Hi all,
Hi,I haven't tested Powell on these (or have time to now unfortunately), but I did previously add a test bench of optimization problems in https://github.com/scipy/scipy/pull/14579 if it's useful.Best,Sara
On Mar 6, 2023, at 4:17 AM, Matthew Brett <matthew.brett@lis.ac.uk> wrote:
Hi,
Sorry to keep on at this one, but it is worrying me.
Do we have anyone reading or known to us who can give a fairly
definitive answer to:
On Thu, Mar 2, 2023 at 10:01 PM Matthew Brett <matthew.brett@lis.ac.uk> wrote:
[snip]
? Is it really true that the modification assists Powell in handling
long skinny valleys? If so, does it work less well in other
situations? Has anyone compared the algorithms across a good range of
problems?
Cheers,
Matthew
Michael J. D. Powell. The newuoa software for unconstrained optimization without
derivatives. In In: Di Pillo G., Roma M. (eds) Large-Scale Nonlinear Optimization,
volume 83, pages 1247–1293. Springer, Boston, MA, 2006.)
The implementation of scipy is pretty old and doesn't incorporoate the changes from the NEWUOA algorihtm from 2006. This one is pretty successfull for non-noisy convex optimization problems.
Luis Miguel Rios and Nikolaos V. Sahinidis. Derivative-free optimization: a review of
algorithms and comparison of software implementations. volume 53, pages 1247–
1293, 2012.Jorge J. Moré and Stefan M. Wild. Benchmarking derivative-free optimization algorithms.
volume 20, pages 172–191, 200.
A narrow valley is problematic, if the step width in the direction made is larger than the valley. A step width parameter or the direc parameter (or scaling) can help
I had problems with COBYLA and even more with NEWUOA (Quadratic optimization) for my noisy function to optimize. I think that the strength is on convex optimization problems.
Best,
Chris
_______________________________________________
SciPy-Dev mailing list -- scipy-dev@python.org
To unsubscribe send an email to scipy-dev-leave@python.org
https://mail.python.org/mailman3/lists/scipy-dev.python.org/
Member address: sarafridov@gmail.com
_______________________________________________
SciPy-Dev mailing list -- scipy-dev@python.org
To unsubscribe send an email to scipy-dev-leave@python.org
https://mail.python.org/mailman3/lists/scipy-dev.python.org/
Member address: christoph.schranz@salzburgresearch.at