HMMLikelihood function

Hidden Markov Model likelihood functions

Function HMMLikelihood computes the log-likelihood via hmm.lnl which is a wrapper for the FORTRAN code hmm_like.f. The function HMMlikelihood is called from optimizer and it in turn calls hmm.lnl after setting up parameters.

For an R version of the HMMLikelihood and related code see R_HMMLikelihood

HMMLikelihood(par,type,xx,xstart,mx,T,freq=1,fct_dmat,fct_gamma,fct_delta,ddl,
                         dml,parameters,debug=FALSE,return.mat=FALSE,sup=NULL,check=FALSE)
       reals(ddl,dml,parameters,parlist,indices=NULL)
       hmm.lnl(x,start,m,T,dmat,gamma,delta,freq,debug)

Arguments

• par: vector of parameter values for log-likelihood evaluation
• type: vector of parameter names used to split par vector into types
• xx: matrix of observed sequences (row:id; column:occasion/time); xx used instead of x to avoid conflict in optimx
• xstart: for each ch, the first non-zero x value and the occasion of the first non-zero value; ; xstart used instead of start to avoid conflict in optimx
• mx: number of states; mx used instead of m to avoid conflict in optimx
• T: number of occasions; sequence length
• freq: vector of history frequencies or 1
• fct_dmat: function to create D from parameters
• fct_gamma: function to create gamma - transition matrix
• fct_delta: function to create initial state distribution
• ddl: design data list of parameters for each id
• dml: list of design matrices; one entry for each parameter; each entry contains fe and re for fixed and random effects
• parameters: formulas for each parameter type
• debug: if TRUE, print out par values and -log-likelihood
• return.mat: If TRUE, returns list of transition, observation and delta arrays.
• sup: list of supplemental information that may be needed by the function but only needs to be computed once; currently only used for MVMS models for dmat
• check: if TRUE, checks validity of gamma, dmat and delta to look for any errors
• x: same as xx but for call to hmm.lnl
• m: same as mx but for call to hmm.lnl
• dmat: observation probability matrices
• gamma: transition matrices
• delta: initial distribution
• parlist: list of parameter strings used to split par vector
• start: same as xstart but for hmm.lnl
• indices: specific indices for computation unless NULL

Returns

HMMLikelihood returns log-likelihood for a single sequence and hmm.lnl returns the negative log-likelihood value for each capture history. reals returns either the column dimension of design matrix for parameter or the real parameter vector

References

Zucchini, W. and I.L. MacDonald. 2009. Hidden Markov Models for Time Series: An Introduction using R. Chapman and Hall, Boca Raton, FL. 275p.

See Also

R_HMMLikelihood

Author(s)

Jeff Laake