[R] BFGS and Neldear-Mead

[Ravi Varadhan]

It is important to pay attention to the convergence message. For example, if
the convergence is due to reaching maximum iteration limit (this is often
the case with Nelder-Mead), then you should increase the maxit from 500 to
1500 or something like that.

Users generally have trouble interpreting the convergence messages.  This is
one of the reasons that we have implemented explicit convergence diagnostics
in "optimx" through first- and second-order KKT conditions. So, you might
want to re-run your problem with "optimx" and check the KKT conditions. 

If there are multiple optima, you can get different answers from properly
converged iterations.

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: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
Behalf Of Petr Savicky
Sent: Friday, March 18, 2011 5:25 PM
To: r-help at r-project.org
Subject: Re: [R] BFGS and Neldear-Mead

On Fri, Mar 18, 2011 at 11:27:58AM -0700, LC-Bea wrote:
> Hello!
> 
> I´m doing this:
> 
> A<-optim(vv,pen.wlsv,method="BFGS")
> B<-optim(vv,pen.wlsv,method="Nelder-Mead")
> 
> the function is the same
> and the initial values too
> 
> the estimate values are different, very different, at the first decimal
> 
> is there an error?

Not necessarily.

> or maybe its because of my data, or funtion to minimize?
> i thought that the values must be almost equal, that the big difference
> between the two algoritms is in the time that they use to find a solution

If the function has several local minima, then different methods
may found different ones.

Can you send more detail about the minimized function?

Petr Savicky.

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.