Questions:
1.How to apply the distribution D in these algorithms proposed in Freund and Schapire's paper?

2.In Schapire and Singer's paper, in the generalized version of AdaBoost, sometimes h(y)'s range can be all real value, sometimes be restricted in [-1,+1], sometimes be restricted in {-1,0,+1}, so I'm confused by this. How to decide the range of h(y)? Will it influence the classfication result?

I'm really interested in section 4 of Freund and Schapire's paper. I think maybe it can also be used in SVM
classifier(in the editing-SVM) to reduce the number of support vectors, so that similarly the SVM algorithm can also be speed up.