help-octave
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

sparse quadratic optimization?


From: fork
Subject: sparse quadratic optimization?
Date: Wed, 3 Apr 2013 18:26:28 +0000 (UTC)
User-agent: Loom/3.14 (http://gmane.org/)

Hi all,

Is there any facility for doing qp() using sparse matrices?  I think
that the compiled backend (__qp__()) is implemented using full
matrices only.

The backstory is that I am minimizing a function in which H is the
result of kronecker multiplication by eye(), so it is mostly sparse.
Because of the Kronecker product I can have H and q (input matrices)
of more than 1000 by 1000, which gets hosed when intermediate
eigenvalues are calculated.

Maybe the issue is that __qp__ doesn't use the ARPACK functions? WOuld
they even work on the algorithm?  From testing on the command line, it
looks like svds() is quite a bit faster on both the sparse matrices
and matrices with lots of zeros but still technically full.

(If qp() could leverage ARPACK and sparsity, that would have payoff
for sqp too)

I will stare at the C code some more, but I have trouble building
Octave on my FreeBSD machine at home, and probably not enough
background to do anything useful anyway.  But maybe I will actually
have a patch one day...

Sorry for the vague barf, but I would love to hear anybody's thoughts.

For reference:

hg.savannah.gnu.org/hgweb/octave/file/28136851099a/libinterp/corefcn/__qp__.cc



reply via email to

[Prev in Thread] Current Thread [Next in Thread]