data-structure

来自「VARHMMBOX, version 1.1, Iead Rezek, Oxfo」· 代码 · 共 52 行

The routines use a common data structure 'hmm' with fields:

hmm.obsmodel		name of observation model
			'Gauss'		- Gaussian
			'GaussCom'	- Gaussian with common cov
			'LIKE'		- Likelihood data, i.e. no obsmodel
			'Poisson'	- Poisson

hmm.train.obsupdate	update observation model (1 or 0)
	 .pupdate	update transition matrix (1 or 0)
	 .init		initialised (1 or 0)
	 .cyc		max number of cycles through data
	 .tol		termination tolerance of likelihood

hmm.data.Xtrain		training data
	.T		length of training sequence		
	.Xtest		testing data

hmm.K			number of hidden states
   .Pi 			initial state probability
   .P			state transition probabilities
   .LPtrain		training log likelihood

hmm.gmmll		loglikelihood of gmm model used for initialisation
hmm.mix			gaussian mixture model trained on same data

hmm.priors.Dir2d_alpha   2-D Dirichlet prior counts for Tx Probabilities
          .Dir_alpha     Dirichlet prior counts for hidden Probabilities @t=0


For 'Gauss' and 'GaussCom' observation models we also have:

hmm.state(k).Mu		mean vector for state k
            .Cov	mean covariance matrix for state k
	    .priors     priors for each state
                   .norm_Mu     Prior for mean: mean (1,dimension(data))
                   .norm_Cov    Prior for mean: covariance
                   .norm_Prec   Prior for mean: precision
                   .Wish_alpha  Prior for Covariance: scale parameter
                   .Wish_B      Prior for Covariance: shape matrix

                   .Wish_k      Prior for Covariance: dimension of shape matrix
For 'Poisson' observation models we also have:

hmm.state(k).lambda	mean/rate of poisson pdf for state k
	    .priors     priors for each state
                   .norm_alpha  Prior for mean: scale
                   .norm_beta    Prior for mean: shape


For 'LIKE' observation models we have no parameters