Hyperparameter Selection in Continual Learning