发明名称 High-speed statistical processing in a database
摘要 Numerically intensive statistical processing of data is implemented as an incremental gradient method within the engine of a database system. Small user-defined functions in the database system calculate an approximate gradient from one term of a linearly separable defined cost resolvable from a single tuple of the database. In this way the optimized data access of the database may be exploited for rapid statistical processing.
申请公布号 US9208203(B2) 申请公布日期 2015.12.08
申请号 US201113253635 申请日期 2011.10.05
申请人 发明人 Ré Christopher M.;Recht Benjamin H.
分类号 G06F17/30 主分类号 G06F17/30
代理机构 Boyle Fredrickson, S.C. 代理人 Boyle Fredrickson, S.C.
主权项 1. A method of improving the operation of an electronic computer including a hardware processor and hardware memory implementing an incremental gradient method and holding a database having multiple tuples, the incremental gradient method being applied for a cost function comprised of linearly separable terms comprising the steps of: (a) loading a gradient function into a user-defined function of a relational database management program associated with the database, the relational database system including a query processor and a database handler, the query processor receiving standard query language queries and the user defined function and a database handler managing a physical structure of a relational database and receiving instructions from the query processor to respond to the queries, the gradient function representing a gradient of a linearly separable term of a cost function, the gradient function applicable to a single tuple at a time; (b) constructing a standard query language query for the relational database management program providing for successive modification of an initial argument by a tuple-by-tuple application of the gradient function to each tuple to successively modify the initial argument, a modification of each successive modification of the initial argument being according to a gradient of the gradient function related to a current tuple; (c) providing the standard query language query to the query processor; (d) executing the standard query language query on the relational database management program by the query processor, wherein executing includes identifying successive tuples of the database and successively modifying the initial argument by a tuple-by-tuple application of the gradient function to each successive tuple of the database; and (e) outputting a result to the user based on modified initial arguments after execution of the standard query language query.
地址