On Fri, 2013-12-13 at 14:48 +0100, Yury V. Zaytsev wrote:
> 
> Is there any built-in facility in NLopt for L1-norm regularization
> that I've missed? Have anyone got L1-norm regularization to work with
> NLopt gradient-based methods before?

After some more googling, I have found that the problem is apparently
solved by a modification of the original L-BFGS-b called Orthant-wise
limited-memory quasi-Newton (OWL-QN):

http://en.wikipedia.org/wiki/Limited-memory_BFGS#OWL-QN

As it appears, there is free implementation out there:

https://github.com/chokkan/liblbfgs (MIT license)

Therefore, I guess I have to re-cast my question as "is that possible to
add this feature to the NLopt implementation"?

Sadly, this library doesn't seem to support even box contraints, which
NLopt can handle with ease...

Thanks,

-- 
Sincerely yours,
Yury V. Zaytsev



_______________________________________________
NLopt-discuss mailing list
[email protected]
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/nlopt-discuss

Reply via email to