Allow BalancingLearner to return arbitrary number of points from 'choose_points'

[jbweston]

(original issue on GitLab)

opened by Joseph Weston (@jbweston) at 2017-09-01T12:42:28.237Z

Currently it is limited to len(self.learners).

We should fix this by implementing _choose_one_point, which selects the learner with the largest loss improvement,
and then calling it n times in choose_points.

Anything cleverer would also be much more complex.