Learning Compositional Sparse Gaussian Processes with a Shrinkage Prior