wolffd@0: seed = 1;
wolffd@0: rand('state', seed);
wolffd@0: randn('state', seed);
wolffd@0: 
wolffd@0: obs_model = 'unique';  % each cell has a unique label (essentially fully observable)
wolffd@0: %obs_model = 'four'; % each cell generates 4 observations, NESW
wolffd@0: 
wolffd@0: % Generate the true network, and a randomization of it
wolffd@0: realnet = mk_map_hhmm('p', 0.9, 'obs_model', obs_model);
wolffd@0: rndnet = mk_rnd_map_hhmm('obs_model', obs_model);
wolffd@0: eclass = realnet.equiv_class;
wolffd@0: U = 1; A = 2; C = 3; F = 4; onodes = 5;
wolffd@0: 
wolffd@0: ss = realnet.nnodes_per_slice;
wolffd@0: T = 100;
wolffd@0: evidence = sample_dbn(realnet, 'length', T);
wolffd@0: ev = cell(ss,T);
wolffd@0: ev(onodes,:) = evidence(onodes,:);
wolffd@0: 
wolffd@0: infeng = jtree_dbn_inf_engine(rndnet);
wolffd@0: 
wolffd@0: if 0
wolffd@0: % suppose we do not observe the final finish node, but only know 
wolffd@0: % it is more likely to be on that off
wolffd@0: ev2 = ev;
wolffd@0: infeng = enter_evidence(infeng, ev2, 'soft_evidence_nodes', [F T], 'soft_evidence',  {[0.3 0.7]'});
wolffd@0: end
wolffd@0: 
wolffd@0: 
wolffd@0: learnednet = learn_params_dbn_em(infeng, {evidence}, 'max_iter', 5);
wolffd@0: 
wolffd@0: disp('real model')
wolffd@0: disp_map_hhmm(realnet)
wolffd@0: 
wolffd@0: disp('learned model')
wolffd@0: disp_map_hhmm(learnednet)
wolffd@0: 
wolffd@0: disp('rnd model')
wolffd@0: disp_map_hhmm(rndnet)
wolffd@0: