Fit an autoregressive HMM with within-state autoregression to data
using a specified function to compute the negative
log-likelihood. This function gets by default minimized using the function optim.
It returns the estimated parameters of the fitted model. The distributional assumptions for the HMM
can be specified using the parameter dist.
Usage
fit_arp_model(
mllk,
data,
theta.star,
N,
p_auto,
dists,
opt_fun = "optim",
scale_kappa = 1,
zero_inf = FALSE,
lambda = 0
)Arguments
- mllk
Negative log-likelihood function that should be minimized.
- data
Data that should be fitted using the HMM.
- theta.star
Unconstrained parameter vector (has to be provided in suitable form). Attention: -Inf values (resulting e.g. from supplying autocorrelation = 0) are not possible for the optimization function.
- N
Number of states.
- p_auto
Vector of degree of autoregression for each distribution, 0 = no autoregression, one value for every state of every variable.
- dists
Vector containing abbreviated names (in R-jargon) of the distributions to be considered in the likelihood computation.
- opt_fun
string - Function that should be used for optimization (default: optim, one of ['optim', 'nlm', 'genoud']).
- scale_kappa
Default 1, Scaling factor for kappa to avoid numerical issues in optimization for large kappa in the von Mises distribution.
- zero_inf
Default FALSE, indicates if the gamma distributed variables should incorporate zero-inflation.
- lambda
Default 0, complexity penalty for lasso regularization of autoregression coefficients.