How Bellhops Leverages Amazon Redshift UDFs for Massively Parallel Data Science