Pity that it is in C++. Otherwise looks like they have made significant progress since I last looked at their stuff.
It is also somewhat complementary to what we have. In our SGD logistic regression code, for instance, I have pushed forward with automatic hyper-parameter tuning, on-line performance metrics and confidence based learning while they have pushed to Pegaso style updates and more perceptron algorithms. Their focus on parsing speed is definitely right on the mark. In off-list work, I have built some fast parsing code and there is a huge benefit to be had, somewhat moderated when you are running dozens to hundreds of classifiers from the same input as with the on-line hyper-parameter tuning. 2010/8/28 Gérard Dupont <[email protected]> > http://code.google.com/p/sofia-ml/ > > On Sat, Aug 28, 2010 at 08:49, Robin Anil <[email protected]> wrote: > > > Wrong List. Bah humbug! > > > > > > On Sat, Aug 28, 2010 at 12:18 PM, Robin Anil <[email protected]> > wrote: > > > > > *http://code.google.com/p/sofia-ml/* > > > * > > > * > > > Ted, Zhao. Both of you might want to take a look at this > > > > > > Robin > > > > > > > > > -- > Gérard Dupont > Information Processing Control and Cognition (IPCC) - EADS DS > http://weblab.ow2.org > > Document & Learning team - LITIS Laboratory >
