hmmDemo: HMM computation demo functions

In marked: Mark-Recapture Analysis for Survival and Abundance Estimation

HMM computation demo functions

Description

Uses fitted hmm model to construct HMM state vectors alpha and phi for demonstration purposes

Usage

hmmDemo(object, ddl = NULL, state.names = NULL, obs.names = NULL)

Arguments

object	fitted hmm model
ddl	design dat list; if NULL it is created
state.names	names for states used to label output; if NULL uses strata.labels + Dead state
obs.names	names for observations used to label output; if NULL uses ObsLevels

Value

hmm demo list which includes 1) lnl - the log-likelihood value, 2) alpha - forward probabilities, 3) beta - backward probabilities, 4) phi - scaled forward probabilities, 5) v- intermediate calculation for phi, 6) dmat - 3-d array with observation probability matrix for each occasion, 7) gamma - 3-d array with state transition probability matrix for each occasion, 8) stateprob - predicted state probabilities, 9) local_decode - state predictions for each occasion and individual, 10) global_decode - state predictions for entire sequence for each individual.

Author(s)

Jeff Laake

Examples

# This example is excluded from testing to reduce package check time
# cormack-jolly-seber model
data(dipper)
mod=crm(dipper,model="hmmcjs")
x=hmmDemo(mod,state.names=c("Alive","Dead"),obs.names=c("Missed","Seen"))
par(mfrow=c(2,1))
barplot(t(x$alpha[45,,]),beside=TRUE,names.arg=x$chforwardstrings)
barplot(t(x$phi[45,,]),beside=TRUE,names.arg=x$chforwardstrings)
# multi-state example showing state predictions
data(mstrata)
mod=crm(mstrata,model="hmmMSCJS",strata.labels=c("A","B","C"))
#' x=hmmDemo(mod)
# state predictions are normalized by likelihood value which = rowSums(alpha*beta)
cat(paste("\nrowsums = ",rowSums(x$alpha[45,,]*x$beta[45,,],na.rm=TRUE)[2],
   "which matches likelihood value",exp(x$lnl[45]),"\n"))
# state predictions given the data
x$stateprob[45,,]

marked documentation built on Oct. 19, 2023, 5:06 p.m.