I'm attempting to implement a multi-class variant of the AdaBoost algorithm called SAMME. I'm fairly sure I have it right, except its performance is abysmal. This is likely due to an incorrect choice of weak learner that isn't good enough; but I need one which can be trained easily in a weighted sense, which limits me somewhat.
Does anyone have enough expertise in this area to be of any help?
Does anyone have enough expertise in this area to be of any help?
