Auto-LfD: Towards Closing the Loop for Learning from Demonstrations