Daniel@0: function [initState, transmat, mu, Nproto, pick] = mhmmParzen_train_observed(obsData, hiddenData, ...
Daniel@0: 						  nstates, maxNproto, varargin)
Daniel@0: % mhmmParzentrain_observed  with mixture of Gaussian outputs from fully observed sequences
Daniel@0: % function [initState, transmat, mu, Nproto] = mhmm_train_observed_parzen(obsData, hiddenData, ...
Daniel@0: %						  nstates, maxNproto)
Daniel@0: %
Daniel@0: %
Daniel@0: % INPUT
Daniel@0: % If all sequences have the same length
Daniel@0: % obsData(:,t,ex) 
Daniel@0: % hiddenData(ex,t)  - must be ROW vector if only one sequence
Daniel@0: % If sequences have different lengths, we use cell arrays
Daniel@0: % obsData{ex}(:,t) 
Daniel@0: % hiddenData{ex}(t)
Daniel@0: %
Daniel@0: % Optional argumnets
Daniel@0: % dirichletPriorWeight - for smoothing transition matrix counts
Daniel@0: % mkSymmetric
Daniel@0: %
Daniel@0: % Output
Daniel@0: % mu(:,q)
Daniel@0: % Nproto(q) is the number of prototypes (mixture components) chosen for state q
Daniel@0: 
Daniel@0: [transmat, initState] = transmat_train_observed(...
Daniel@0:     hiddenData, nstates, varargin{:});
Daniel@0: 
Daniel@0: % convert to obsData(:,t*nex)
Daniel@0: if ~iscell(obsData)
Daniel@0:   [D T Nex] = size(obsData);
Daniel@0:   obsData = reshape(obsData, D, T*Nex);
Daniel@0: else
Daniel@0:   obsData = cat(2, obsData{:});
Daniel@0:   hiddenData = cat(2, hiddenData{:});
Daniel@0: end
Daniel@0: [mu, Nproto, pick] = parzen_fit_select_unif(obsData, hiddenData(:), maxNproto);
Daniel@0: 
Daniel@0: