If your problem is small enough just use a grid of starting values
and run your optimization on each one and then take the best.
On 3/6/07, Dae-Jin Lee <lee.daejin at gmail.com>
wrote:> Hi all !
>
> I've been trying to maximize a likelihood using optim( ) function, but
it
> seems that the function has several local maxima. I've tried in my
algorithm
> with different starting values and depending on them "optim"
obtains
> different results...
>
> I use the "L-BFGS-B" method setting the lower values as 1e-06,
because my
> parameters must be strictly positive. Also tried a log() transformation to
> ensure that my parameters are positive. Don't know if this is useful in
this
> case... (also with notLog and notExp functions of mgcv package)
>
> the function nlminb( ) also have the same problems.
>
>
> ?Is there any thing I'm not considering? I mean other methods instead
of
> "L-BFGS-B"?
>
> How can I do to take a strategy to begin with "good" starting
values?
>
>
> Thanks in advance
>
> Dae-Jin
>
> PS: I'm trying to fit a mixed model with REML and several random
effects, so
> I maximize over several parameters...
>
> [[alternative HTML version deleted]]
>
>
> ______________________________________________
> R-help at 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
> and provide commented, minimal, self-contained, reproducible code.
>
>