camir-aes2014: toolboxes/FullBNT-1.0.7/bnt/examples/dynamic/HHMM/Map/learn

first hg version after svn

rev	line source
wolffd@0	1 seed = 1;
wolffd@0	2 rand('state', seed);
wolffd@0	3 randn('state', seed);
wolffd@0	4
wolffd@0	5 obs_model = 'unique'; % each cell has a unique label (essentially fully observable)
wolffd@0	6 %obs_model = 'four'; % each cell generates 4 observations, NESW
wolffd@0	7
wolffd@0	8 % Generate the true network, and a randomization of it
wolffd@0	9 realnet = mk_map_hhmm('p', 0.9, 'obs_model', obs_model);
wolffd@0	10 rndnet = mk_rnd_map_hhmm('obs_model', obs_model);
wolffd@0	11 eclass = realnet.equiv_class;
wolffd@0	12 U = 1; A = 2; C = 3; F = 4; onodes = 5;
wolffd@0	13
wolffd@0	14 ss = realnet.nnodes_per_slice;
wolffd@0	15 T = 100;
wolffd@0	16 evidence = sample_dbn(realnet, 'length', T);
wolffd@0	17 ev = cell(ss,T);
wolffd@0	18 ev(onodes,:) = evidence(onodes,:);
wolffd@0	19
wolffd@0	20 infeng = jtree_dbn_inf_engine(rndnet);
wolffd@0	21
wolffd@0	22 if 0
wolffd@0	23 % suppose we do not observe the final finish node, but only know
wolffd@0	24 % it is more likely to be on that off
wolffd@0	25 ev2 = ev;
wolffd@0	26 infeng = enter_evidence(infeng, ev2, 'soft_evidence_nodes', [F T], 'soft_evidence', {[0.3 0.7]'});
wolffd@0	27 end
wolffd@0	28
wolffd@0	29
wolffd@0	30 learnednet = learn_params_dbn_em(infeng, {evidence}, 'max_iter', 5);
wolffd@0	31
wolffd@0	32 disp('real model')
wolffd@0	33 disp_map_hhmm(realnet)
wolffd@0	34
wolffd@0	35 disp('learned model')
wolffd@0	36 disp_map_hhmm(learnednet)
wolffd@0	37
wolffd@0	38 disp('rnd model')
wolffd@0	39 disp_map_hhmm(rndnet)
wolffd@0	40

Mercurial > hg > camir-aes2014