Compute the negative log-likelihood, using one or several specified distributions. The distributions have to be specified by their commonly known abbreviation in R, e.g. one of ['gamma', 'vm', 'pois', 'binom',...]. The named list of parameters (one value for each parameter and for each state) have to be in suitable form, i.e. a vector of the working parameters. In the likelihood computation, contemporaneous independence is assumed. Includes an optional penalization term \(\lambda\) for parameter selection of \(p\).
Usage
mllk(
theta.star,
dists,
x,
N,
p_auto,
lambda = 0,
scale_kappa = 1,
zero_inf = FALSE,
alt_data = NULL
)Arguments
- theta.star
Vector of parameters in the following order: 1) Off-diagonal entries of TPM, 2) Distribution parameters for each state (each distribution at a time, i.e. first all parameters of dist1, then all parameters of dist2 etc.), 3) Autoregression parameters (each distribution at a time, i.e. first all autoregression parameters of dist1, then all autocorrelation parameters of dist2 etc.) -> degree parameters for each state of each variable.
- dists
Vector containing abbreviated names (in R-jargon) of the distributions to be considered in the likelihood computation.
- x
List of Data vectors or matrices for which the negative log-likelihood should be computed. Each list element corresponds to one time-series. All time-series are pooled with complete pooling.
- N
Number of states.
- p_auto
Vector of autoregression degrees, one value for each state of each variable (in case of penalization choose upper bound of number of parameters).
- lambda
Complexity penalty (≥0) for autoregression parameters \(\phi\).(default: 0 no penalization).
- scale_kappa
Default 1, Scaling factor for kappa to avoid numerical issues in optimization for large kappa.
- zero_inf
Default FALSE, indicates if the gamma distributed variables should incorporate zero-inflation.
- alt_data
Default NULL, provide data here if variable name x is already taken in wrapper function.