Subject: LBFGS for use in large maximum likelihood problem
From: Robert Dodier
Date: Fri, 22 Aug 2008 18:26:42 -0600
On Fri, Aug 22, 2008 at 2:04 PM, dlakelan <dlakelan at street-artists.org> wrote:
> "`optimize' has met up with a special form - answer may be wrong."
Dunno what that's about.
> Also, can I use optimize on the figure of merit expression for LBFGS to get
> faster results? Is there a way to hand LBFGS an expression for the gradient
> (also optimized)? The manual doesn't mention anything, but I thought there
> was some work done on this?
Yes, you can provide the gradient to LBFGS. There are some examples
in the comment header of share/lbfgs/maxima-lbfgs.lisp. I guess I should
revise the reference manual. Let me know how it works out for you.
best
Robert Dodier