Sparse Bayesian learning with Stein’s unbiased risk estimator based hyperparameter optimization