[Rd] bounds violations, infinite loops in optim/L-BFGS-B (PR#671)

[ripley@stats.ox.ac.uk]

> From: bolker@zoo.ufl.edu
> Date: Tue, 26 Sep 2000 16:14:13 +0200 (MET DST)
> 
>   I'm having some trouble with optim(method="L-BFGS-B"),
> and I'm not sure I have the ability to track down and fix
> what seem to be bugs within optim().
>   I'm bootstrapping an original data set and fitting a model
> to each bootstrapped data set.  For some bootstrapped samples,
> optim() sets negative parameter values (despite the fact that
> I have explicitly set non-zero lower bounds on the parameters)
> and chokes.  For other samples, it appears to get into an infinite
> loop (ignoring the finite value of maxit).
>  
>   Functions and samples that provoke the problem are below: I
> was able to reproduce the problems running this with --vanilla.
> 
>   Not exactly a bug but: there is no tracing information built
> into L-BFGS-B (setting trace=TRUE has no effect).

It's a feature, on the TODO list to be added one day.

>   Also, a general question: are both nlm() and optim() going to
> be around indefinitely?  Should I be using one or the other?

There are no plans to remove either.  `indefinitely' is not part
of the non-warranty.
 
>   For the moment I've managed to (sort of) get around the problem
> with try(), but the infinite loops are a real nuisance ...

On Solaris, I get neither negative values nor an infinite loop
on your examples. 

On Linux RH6.2/gcc 2.95.2 I get

NA 1 NA 1 1 1 1 0.6831783 1 0.6831783 1 NA 
NA -Inf NA -Inf -Inf -Inf -Inf -160.2173 -Inf -107.0805 -Inf NA 
Error in optim(c(min(boot.total) - 1, 100, 1), nllfun2g.boot, lower = rep(fuzz,  
: 
        L-BFGS-B needs finite values of fn

for the first, and probably a loop for the second.  I think this is
the usual problem with inconsistent internal precision on Linux
compilers, so try compiling optim.c with -ffloat-store to make gcc
IEEE-compliant.  (At least, that's what the Linux gcc man page says.)

Can you provide some evidence for negative parameter values?  That's
not supposed to happen, but as I rarely fail to supply derivatives,
I have not tested it much.

Finally, I think you have omitted to supply scaling consts for your problem,
and optim will work a lot better if you do, as it will also do if
you can supply analytical derivatives.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._