Re: Optimization of large nonlinear models

    I would strongly recommend that you take a look at the RTMB package 
... it's a very low-threshold way to get autodiff gradients and near-C++ 
speed from your objective function.

https://cran.r-project.org/web/packages/RTMB/vignettes/RTMB-introduction.html

https://kaskr.r-universe.dev/articles/RTMB/RTMB-tips.html

On 12/30/25 07:32, Ruben Roa Ureta via R-help wrote:
> Thanks Ben, John, Richard
> You confirm my experience: above certain amount of RAM there are no improvements in speed.
> It cost me 3K euros to learn that, but the machine will be good for image processing.
> So for largish nonlinear models, over a 100 parameters, optimization in R would best be done with analytical gradients or by calling code written in C++ (autodiff TMB or ADMB) or FORTRAN.
> It seems to be the best option, as parallelization of computations with the obj. function would be entering unknown territory.
> For intermediate problems with 50-100 parameters, even a laptop with 36 GB RAM finish in a few hours, which is good enough for me.
> Regards.
> R.
> 
> ---
> Ruben H. Roa-Ureta, Ph. D.
> Consultant in Statistical Modeling
> ORCID ID 0000-0002-9620-5224
> 
> ______________________________________________
> [email protected] mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide https://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

--