Mercurial > hg > camir-aes2014
comparison toolboxes/FullBNT-1.0.7/bnt/examples/dynamic/HHMM/Map/learn_map.m @ 0:e9a9cd732c1e tip
first hg version after svn
author | wolffd |
---|---|
date | Tue, 10 Feb 2015 15:05:51 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:e9a9cd732c1e |
---|---|
1 seed = 1; | |
2 rand('state', seed); | |
3 randn('state', seed); | |
4 | |
5 obs_model = 'unique'; % each cell has a unique label (essentially fully observable) | |
6 %obs_model = 'four'; % each cell generates 4 observations, NESW | |
7 | |
8 % Generate the true network, and a randomization of it | |
9 realnet = mk_map_hhmm('p', 0.9, 'obs_model', obs_model); | |
10 rndnet = mk_rnd_map_hhmm('obs_model', obs_model); | |
11 eclass = realnet.equiv_class; | |
12 U = 1; A = 2; C = 3; F = 4; onodes = 5; | |
13 | |
14 ss = realnet.nnodes_per_slice; | |
15 T = 100; | |
16 evidence = sample_dbn(realnet, 'length', T); | |
17 ev = cell(ss,T); | |
18 ev(onodes,:) = evidence(onodes,:); | |
19 | |
20 infeng = jtree_dbn_inf_engine(rndnet); | |
21 | |
22 if 0 | |
23 % suppose we do not observe the final finish node, but only know | |
24 % it is more likely to be on that off | |
25 ev2 = ev; | |
26 infeng = enter_evidence(infeng, ev2, 'soft_evidence_nodes', [F T], 'soft_evidence', {[0.3 0.7]'}); | |
27 end | |
28 | |
29 | |
30 learnednet = learn_params_dbn_em(infeng, {evidence}, 'max_iter', 5); | |
31 | |
32 disp('real model') | |
33 disp_map_hhmm(realnet) | |
34 | |
35 disp('learned model') | |
36 disp_map_hhmm(learnednet) | |
37 | |
38 disp('rnd model') | |
39 disp_map_hhmm(rndnet) | |
40 |