2013/1/10 Edward J. Yoon <[email protected]> > Cool. > > BTW, what is our plan about the future of math classes? Personally I'd > like to package all math classes into ML module and so provide them > together. >
IMHO either have them in ML module (preferred right now) or have a separate math module (likely to be better for the future). Tommaso > > On Thu, Jan 10, 2013 at 5:12 PM, Tommaso Teofili > <[email protected]> wrote: > > Hi all > > > > I recently played a bit with classification algorithms so I think it may > > make sense to start adding a BSP version of them in our machine learning > > package. > > As a first step though I would like to add a couple of adapter interfaces > > in order to be able to use existing (BSP or not) classifiers inside our > BSP > > algorithms. This would be useful, for example, for meta learning > algorithms > > like weighted majority [1]. > > Apart from that we could add some basic algorithms first: > > - perceptron > > - naive bayes > > > > and then move to the more tricky ones (neural networks, decision tress, > > random forest, svm, etc.). > > > > Note that currently we have logistic regression with gradient descent for > > classification. > > > > Have a nice day, > > Tommaso > > > > > > [1] : http://en.wikipedia.org/wiki/Weighted_Majority_Algorithm > > > > -- > Best Regards, Edward J. Yoon > @eddieyoon >
