A multiclass implementation is fine as is a binary implementation. If you are interested in the multi-class approach, you should take a look at this paper:
http://arxiv.org/abs/0902.3176 The upshot is that tournaments of binary classifiers may be better for some applications. On Thu, May 14, 2009 at 11:33 AM, Danny-Michael Busch <da...@kurbel.net>wrote: > In the paper it is said that the neuronal network would map only to two > classes. > What I imagined was more like a training of networks towards arbitrary > (user definable) categories / classes. > Would that be o.k. as well? ;) > If not, what are the two classes that should be utilized for mahout? > >
