[Rd] Optimization in R

Petr Savicky savicky at cs.cas.cz
Mon Aug 6 08:48:19 CEST 2007

I am sorry for omitting a citation in my previous post.
The complete message is as follows (my text unchanged). PS

I would like to add a remark and a question.


There is a part of R, which allows the user to select
among several methods for the same task and also to add
his own C code: random number generation. However, the interface
for optimization is more complex. In my opinion, looking
for a unified interface for this is desirable, but it is
a research problem, not a suggestion for an immediate
code modification.


Is there a way how to optimize a function written in C
using optim? This would be very useful, if the optimization
needs a lot of iterations. This may be done by defining 
an R function, which does nothing more than calling .C with
appropriate parameters,
but this looses efficiency. A more efficient solution
could be adding a specified entry point (or several, if derivatives
are also available), similar as in the user defined random number
generator. Then, a parameter of optim could control, whether
the function to be optimized is fn or the C entry point.

Petr Savicky.

On Sat, Aug 04, 2007 at 06:56:47PM -0400, Gabor Grothendieck wrote:
> I don't have an example of that but that does not make it less
> desirable.  If one wants to use method 1, 2 or 3 then one can
> use optim with a method= but if one wants to use methods 4
> or 5 then one must use an entirely different function.  Surely
> it would be better to be consistent from the user's viewpoint
> and allow all of them to work consistently through the same
> interface.
> On 8/4/07, Duncan Murdoch <murdoch at stats.uwo.ca> wrote:
> > On 04/08/2007 2:53 PM, Gabor Grothendieck wrote:
> > > The example of generic functions.
> >
> > Show me an example where we have a list of ways to do a calculation
> > passed as an argument (analogous to the method argument of optim), where
> > the user is allowed to add his own function to the list.
> >
> > Duncan Murdoch
> > >

More information about the R-devel mailing list