As I'm at least partly responsible for CG in optim, and packager of Rcgmin, I'll recommend the latter based on experience since it was introduced. I've so far seen no example where CG does better than Rcgmin, though I'm sure there are cases to be found. However, Ben is right that if ADMB does so well (it uses effectively analytic derivatives), then use it. Rcgmin really wants you to provide gradient code, and that is work. JN
On 07/14/2011 06:00 AM, r-help-request at r-project.org wrote:
Message: 85 Date: Wed, 13 Jul 2011 20:20:47 +0000 From: Ben Bolker <bbolker at gmail.com> To: <r-help at stat.math.ethz.ch> Subject: Re: [R] Very slow optim() Message-ID: <loom.20110713T221924-956 at post.gmane.org> Content-Type: text/plain; charset="utf-8" Hamazaki, Hamachan (DFG <toshihide.hamazaki <at> alaska.gov> writes:
Dear list, I am using optim() function to MLE ~55 parameters, but it is very slow to
converge (~ 25 min), whereas I can do
the same in ~1 sec. using ADMB, and ~10 sec using MS EXCEL Solver. Are there any tricks to speed up? Are there better optimization functions?
There's absolutely no way to tell without knowing more about your code. You
might try method="CG":
Method ?"CG"? is a conjugate gradients method based on that by
Fletcher and Reeves (1964) (but with the option of Polak-Ribiere
or Beale-Sorenson updates). Conjugate gradient methods will
generally be more fragile than the BFGS method, but as they do not
store a matrix they may be successful in much larger optimization
problems.
If ADMB works better, why not use it? You can use the R2admb
package (on R forge) to wrap your ADMB calls in R code, if you
prefer that workflow.
Ben