A Data Cube Algebra Engine for Data Mining
The interaction between Data Surveyor and its DBMS backends is described using an extended relational algebra, the Data Cube Algebra, to encode the mining requests and a drill engine produces optimized code for several database back-ends.
Abstract
M.L. Kersten, A.P.J.M. Siebes CWI, Amsterdam, The Netherlands M. Holsheimer , F. Kwakkel Data Distilleries, Amsterdam, The Netherlands Abstract On line data mining products, such as Data Surveyor, illustrate that an extensible architecture to accommodate a variety of mining algorithms and database interconnectivity is technically feasible. In this paper we describe the interaction between Data Surveyor and its DBMS backends using an extended relational algebra, the Data Cube Algebra, to encode the mining requests. Subsequently, a drill engine produces optimized code for several database back-ends. Amongst others, the optimizer exploits commonalities amongst multiple query batches and target platform speci c optimizations rules. The e ectiveness of several strategies is illustrated using the Monet database engine.