wolffd@0: % Compare the speeds of various inference engines on the water DBN
wolffd@0: 
wolffd@0: [bnet, onodes] = mk_water_dbn;
wolffd@0: 
wolffd@0: T = 3;
wolffd@0: 
wolffd@0: engine = {};
wolffd@0: engine{end+1} = jtree_unrolled_dbn_inf_engine(bnet, T, onodes);
wolffd@0: engine{end+1} = hmm_inf_engine(bnet, onodes);
wolffd@0: engine{end+1} = frontier_inf_engine(bnet, onodes);
wolffd@0: engine{end+1} = jtree_dbn_inf_engine(bnet, onodes);
wolffd@0: engine{end+1} = bk_inf_engine(bnet, 'exact', onodes);
wolffd@0: 
wolffd@0: engine{end+1} = bk_inf_engine(bnet, 'ff', onodes);
wolffd@0: engine{end+1} = bk_inf_engine(bnet, { [1 2], [3 4 5 6], [7 8] }, onodes);
wolffd@0: 
wolffd@0: N = length(engine);
wolffd@0: exact = 1:5;
wolffd@0: 
wolffd@0: 
wolffd@0: filter = 0;
wolffd@0: err = cmp_inference(bnet, onodes, engine, exact, T, filter);
wolffd@0: 
wolffd@0: % elapsed times for enter_evidence  (matlab 5.3 on PIII with 256MB running Redhat linux)  
wolffd@0: 
wolffd@0: % T = 5, 4/20/00
wolffd@0: %     0.6266 unrolled *
wolffd@0: %     0.3490 hmm *
wolffd@0: %     1.1743 frontier
wolffd@0: %     1.4621 old frontier
wolffd@0: %     0.3270 fast frontier *
wolffd@0: %     1.3926 jtree 
wolffd@0: %     1.3790 bk 
wolffd@0: %     0.4916 fast bk
wolffd@0: %     0.4190 fast bk compiled
wolffd@0: %     0.3574 fast jtree *
wolffd@0: 
wolffd@0: 
wolffd@0: err = cmp_learning(bnet, onodes, engine, exact, T);
wolffd@0: 
wolffd@0: % elapsed times for learn_params_dbn_em (matlab 5.3 on PIII with 256MB running Redhat linux)  
wolffd@0: 
wolffd@0: % T = 5, 2cases, 2 iter, 4/20/00
wolffd@0: % 3.5750 unrolled
wolffd@0: % 3.7475 hmm
wolffd@0: % 2.1452 fast frontier
wolffd@0: % 2.5724 fast bk compiled
wolffd@0: % 2.3387 fast jtree