Posterior state probabilities of an hmm fit

The smoothed occupancy P(S_t = k | y) for every row of the data, from a forward-backward pass at the estimates. This is the frmtmb::mixture() analog mixture_probs(), one rung up: an HMM's per-row state probability conditions on the WHOLE sequence, not on that row alone, which is why it needs a backward pass and cannot come out of the family's density.

Usage

hmm_probs(fit)

Arguments

fit

A frmtmb_fit with an hmm() family.

Value

An n x K matrix of probabilities whose rows sum to one, with columns named state1 ... stateK.

Details

Under a random-effect model the probabilities are conditional on the random-effect modes, as every other post-fit quantity in the package is. Rows whose response is NA still get a probability: the chain passes through them and the neighboring observations inform the state.

See also

hmm_viterbi(), 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)
p <- hmm_probs(fit)
head(p)
#>            state1       state2
#> [1,] 9.999989e-01 1.121901e-06
#> [2,] 1.000000e+00 1.786469e-08
#> [3,] 9.999676e-01 3.235686e-05
#> [4,] 4.209560e-08 1.000000e+00
#> [5,] 2.937353e-10 1.000000e+00
#> [6,] 8.655297e-10 1.000000e+00
rowSums(p)[1:3]
#> [1] 1 1 1
mean(max.col(p) == s)
#> [1] 1