Unable to parse email body. Email id is 5197
Bruce J. Weimer, MD <bjweimer <at> charter.net> writes: > > > I'm using LWWPE. Can anyone recommend a > (preferably free) linear algebra package? I've Googled some, but I'd > prefer a specific recommendation - if anyone has experience with one that > they particularly like? > > Thanks in advance. > > Bruce. > > Bruce, I've done some research on this. It would help if you could provide the list with what your requirements are. Do you need sparse matrices, dense ones, what kinds of operations do you need to perform, how big are the matrices, etc. Generally, it would be nice if Matlisp was supported, because I've found that floating point math isn't as fast as with CMUCL or SBCL. This would be my favorite option, if it existed. I've looked into what it would take to port Matlisp to LispWorks. The Matlisp ffi interface is abstracted pretty well into a single file per implementation. Being somewhat familiar with LispWork's ffi, I estimated it be at least several days of solid work, if there's not show stopper issues. You might want to ask the same question in the Matlisp user group, as sometimes there are general discussions about linear algebra and Lisp there. You might also want to ask if someone's already ported Matlisp to LispWorks and would be willing to share it. Without knowing your requirements, here's some other options: Femlisp. It's pretty full featured and has a subset of what's defined in Matlisp, although written in CL. As the author points out, Femlisp using LispWorks will be slow. As an example, solving a system of 1000 equations using Matlisp/CMUCL using LAPACK takes about 2.5 seconds and 0.5 seconds if built with ATLAS. Solving the same size system using Femlisp/LispWorks takes about 120 seconds and 4 seconds on SBCL/Femlisp. MATLAB on the same computer & OS takes about 0.5 seconds as well, since it internally uses ATLAS that's not suprising. I've noticed this same 10-20x speed difference in some of the computer language shootout benchmarks as well. Note that Femlisp isn't officially supported for LispWorks, as it took about 5 very trivial modifications to get it to build. Another not free, but inexpensive option is to get the Numerical Recipies CDROM. It's rarely mentioned, but it has Common Lisp translations of all of the functions in the Fortran 1st edition. You can't freely distribute it, but if you're just running privately, that might work for you. I've examined and slightly tested its code and you'll have the same slowness on LW that Femlisp has. The CLEM package has some basic matrix operations, nothing like solving linear systems of equations, but if your needs are light it might be ok. NLISP is coming along nicely. It's architected to bind to the GNU GSL library. It currently works with CMUCL, but the author just ported it to the new CFFI package which does have LispWorks support. Good luck. Glenn Ehrlich
Martin Simmons <martin@lispworks.com> writes: > The main floating point optimization trick for LispWorks is to use > > (optimize (float 0)) Why zero? In my mind, it would be more intuitive if 3 was the value for maximum optimization of floating point operations. -- Mvh/Regards Peder O. Klingenberg Netfonds ASA
Unable to parse email body. Email id is 5206