Hi, In my earlier post I eluded to a situation where that would be useful. In nlme, there is a choice of optimizers, minpack.lm has Levenberg-Marquardt, while nlminb has the port routines. For the same starting values, different optimizers will present different solutions, having a common interface would make fitting with multiple optimizers very attractive.
Also the inverse Hessian would be useful, for cases where the Hessian is ill conditioned a little regularization goes a long way. I believe the package accuracy has a nice solution. Nicholas On 04/08/2007 2:23 PM, Gabor Grothendieck wrote: > For the same reason that generic functions exist. They don't have > a lot of common code but it makes easier to use. Perhaps the argument > is not as strong here since the class tends to be implicit whereas the > method is explicit but it would still be a convenience. Can you give other examples where we do this? The ones I can think of (graphics drivers and finalizers) involve a large amount of common machinery that it's difficult or impossible for the user to duplicate. That's not the case here. Duncan Murdoch ______________________________________________ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
