The function performs Viterbi algorithm (Viterbi, 1967). It can be applied to a ContObservHMM object after sufficient number of Baum-welch iterations (function baumwelchcont).
The function sets an initial Hidden Markov Model object with initial set of model parameters. It returns the object of class ContObservHMM that can be analysed with Baum-Welch (function baumwelchcont) and Viterbi algorithms (viterbicont).
● Data Source:
CranContrib
● Keywords: Hidden Markov Model, Time series
● Alias: hmmsetcont, plot.ContObservHMM, print.ContObservHMM, summary.ContObservHMM
●
1 images
The function simulates (i) two observation processes that correspond to the distributions of the two states in an HMM, (ii) the underlying Markov process, and (iii) an observation process that correspond to an HMM in terms of both the underlying Markov and observations processes. The function uses the last-iteration parameters, when the HMM-object was modified by function baumwelchcont previously.
The function plots the Gaussian probability density functions from the means and variances of the whole data set, the two sub-sets corresponding to the two Markov chain states, and additionally from the HMM model (i.e. the means and variances taken form the last Baum-Welch iteration).
The function performs Baum-Welch algorithm with Gaussian PDFs (Baum et al, 1970; Rabiner, 1989). It allows to control the model parameters after each iteration, and accumulates the information on the model evolution. The intended use is to perform repeated executions and to save the returned object into the argument object (see examples below).
● Data Source:
CranContrib
● Keywords: Baum-Welch, Hidden Markov Model
● Alias: baumwelchcont
●
1 images
The package includes the functions designed to analyse continuous observations processes with the Hidden Markov Model approach. They include Baum-Welch and Viterbi algorithms and additional visualisation functions. The observations are assumed to have Gaussian distribution and to be weakly stationary processes. The package was created for analyses of financial time series, but can also be applied to any continuous observations processes.