hhsmmfit function

hhsmm model fit

Fits a hidden hybrid Markov-semi-Markov model to a data of class "hhsmmdata" and using an initial model created by hhsmmspec or initialize_model

hhsmmfit(
  x,
  model,
  mstep = NULL,
  ...,
  M = NA,
  par = list(maxit = 100, lock.transition = FALSE, lock.d = FALSE, lock.init = FALSE,
    graphical = FALSE, verbose = TRUE)
)

Arguments

• x: a data of class "hhsmmdata", which can also contain missing values (NA or NaN)

• model: an initial model created by hhsmm.spec or initialize_model

• mstep: the M step function for the EM algorithm, which also can be given in the model

• ...: additional parameters for the dens.emission and mstep functions

• M: the maximum duration in each state

• par: additional list of control parameters of the hhsmmfit function including the following items:

  • maxit the maximum number of iterations for the EM algorithm
  • lock.transition logical. if TRUE the transition matrix will not be updated through the EM algorithm
  • lock.d logical. if TRUE the sojourn probability matrix d will not be updated through the EM algorithm
  • lock.init logical. if TRUE the initial probability vector will not be updated through the EM algorithm
  • graphical logical. if TRUE a plot of the sojourn probabilities will be plotted through the EM algorithm
  • verbose logical. if TRUE the outputs will be printed

Returns

a list of class "hhsmm" containing the following items:

• loglike the log-likelihood of the fitted model
• AIC the Akaike information criterion of the fitted model
• BIC the Bayesian information criterion of the fitted model
• model the fitted model
• estep_variables the E step (forward-backward) probabilities of the final iteration of the EM algorithm
• M the maximum duration in each state
• J the number of states
• NN the vector of sequence lengths
• f the emission probability density function
• mstep the M step function of the EM algorithm
• yhat the estimated sequence of states

Examples

J <- 3
initial <- c(1, 0, 0)
semi <- c(FALSE, TRUE, FALSE)
P <- matrix(c(0.8, 0.1, 0.1, 0.5, 0, 0.5, 0.1, 0.2, 0.7), nrow = J, 
byrow = TRUE)
par <- list(mu = list(list(7, 8), list(10, 9, 11), list(12, 14)),
sigma = list(list(3.8, 4.9), list(4.3, 4.2, 5.4), list(4.5, 6.1)),
mix.p = list(c(0.3, 0.7), c(0.2, 0.3, 0.5), c(0.5, 0.5)))
sojourn <- list(shape = c(0, 3, 0), scale = c(0, 10, 0), type = "gamma")
model <- hhsmmspec(init = initial, transition = P, parms.emis = par,
dens.emis = dmixmvnorm, sojourn = sojourn, semi = semi)
train <- simulate(model, nsim = c(10, 8, 8, 18), seed = 1234, 
remission = rmixmvnorm)
clus = initial_cluster(train, nstate = 3, nmix = c(2 ,2, 2),ltr = FALSE,
final.absorb = FALSE, verbose = TRUE)
initmodel1 = initialize_model(clus = clus, sojourn = "gamma", 
M = max(train$N), semi = semi)
fit1 = hhsmmfit(x = train, model = initmodel1, M = max(train$N))

References

Guedon, Y. (2005). Hidden hybrid Markov/semi-Markov chains. Computational statistics and Data analysis, 49(3), 663-688.

OConnell, J., & Hojsgaard, S. (2011). Hidden semi Markov models for multiple observation sequences: The mhsmm package for R. Journal of Statistical Software, 39(4), 1-22.

Author(s)

Morteza Amini, morteza.amini@ut.ac.ir , Afarin Bayat, aftbayat@gmail.com