When training, look if there is bayesian search results, if yes use this. Exception: forest_size use the one given by parser if applicable