function [classes, B_t, e_t] = k_adaboost_training(patterns, targets, depth, T)

P = patterns;
nData = length(P);
fraction = round(nData/2);
D_t = (1/nData)*ones(1,nData);

for t = 1:T

%Select training subset, make tree
inds = sample_discrete(D_t, 1, fraction);
train_patterns = P(inds,:);
train_targets = targets(inds);
[test_labels, train_labels, error, tree_t] = k_make_tree_n_test(train_patterns, train_targets, P, targets, depth);

%Store results
classes{ t } = tree_t;
e_t(t) = error(:,1)+eps;
B_t(t) = (error(:,1)+eps)/(1-error(:,1));

%Apply new weights for next subset
inds_correct = find(test_labels == targets.');
D_t(inds_correct) = D_t(inds_correct).*B_t(t);
D_t = D_t./sum(D_t);

end