Review of optimization problems inspired by the NIST nonlinear least squares test problems

In NISTopt: Nonlinear least squares examples from NIST in form of functions for optim() and optimx()

Motivation

http://www.itl.nist.gov/div898/strd/nls/nls_info.shtml presents a number of nonlinear regression (nonlinear least squares) problems that are more or less difficult to solve numerically. Doug Bates prepared an R package NISTnls which adapted these problems to R and tested them with the nls() function of which he was a major author. I built a related package to allow the same problems to be approached as unconstrained function minimization problems, which I named NISTopt. However, this package was, when I prepared it in 2009, incomplete, particularly in respect to the manuals (Rd files). This vignette is an attempt to better document that effort and in the process force a review of the package.

Note that the problems here are NOT good illustrations of tasks to which the various function minimization (also called "optimization", though that generally implies that there are also constraints) tools should be applied. However, they do suggest limits of performance of these tools.

Approach

?? We need to be able to set things up so that a separate command is not needed for each start or example. Need to think about this. Also the "setup" approach doesn't seem too helpful.

NISTopt documentation built on July 25, 2017, 3:01 p.m.