Daniel@0: % Compare the speeds of various inference engines on the DBN in Kjaerulff
Daniel@0: % "dHugin: A computational system for dynamic time-sliced {B}ayesian networks",
Daniel@0: % Intl. J. Forecasting 11:89-111, 1995.
Daniel@0: %
Daniel@0: % The intra structure is (all arcs point downwards)
Daniel@0: %
Daniel@0: %  1 -> 2
Daniel@0: %   \  /
Daniel@0: %     3
Daniel@0: %     |
Daniel@0: %     4
Daniel@0: %    / \
Daniel@0: %   5   6
Daniel@0: %   \  /
Daniel@0: %     7
Daniel@0: %     |
Daniel@0: %     8
Daniel@0: %
Daniel@0: % The inter structure is 1->1, 4->4, 8->8
Daniel@0: 
Daniel@0: seed = 0;
Daniel@0: rand('state', seed);
Daniel@0: randn('state', seed);
Daniel@0: 
Daniel@0: ss = 8;
Daniel@0: intra = zeros(ss);
Daniel@0: intra(1,[2 3])=1;
Daniel@0: intra(2,3)=1;
Daniel@0: intra(3,4)=1;
Daniel@0: intra(4,[5 6])=1;
Daniel@0: intra([5 6], 7)=1;
Daniel@0: intra(7,8)=1;
Daniel@0: 
Daniel@0: inter = zeros(ss);
Daniel@0: inter(1,1)=1;
Daniel@0: inter(4,4)=1;
Daniel@0: inter(8,8)=1;
Daniel@0: 
Daniel@0: ns = 2*ones(1,ss);
Daniel@0: onodes = 2;
Daniel@0: bnet = mk_dbn(intra, inter, ns, 'observed', onodes, 'eclass2', (1:ss)+ss);
Daniel@0: for i=1:2*ss
Daniel@0:   bnet.CPD{i} = tabular_CPD(bnet, i);
Daniel@0: end
Daniel@0: 
Daniel@0: T = 4;
Daniel@0: 
Daniel@0: engine = {};
Daniel@0: engine{end+1} = jtree_unrolled_dbn_inf_engine(bnet, T);
Daniel@0: engine{end+1} = jtree_dbn_inf_engine(bnet);
Daniel@0: engine{end+1} = smoother_engine(jtree_2TBN_inf_engine(bnet));
Daniel@0: %engine{end+1} = smoother_engine(hmm_2TBN_inf_engine(bnet)); % observed nodes have children
Daniel@0: 
Daniel@0: inf_time = cmp_inference_dbn(bnet, engine, T)
Daniel@0: learning_time = cmp_learning_dbn(bnet, engine, T)