1 files changed, 40 insertions, 0 deletions
diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/learn_map.m b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/learn_map.m
new file mode 100644
index 00000000..ac36586a
--- /dev/null
+++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/learn_map.m
@@ -0,0 +1,40 @@
+seed = 1;
+rand('state', seed);
+randn('state', seed);
+
+obs_model = 'unique';  % each cell has a unique label (essentially fully observable)
+%obs_model = 'four'; % each cell generates 4 observations, NESW
+
+% Generate the true network, and a randomization of it
+realnet = mk_map_hhmm('p', 0.9, 'obs_model', obs_model);
+rndnet = mk_rnd_map_hhmm('obs_model', obs_model);
+eclass = realnet.equiv_class;
+U = 1; A = 2; C = 3; F = 4; onodes = 5;
+
+ss = realnet.nnodes_per_slice;
+T = 100;
+evidence = sample_dbn(realnet, 'length', T);
+ev = cell(ss,T);
+ev(onodes,:) = evidence(onodes,:);
+
+infeng = jtree_dbn_inf_engine(rndnet);
+
+if 0
+% suppose we do not observe the final finish node, but only know 
+% it is more likely to be on that off
+ev2 = ev;
+infeng = enter_evidence(infeng, ev2, 'soft_evidence_nodes', [F T], 'soft_evidence',  {[0.3 0.7]'});
+end
+
+
+learnednet = learn_params_dbn_em(infeng, {evidence}, 'max_iter', 5);
+
+disp('real model')
+disp_map_hhmm(realnet)
+
+disp('learned model')
+disp_map_hhmm(learnednet)
+
+disp('rnd model')
+disp_map_hhmm(rndnet)
+