wolffd@0: function bnet = mk_chmm(N, Q, Y, discrete_obs, coupled, CPD)
wolffd@0: % MK_CHMM Make a coupled Hidden Markov Model
wolffd@0: %
wolffd@0: % There are N hidden nodes, each connected to itself and its two nearest neighbors in the next
wolffd@0: % slice (apart from the edges, where there is 1 nearest neighbor).
wolffd@0: %
wolffd@0: % Example: If N = 3, the hidden backbone is as follows, where all arrows point to the righ+t
wolffd@0: %
wolffd@0: % X1--X2
wolffd@0: %   \/ 
wolffd@0: %   /\
wolffd@0: % X2--X2
wolffd@0: %   \/ 
wolffd@0: %   /\
wolffd@0: % X3--X3
wolffd@0: %
wolffd@0: % Each hidden node has a "private" observed child (not shown).
wolffd@0: %
wolffd@0: % BNET = MK_CHMM(N, Q, Y)
wolffd@0: % Each hidden node is discrete and has Q values.
wolffd@0: % Each observed node is a Gaussian vector of length Y.
wolffd@0: %
wolffd@0: % BNET = MK_CHMM(N, Q, Y, DISCRETE_OBS)
wolffd@0: % If discrete_obs = 1, the observations are discrete (values in {1, .., Y}).
wolffd@0: %
wolffd@0: % BNET = MK_CHMM(N, Q, Y, DISCRETE_OBS, COUPLED)
wolffd@0: % If coupled = 0, the chains are not coupled, i.e., we make N parallel HMMs.
wolffd@0: %
wolffd@0: % BNET = MK_CHMM(N, Q, Y, DISCRETE_OBS, COUPLED, CPDs)
wolffd@0: % means use the specified CPD structures instead of creating random params.
wolffd@0: %  CPD{i}.CPT, i=1:N specifies the prior
wolffd@0: %  CPD{i}.CPT, i=2N+1:3N specifies the transition model
wolffd@0: %  CPD{i}.mean, CPD{i}.cov, i=N+1:2N specifies the observation model if Gaussian
wolffd@0: %  CPD{i}.CPT, i=N+1:2N if discrete
wolffd@0: 
wolffd@0: 
wolffd@0: if nargin < 2, Q = 2; end
wolffd@0: if nargin < 3, Y = 1; end
wolffd@0: if nargin < 4, discrete_obs = 0; end
wolffd@0: if nargin < 5, coupled = 1; end
wolffd@0: if nargin < 6, rnd = 1; else rnd = 0; end
wolffd@0:   
wolffd@0: ss = N*2;
wolffd@0: hnodes = 1:N;
wolffd@0: onodes = (1:N)+N;
wolffd@0: 
wolffd@0: intra = zeros(ss);
wolffd@0: for i=1:N
wolffd@0:   intra(hnodes(i), onodes(i))=1;
wolffd@0: end
wolffd@0: 
wolffd@0: inter = zeros(ss);
wolffd@0: if coupled
wolffd@0:   for i=1:N
wolffd@0:     inter(i, max(i-1,1):min(i+1,N))=1;
wolffd@0:   end
wolffd@0: else
wolffd@0:   inter(1:N, 1:N) = eye(N);
wolffd@0: end  
wolffd@0: 
wolffd@0: ns = [Q*ones(1,N) Y*ones(1,N)]; 
wolffd@0: 
wolffd@0: eclass1 = [hnodes onodes];
wolffd@0: eclass2 = [hnodes+ss onodes];
wolffd@0: if discrete_obs
wolffd@0:   dnodes = 1:ss;
wolffd@0: else
wolffd@0:   dnodes = hnodes;
wolffd@0: end
wolffd@0: bnet = mk_dbn(intra, inter, ns, 'discrete', dnodes, 'eclass1', eclass1, 'eclass2', eclass2, ...
wolffd@0: 	      'observed', onodes);
wolffd@0: 
wolffd@0: if rnd
wolffd@0:   for i=hnodes(:)'
wolffd@0:     bnet.CPD{i} = tabular_CPD(bnet, i);
wolffd@0:   end
wolffd@0:   for i=onodes(:)'
wolffd@0:     if discrete_obs
wolffd@0:       bnet.CPD{i} = tabular_CPD(bnet, i);
wolffd@0:     else
wolffd@0:       bnet.CPD{i} = gaussian_CPD(bnet, i);
wolffd@0:     end
wolffd@0:   end
wolffd@0:   for i=hnodes(:)'+ss
wolffd@0:     bnet.CPD{i} = tabular_CPD(bnet, i);
wolffd@0:   end
wolffd@0: else
wolffd@0:   for i=hnodes(:)'
wolffd@0:     bnet.CPD{i} = tabular_CPD(bnet, i, CPD{i}.CPT);
wolffd@0:   end
wolffd@0:   for i=onodes(:)'
wolffd@0:     if discrete_obs
wolffd@0:       bnet.CPD{i} = tabular_CPD(bnet, i, CPD{i}.CPT);
wolffd@0:     else
wolffd@0:       bnet.CPD{i} = gaussian_CPD(bnet, i, CPD{i}.mean, CPD{i}.cov);
wolffd@0:     end
wolffd@0:   end
wolffd@0:   for i=hnodes(:)'+ss
wolffd@0:     bnet.CPD{i} = tabular_CPD(bnet, i, CPD{i}.CPT);
wolffd@0:   end
wolffd@0: end
wolffd@0: 
wolffd@0: