OONumerics User :

From: Michael Lehn (michael.lehn_at_[hidden])
Date: 2007-07-20 11:03:49


Hi Georg,

of course I only can tell you about what FLENS is capable or not. So
this is no advice or suggestion ...

So the short answer: out of the box FLENS can not solve your problem.

The longer answer: FLENS was designed to provide a clean separation
between its interface for "end-user" (who are
still programmers) and an implementation providing the actual
functionality. This makes FLENS very easy to extend.

However you have to define a more detailed list of operations that
you need:
- If there are existing (lower-level) libraries (like http://
www.netlib.org/svdpack/svdpack) that already provide an
   appropriate implementation these can be used as back-end for the
computations. This is possible without loss of
   performance.
- If you really want an efficient solution you also have to consider
things like: do I need all eigenvalues, or only the
   five largest eigenvalues? Things like this have an impact on what
method is most suited, e.g. iterative method vs.
   direct method. And this has an impact on what storage format is
most suitable to store the sparse matrix (will there
   be fill-ins or not?).
- the pseudo-inverse of a sparse matrix is in general a dense
matrix. So do you really need the pseudo-inverse
   explicitly? Or do you merely want to solve a least square
problem? In the latter case it is not necessary to setup
   the pseudo-inverse explicitly (and thereby loosing sparsity).

So if you can provide more detailed information it would be easier to
estimate how much effort it would take. I guess
the same holds no matter what library you intend to use.

Cheers,
Michael

----
http://flens.sf.net
Am 20.07.2007 um 14:18 schrieb Georg Holzmann:
> Hallo list!
>
> (Sorry if this was already answered, but I didn't find something in  
> the
> archives)
>
> I am searching for a fast open-source library for sparse matrix
> processing (in C++, or otherwise C - on linux, but should be
> crossplatform on win/linux/osx) for realtime audio - so the prime
> importance is efficiency (of course also stability, usability, ...;)
>
> I need to perform the following calculations: eigenvalue calculation,
> standard arithmetic operations, pseudo-inverse of a matrix - I found
> lots of libraries in the net (boost::UBlas, FLENS, ATLAS, ...), but  
> any
> comment on which one to choose would be highly appreciated !
>
> Thanks in advance,
> LG
> Georg
> _______________________________________________
> oon-list mailing list
> oon-list_at_[hidden]
> http://www.oonumerics.org/mailman/listinfo.cgi/oon-list