Efficient effects throughout state-space models through flexible learning in on-line Monte Carlo hope maximization.