em: Parameter estimation by EM algorithm

In AdvancedR-2021/hmm: Hidden Markov Models

Description

Calculates approximate MLE's of all involved parameters, i.e. initial distribution probabilities, transition probabilities and whichever parameters are required for the marginal distribution of the emissions. Not expected to be called by the user directly, instead being called during the initialization of an hmm object.

Usage

em(
  obs,
  gamma,
  delta,
  lls,
  param_lls,
  lls_mle,
  epsilon = 1e-05,
  max_iter = 1000,
  ...
)

em.binom(obs, gamma, delta, size, prob, ...)

em.exponential(obs, gamma, delta, rate, ...)

em.normal(obs, gamma, delta, mean, sd, ...)

em.poisson(obs, gamma, delta, lambda, ...)

Arguments

obs	The observed data
gamma	Initial value of transition matrix
delta	Initial value of initial distribution
lls	List of marginal densities
param_lls	List of parameters for marginal densities
lls_mle	List of functions that generate MLE's
epsilon	Threshold to stop algorithm when difference in log-likelihood is less than this
max_iter	Maximum number of iterations before stopping
...	Additional arguments, mainly here to avoid "unused argument"-errors.

Value

A list of log-likelihoods, parameter estimates and number of iterations.

Functions

• em.binom: EM-algorithm for binomial emissions

• em.exponential: EM-algorithm for exponential emissions

• em.normal: EM-algorithm for normal emissions

• em.poisson: EM-algorithm for Poisson emissions

AdvancedR-2021/hmm documentation built on Dec. 17, 2021, 7:41 a.m.