Make scoring consistent in optimization and selection
Should complete_cases() be run, such that only complete cases across all horizons (and models) are included. How to best make this? Perhaps use the score function everywhere, then it's easy to say that it, with default parameters, is used everywhere.