30 likes | 200 Views
You Probably Expect Me to Say Something About Datalog. Recursion turns out to be much more important in big-data operations than we thought in 1980. Examples : PageRank, gradient descent, dimensionality reduction, SVM, etc., etc. Many new extended Datalog -like languages are coming into use.
E N D
You Probably Expect Me to Say Something About Datalog • Recursion turns out to be much more important in big-data operations than we thought in 1980. • Examples: PageRank, gradient descent, dimensionality reduction, SVM, etc., etc. • Many new extended Datalog-like languages are coming into use. • Examples: LogicBlox, Scalops. • Too many dislike logic, need SQL. • Better recursion in SQL needed. • Nonlinear, nonstratified.
Nested Relations • Tree-like structures popular outside SQL. • Record (tuple) types + collection types like repeating groups, optional elements, arrays, etc., built recursively. • Examples: • Hierarchical model without pointers. • DTD’s without refs. • Google Protocol buffers. • ODL classes without relationships. • NoSQL types as in BigTable, Hbase, etc.
Adapting SQL • In a sense, this is all in Object-Relational SQL, but constructs often cumbersome. • Start from the assumption that all rows are tree-like, with simple notations to navigate trees. • Following links OK, but efficiency?? • Semantics gets tricky fast. • Example: (A, B*, C*), with selection B < C. • XQuery would interpret existentially. • If you flatten (take X-product of B and C), and then select, you get a different structure.