BaumWelchT: BaumWelchT HMM

In gui11aume/HMMt: HMM with Student's t emissions

Description

Estimate the HMM with emissions distributed as multi-dimensional correlated T variables

Usage

  BaumWelchT(x, series.length, m = 2, Q, mu, S, nu = TRUE,
    model, initial.prob, maxiter = 500, overflow = 1e-09,
    num.inst = 1e-09, tol = 1e-05, dig = 3)

Arguments

x	matrix of observations.
series.length	length of independent blocks in matrix x.
m	number of states.
Q	transition matrix.
mu	means of the emissions in different states.
S	correlation matrix of the emissions.
nu	degrees of freedom (in all states).
model	a model matrix for parameter estimation.
initial.prob	state probability at the start of series.
maxiter	maximum number of iterations before returning.
overflow	the probability ratio in case of numeric overflow.
num.inst	threshold for no update (see E.step()).
tol	threshold to difference in log-likelihood before returning.
dig	numeric precision for parameter estimation.

Value

the estimate of the HMM with emissions distributed as multi dimensional correlated T variables.

Author(s)

Guillaume Filion. date: June 17, 2011.

Examples

library(HMMt)
x <- c(rt(1000, df=3), rt(1000, df=3)+1)

# x has a t distribution with a jump at position 1001.
plot(x, type = 'l')

# Check the output of BaumWelchT.
BaumWelchT(x)

# See that it usually finds the transition.
lines(BaumWelchT(x)$ViterbiPath-1, col=2)

gui11aume/HMMt documentation built on May 5, 2019, 9:19 a.m.