OBIC calculates BIC, AIC, approximate log-likelihood and plots the
log-likelihood for all iterations. The log-likelihood plot should
be flat to show convergence to a stationary distribution. Minimize
the AIC and BIC for the *best* model and maximize PLS. The log likelihood is approximate in that
it is calculated by marginalizing over the current chain of hidden states instead
of using a recurrsive algorithm to compute it; every iterations produces an estimation
of the log-likelihood. If yhold is provided the preditive log score (PLS) is also
given.
Usage
OBIC(nhmmobj, outfile = NULL)
Arguments
nhmmobj
an object created from the NHMM function
outfile
a directory to put the .png plot
Value
BIC
output: AIC, BIC, PLS [if yhold data was provided], log-likelihood to the GUI and a plot of the log-likelihood
Details
Predictive Log Score: mean(log( E(p(yhold|...))) The expectation is over all of the
iterations of the algorithm. And the mean is over the pT count of yhold. The scale of
the PLS is in the unit of t (usually days).