diff toolboxes/FullBNT-1.0.7/bnt/examples/dynamic/mhmm1.m @ 0:e9a9cd732c1e tip

first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/toolboxes/FullBNT-1.0.7/bnt/examples/dynamic/mhmm1.m	Tue Feb 10 15:05:51 2015 +0000
@@ -0,0 +1,85 @@
+% Make an HMM with mixture of Gaussian observations
+%    Q1 ---> Q2
+%  /  |   /  |
+% M1  |  M2  | 
+%  \  v   \  v
+%    Y1     Y2 
+% where Pr(m=j|q=i) is a multinomial and Pr(y|m,q) is a Gaussian     
+
+%seed = 3;
+%rand('state', seed);
+%randn('state', seed);
+
+intra = zeros(3);
+intra(1,[2 3]) = 1;
+intra(2,3) = 1;
+inter = zeros(3);
+inter(1,1) = 1;
+n = 3;
+
+Q = 2; % num hidden states
+O = 2; % size of observed vector
+M = 2; % num mixture components per state
+
+ns = [Q M O];
+dnodes = [1 2];
+onodes = [3];
+eclass1 = [1 2 3];
+eclass2 = [4 2 3];
+bnet = mk_dbn(intra, inter, ns, 'discrete', dnodes, 'eclass1', eclass1, 'eclass2', eclass2, ...
+	      'observed', onodes);
+
+prior0 = normalise(rand(Q,1));
+transmat0 = mk_stochastic(rand(Q,Q));
+mixmat0 = mk_stochastic(rand(Q,M));
+mu0 = rand(O,Q,M);
+Sigma0 = repmat(eye(O), [1 1 Q M]);
+bnet.CPD{1} = tabular_CPD(bnet, 1, prior0);
+bnet.CPD{2} = tabular_CPD(bnet, 2, mixmat0);
+%% we set the cov prior to 0 to give same results as HMM toolbox
+%bnet.CPD{3} = gaussian_CPD(bnet, 3, 'mean', mu0, 'cov', Sigma0, 'cov_prior_weight', 0);
+% new version of HMM toolbox uses the same default prior on Gaussians as BNT
+bnet.CPD{3} = gaussian_CPD(bnet, 3, 'mean', mu0, 'cov', Sigma0);
+bnet.CPD{4} = tabular_CPD(bnet, 4, transmat0);
+
+
+
+T = 5; % fixed length sequences
+
+engine = {};
+engine{end+1} = hmm_inf_engine(bnet);
+engine{end+1} = smoother_engine(jtree_2TBN_inf_engine(bnet));
+engine{end+1} = smoother_engine(hmm_2TBN_inf_engine(bnet));
+if 0
+engine{end+1} = jtree_unrolled_dbn_inf_engine(bnet, T);
+%engine{end+1} = frontier_inf_engine(bnet);
+engine{end+1} = bk_inf_engine(bnet, 'clusters', 'exact');
+engine{end+1} = jtree_dbn_inf_engine(bnet);
+end
+
+inf_time = cmp_inference_dbn(bnet, engine, T);
+
+ncases = 2;
+max_iter = 2;
+[learning_time, CPD, LL, cases] = cmp_learning_dbn(bnet, engine, T, 'ncases', ncases, 'max_iter', max_iter);
+
+% Compare to HMM toolbox
+
+data = zeros(O, T, ncases);
+for i=1:ncases
+  data(:,:,i) = reshape(cell2num(cases{i}(onodes,:)), [O T]);
+end
+tic;
+[LL2, prior2, transmat2, mu2, Sigma2, mixmat2] = ...
+    mhmm_em(data, prior0, transmat0,  mu0, Sigma0, mixmat0, 'max_iter', max_iter);
+t=toc;
+disp(['HMM toolbox took ' num2str(t) ' seconds '])
+
+for e = 1:length(engine)
+  assert(approxeq(prior2, CPD{e,1}.CPT))
+  assert(approxeq(mixmat2, CPD{e,2}.CPT))
+  assert(approxeq(mu2, CPD{e,3}.mean))
+  assert(approxeq(Sigma2, CPD{e,3}.cov))
+  assert(approxeq(transmat2, CPD{e,4}.CPT))
+  assert(approxeq(LL2, LL{e}))
+end