On Tue, Mar 06, 2012 at 10:05:19PM +0100, Gael Varoquaux wrote: > Hi Vlad and Andy, > > First of all, Vlad, thanks for offering to take up the less sexy work. I > don't think that maintenance can qualify as a GSOC project. A big > refactor could, but I don't think that we need one right now.
What about a "speed freak"-themed GSOC? Some things I can think of: a) sparse coding is about 2 orders of magnitude slower than competing implementations right now, making it kind of useless except in toy 1996-sized situations (I'm supposed to find a way to benchmark this for Alex, but I can tell you that the situation is fairly bad currently, compared to e.g. SPAMS; Olivier said this had something to do with a badly optimized convergence check) b) a few things things could benefit from Cython rewrites c) several things could probably benefit from Cython's newfound OpenMP capabilities to parallelize across different GIL-free worker threads d) patching liblinear to work with dense matrices e) patching LOTS of things to accept and operate natively on float32 (this can be quite important when working with huge data where doubles simply aren't an option -- Cython fused types will make this simpler in some cases) > With regards to reaching 1.0 release, one thing that I would really like > to achieve is a real online learning API and set of examples. The reason > that I would like to get this done before 1.0 is that I feel that going > through such design could force us to reconsider some of our API, and > that should happen before 1.0. I would love to see such a project > proposed by someone who has enough experience. Hopefuly we should be able > to find mentors for it. I think you're absolutely right on this front. Online learning does require a different mindset from an API design perspective, we're only getting the hang of it ourselves and it's basically all we ever do... David ------------------------------------------------------------------------------ Keep Your Developer Skills Current with LearnDevNow! The most comprehensive online learning library for Microsoft developers is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3, Metro Style Apps, more. Free future releases when you subscribe now! http://p.sf.net/sfu/learndevnow-d2d _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
