[R] EM algorithm to fit circular mix of uniform+Von Mises

[Ravi Varadhan]

Hi Mike,

I have an EM algorithm code for a binary von Mises mixture with 5 parameters: mixing proportion (p), 2 locations (m1, m2), and 2 dispersion parameters (k1, k2).  Of course, your model is nested within this one, where k1=Inf, m1 = arbitrary, m2=0.  You should be able to modify my code easily to fit this reduced model.  This will also allow you to perform a likelihood ratio test for whether the 5-parameter model is better than the 2-parameter model (approximately chi-squared with 3 d.o.f.).  

Alternatively, you can directly estimate the 2-parameter model by maximizing the log-likelihood subject to constraints on the 2 parameters.  This is quite easy to do, but will not allow you comparison with a more general model.

Ravi.
____________________________________________________________________

Ravi Varadhan, Ph.D.
Assistant Professor,
Division of Geriatric Medicine and Gerontology
School of Medicine
Johns Hopkins University

Ph. (410) 502-2619
email: rvaradhan at jhmi.edu


----- Original Message -----
From: Mike Lawrence <Mike.Lawrence at dal.ca>
Date: Saturday, November 7, 2009 6:38 pm
Subject: [R] EM algorithm to fit circular mix of uniform+Von Mises
To: r-help at stat.math.ethz.ch


> Hi all,
> 
> I'm curious if anyone has coded an Expectation-Maximization algorithm
> that could help me model some circular data I have. I'd like to model
> it as a mixture of uniform and Von Mises centered on 0, so the only
> free parameters is the mixing proportion and the kappa of the Von
> Mises. I couldn't find anything in the contributed packages that
> seemed to suit this purpose. Any pointers would be greatly
> appreciated!
> 
> Cheers,
> 
> Mike
> 
> -- 
> Mike Lawrence
> Graduate Student
> Department of Psychology
> Dalhousie University
> 
> Looking to arrange a meeting? Check my public calendar:
> 
> 
> ~ Certainty is folly... I think. ~
> 
> ______________________________________________
> R-help at r-project.org mailing list
> 
> PLEASE do read the posting guide 
> and provide commented, minimal, self-contained, reproducible code.