#votes

Question 1: Select all that apply

Which statements about AdaBoost with decision stumps are true? Select all that apply.

Its votes give the largest possible training margins of any weighting of the same stumps1
Given enough rounds it always reaches zero training error, whatever the weak learners2
The votes $\alpha_k$ it assigns are never negative3
The training errors of the new stump and of the combined classifier move in concert4
The weighted error of each new stump tends to increase over the rounds5