Release from active learning/model selection dilemma: optimizing sample points and models at the same time