The case for learned index structures – part I