On Mon, 4 Aug 2003, Jean Fan wrote: > Dear Professor Ripley, > > I'm little confused of your reply: do you mean that GRG would > not be a standard optimization algorithm, so it couldn't be > better than what exist in R ?
Not at all. I meant what I actually said (and I said nothing about which was better). To spell it out: First optim() has five different methods, so Excel's could be at most essentially the same as *one* of them. Second, none of them are called GRG2 by their authors. Third, GRG2 by that name does not appear in the index of my books. > I'm not a specialist of numerical optimization algorithms, > but it seems that GRG is actually implemented in several > specialized optimisation toolbox (sure generally commercial), > not only the limited one in Excel. > And with google, search "GRG generalized reduced gradient" is > giving 424 links. So now you will be able to work out how it differs, it at all, from the methods of R (which are fully documented). But looking at just the first reference suggests that GRG is not intended for unconstrained or box-constrained problems, those covered by optim(), and that it is a class of methods rather than an actual algorithm. > > -- > Fan > > > I've found that the discussions are interesting, generally > > speaking, peoples seem equally confident on R's optim/nlm and > > Excel's solver. > > > > The authors of the algorithm GRG2 (Generalized Reduced Gradient) > > are not cited in the documentation of optim(), so I'm wondering if > > the optimization algorithm implemented in Excel is "fondamentally" > > the same than that in R ? > > I don't suppose Excel cites the method*s* used in optim() either, > but GRG2 is not in the index of my copies of any of the standard texts on numerical > optimization. -- Brian D. Ripley, [EMAIL PROTECTED] 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 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595 ______________________________________________ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help
