Skip to contents

Most likely state path of an hmm fit (Viterbi)

Source: R/hmm.R
hmm_viterbi.Rd

Global decoding: the single state sequence with the highest posterior probability, per sequence, by the Viterbi recursion. This is not the same as taking the most likely state at each row separately (max.col(hmm_probs(fit)), local decoding), and the two can disagree

  • only the Viterbi path is guaranteed to be a possible path under the transition matrix.

Usage

hmm_viterbi(fit)

Arguments

fit

A frmtmb_fit with an hmm() family.

Value

An integer vector, one state index per row of the data.

Details

State LABELS are arbitrary: the likelihood is invariant to permuting the states, so which run calls a state "1" is not meaningful across fits. Compare paths after matching states by their fitted means.

See also

hmm_probs(), hmm()

Examples

set.seed(12)
dd <- data.frame(id = 1, t = 1:120)
s <- integer(120); s[1] <- 1L
G <- matrix(c(0.92, 0.08, 0.15, 0.85), 2, 2, byrow = TRUE)
for (t in 2:120) s[t] <- sample.int(2, 1, prob = G[s[t - 1], ])
dd$y <- rnorm(120, c(0, 3)[s], 0.5)
fit <- frm(bf(y ~ 1),
           family = hmm(K = 2, gaussian(), time = t, group = id),
           data = dd)
v <- hmm_viterbi(fit)
table(v, truth = s)
#>    truth
#> v    1  2
#>   1 58  0
#>   2  0 62