Opening the Black Box of Neural Nets: Case Studies in Stop/top Discrimination