wolffd@0: function [transmat, initState] = transmat_train_observed(labels,  nstates, varargin)
wolffd@0: % transmat_train_observed ML estimation from fully observed data
wolffd@0: % function [transmat, initState] = transmat_train_observed(labels,  nstates, varargin)
wolffd@0: %
wolffd@0: % If all sequences have the same length
wolffd@0: % labels(ex,t)
wolffd@0: % If sequences have different lengths, we use cell arrays
wolffd@0: % labels{ex}(t)
wolffd@0: 
wolffd@0: [dirichletPriorWeight, mkSymmetric, other] = process_options(...
wolffd@0:     varargin, 'dirichletPriorWeight', 0, 'mkSymmetric', 0);
wolffd@0: 
wolffd@0: if ~iscell(labels)
wolffd@0:   [numex T] = size(labels);
wolffd@0:   if T==1
wolffd@0:     labels = labels';
wolffd@0:   end
wolffd@0:   %fprintf('T=%d, numex=%d\n', T, numex);
wolffd@0:   labels = num2cell(labels,2); % each row gets its own cell
wolffd@0: end
wolffd@0: numex = length(labels);
wolffd@0: 
wolffd@0: counts = zeros(nstates, nstates);
wolffd@0: counts1 = zeros(nstates,1);
wolffd@0: for s=1:numex
wolffd@0:   labs = labels{s}; labs = labs(:)';
wolffd@0:   dat = [labs(1:end-1); labs(2:end)];
wolffd@0:   counts = counts + compute_counts(dat, [nstates nstates]);
wolffd@0:   q = labs(1);
wolffd@0:   counts1(q) = counts1(q) + 1;
wolffd@0: end
wolffd@0: pseudo_counts = dirichletPriorWeight*ones(nstates, nstates);
wolffd@0: if mkSymmetric
wolffd@0:   counts = counts + counts';
wolffd@0: end
wolffd@0: transmat = mk_stochastic(counts + pseudo_counts);
wolffd@0: initState = normalize(counts1 + dirichletPriorWeight*ones(nstates,1));
wolffd@0: 
wolffd@0: