s-news
[Top] [All Lists]

EM with nlminb?

To: s-news@lists.biostat.wustl.edu
Subject: EM with nlminb?
From: jagadish rangrej <jrangrej@yahoo.com>
Date: Tue, 30 Mar 2004 08:46:43 -0800 (PST)
In-reply-to: <000101c415d0$c34fe330$641a7680@meth.hhdev.psu.edu>
Dear all!!

I have an issue in finding EM estimates, I have setup
and EM algorithm for my problem, but it finds
different optimal solutions when starting with
different starting-points of the parameters.

I was curious to know if EM by Dempster and Rubin
asures of convergence to global optimal solution? or
does it depends on the likelihood function? 


-My function is based on two stage model with 1st
stage being multinomial and the 2nd stage being
binomial given (n,y) (where n_i's are missing) . and I
am trying to optimize the combined likelihood in the
M-step using "nlminb".  E-step consists of just
estimating the n_i's using the expected values in
multinomial distribution.

Any suggestion or reference will be highly
appreciated. I am sorry for being little out of
context question.

Many thanks in advance,
Jag


__________________________________
Do you Yahoo!?
Yahoo! Finance Tax Center - File online. File on time.
http://taxes.yahoo.com/filing.html

<Prev in Thread] Current Thread [Next in Thread>