From QERM Wiki
Jump to: navigation, search
Performance of optim vs. nlminb optimizing function in R used to compare the final objective function (OF) on 120 trials of each method on the same (5-parameter) problem with a small (too small) data set. In each trial, starting values were varied randomly from a good vector. All trials in which either algorithm arrived at OF > 1.1 * OF(best) were discarded.


Most people seeking to do optimization in R eventually find optim. However some caution should be used. First, buried in the help for optim is this note: "optim will work with one-dimensional pars, but the default method does not work well (and will warn). Use optimize instead."

For those looking for a multi-criteria optimizer, mco is a good option, and frequently finds better solutions than optim for single criteria problems.

A comparison by Mike Prager of the NMFS Southeast Fisheries Science Center found that the function nlminb outperformed optim as shown in the figure at right.

It should also be noted that for more complex minimization problems AD Model Builder has been shown to be superior to R.

See Also

Optimization Task View

R tips

AD Model Builder

Personal tools