag(1,2)=1; dag(1,3)=1; dag(1,4)=1;
Daniel@0: dag(2,5)=1; dag(3,6)=1; dag(4,7)=1;
Daniel@0: dag(5,8)=1; dag(6,8)=1; dag(6,9)=1; dag(7,9) = 1;
Daniel@0: 
Daniel@0: ns = [5 4 3 2 2 1 2 2 2]; % vector-valued nodes
Daniel@0: %ns = ones(1,9); % scalar nodes
Daniel@0: dnodes = [];
Daniel@0: 
Daniel@0: bnet = mk_bnet(dag, ns, 'discrete', []);
Daniel@0: rand('state', 0);
Daniel@0: randn('state', 0);
Daniel@0: for i=1:N
Daniel@0:   bnet.CPD{i} = gaussian_CPD(bnet, i);
Daniel@0: end
Daniel@0: 
Daniel@0: clear engine;
Daniel@0: engine{1} = gaussian_inf_engine(bnet);
Daniel@0: engine{2} = jtree_inf_engine(bnet);
Daniel@0: 
Daniel@0: [err, time] = cmp_inference_static(bnet, engine);
Daniel@0: 
Daniel@0: Nsamples = 100;
Daniel@0: samples = cell(N, Nsamples);
Daniel@0: for s=1:Nsamples
Daniel@0:   samples(:,s) = sample_bnet(bnet);
Daniel@0: end
Daniel@0: bnet2 = learn_params(bnet, samples);