On Wed, 24 Nov 1999, Jim Lindsey wrote:
> OK I figured it out. The attributes have to be on the value returned,
> not on the function itself.
Yes, ms-style (and not surprising given that Doug Bates' hand was in both
implementations). I would prefer to be able to specify separate
functions for the gradient and hessian, at least as an alternative.
> I tried a linear (not log linear) regression with Poisson, using
> only the gradient, not the hessian. It took one extra iteration (13
> instead of 12) and gave twice as many warnings (28 instead of 16)
> compared to not supplying the gradient. The answers are identical to 5
> decimal places.
As a different example, I am currently implementing polr from MASS in R.
Without gradients, it takes a few hundred iterations if it converges at
all. With gradients, it takes less than 10 and seems pretty reliable.
I have several problems where gradients are essential, so it great to have
the facility in nlm for those.
Brian
--
Brian D. Ripley, ripley at stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._