Re: [R] optim() non-finite finite-difference value

From: Spencer Graves <spencer.graves_at_pdf.com>
Date: Sun 24 Apr 2005 - 02:36:54 EST

          I just encountered that myself using optim in S-Plus 6.1. For a certain set of parameter values, the function I wanted to minimize returned NaN (not a number). To fix this problem, I set a trace with something like options(trace=99) using, I think, method = "CG". This printed out all the parameter values tested including the one that generated the offense. I then took that set of parameter values to the function to be optimized and figured out a way to limit the range of the computations. I used a construct like the following:

          good.num.limit = c(.Machine$double.xmin, .Machine$double.xmax)^(1/3)

Then I constrained certain positive numbers to lie within these bounds, thereby avoiding 0 and Inf.

          Then I tested the result using method = "Nelder-Mead", "BFGS", and "CG". For my particular installation, for some unknown reason, Nelder-Mead bombed S-Plus, BFGS returned a nonsense answer, but CG was sensible. Just a few minutes ago, I restarted my (several hour) job. I think it should be more likely to finish now.

	  hope this helps.
	  spencer graves

Werner Bier wrote:
> Dear all,
>
> I am using the optim() function which it stops with the following error messagge:
>
> error in optim(...) non-finite finite-difference value
>
> I was wondering if somebody might suggest me a way to fix it please.
>
> Thanks in advance to all of you.
> Kind regards,
> Tom
>
> __________________________________________________
>
>
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help@stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html



R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Sun Apr 24 02:39:43 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:31:25 EST