inoueakimitsu
commented
2 years ago A policy of treating all classes equally is possible.
Each class defines the priority of the bags. For each bag, the class that gives the highest priority shall be the initial label for that bag.
The pseudo code is here:
n_instance[i_bag, i_class] = lower_bound[i_bag, i_class] or upper_bound[i_bag, i_class]
bag_order[:, i_class] = rankdata(n_instance[:, i_class])
initial_bag_labels[i_bag] = argmax(bag_order[i_bag, :])
For data with large imbalance between classes, the initial labels may be inappropriate.
For example, rare classes cannot be assigned as initial labels and therefore cannot be trained.