Closed willwerscheid closed 3 years ago
It doesn't do well for large x due to the stepmax setting for nlm (which I need to avoid overflow errors). I think I'll have to add pre- and postscaling functions so that optimization is done over (for example) x / median(s)
after scaling, set default stepmax (nlm) to 1 and rmax (trust) to 1; otherwise they fail too easily
we haven't sufficiently tested scaling