[Depmix-commits] r517 - pkg/depmixS4/man

[noreply at r-forge.r-project.org]

Author: ingmarvisser
Date: 2012-06-12 16:31:19 +0200 (Tue, 12 Jun 2012)
New Revision: 517

Modified:
   pkg/depmixS4/man/em.control.Rd
Log:
Updated help page of em.control to explain the procedure for generating starting values using the Dirichlet distribution.

Modified: pkg/depmixS4/man/em.control.Rd
===================================================================
--- pkg/depmixS4/man/em.control.Rd	2012-06-12 13:14:18 UTC (rev 516)
+++ pkg/depmixS4/man/em.control.Rd	2012-06-12 14:31:19 UTC (rev 517)
@@ -31,24 +31,35 @@
 The argument \code{crit} sets the convergence criterion to either the
 relative change in the log-likelihood or the absolute change in the
 log-likelihood.  The relative likelihood criterion (the default) assumes
-convergence on iteration \eqn{i}{i} when \eqn{\frac{\log L(i) - \log
-L(i-1)}{\log L(i-1)} < tol}{\frac{\log L(i) - \log L(i-1)}{\log L(i-1)} <
-tol}.  The absolute likelihood criterion assumes convergence on iteration
-\eqn{i}{i} when \eqn{\log L(i) - \log L(i-1) < tol}{(logLik(i) -
-logLik(i-1)) < tol}.  Use \code{crit="absolute"} to invoke the latter
+convergence on iteration \eqn{i}{i} when 
+\eqn{\frac{\log L(i) - \log L(i-1)}{\log L(i-1)} < tol}{ (log L(i) - log L(i-1))/(log L(i-1)) < tol}.  
+The absolute likelihood criterion assumes convergence on iteration
+\eqn{i}{i} when \eqn{\log L(i) - \log L(i-1) < tol}{(log L(i) - log L(i-1)) < tol}.  
+Use \code{crit="absolute"} to invoke the latter
 convergence criterion.  Note that in that case, optimal values of the 
-tolerance parameter \code{tol} scale with the value of the log-likelihood. 
+tolerance parameter \code{tol} scale with the value of the log-likelihood (and these are not changed automagically). 
 
 Argument \code{random.start} This is used for a (limited) random
 initialization of the parameters.  In particular, if
 \code{random.start=TRUE}, the (posterior) state probabilities are
-randomized at iteration 0 (using a uniform distribution).  Random
-initialization is useful when no initial parameters can be given to
-distinguish between the states.  It is also useful for repeated estimation
-from different starting values.
+randomized at iteration 0 (using a uniform distribution), i.e. the 
+\eqn{\gamma} variables (Rabiner, 1989) are sampled from the Dirichlet
+distribution with a (currently fixed) value of
+\eqn{\alpha=0.1}; this results in values for each row of \eqn{\gamma}
+that are quite close to zero and one; note that when these values are
+chosen at zero and one, the initialization is similar to that used in
+\code{kmeans}.  Random initialization is useful when no initial parameters can be
+given to distinguish between the states.  It is also useful for repeated
+estimation from different starting values.
 
 }
 
+\references{
+	Lawrence R. Rabiner (1989).  A tutorial on hidden Markov models and
+	selected applications in speech recognition.  \emph{Proceedings of
+	IEEE}, 77-2, p.  267-295.
+}
+
 \value{
 	
 	\code{em.control} returns a list of the control parameters.