120 likes | 264 Views
SPRINT: A Scalable Parallel Classifier for Data Mining. Presenter : Yu-hui Huang Authors : John Shafer , Rakesh Agrawal Manish Mehta. 國立雲林科技大學 National Yunlin University of Science and Technology. VLDB 1996. Outline. Motivation Objective Methodology Experiment Conclusion.
E N D
SPRINT: A Scalable Parallel Classifier for Data Mining Presenter : Yu-hui Huang Authors : John Shafer , Rakesh Agrawal Manish Mehta 國立雲林科技大學 National Yunlin University of Science and Technology VLDB 1996
Outline • Motivation • Objective • Methodology • Experiment • Conclusion
Motivation • Run time is expensive • must remain memory resident at all times. • Require large memory Data set
Objective • Construct a algorithm can to handle large datasets • Allowing many processors to work together
Methodology-SPRINT 27.5 <--------------------------------------------------------------------------
Methodology-SLIQ • SLIQ: • Parallelizing SLIQ: • SLIQ/R: the class list is replicated in the memory of every processor • SLIQ/D: Each.processor therefore contains only l/Nth of the class list.
Conclusion • The SPRINT is no memory restrictions • Run time is very fast , compare with previous algorithm. 10
Comments • Advantage • … • Drawback • …. • Application • medical diagnosis , fraud detection, retail target marketing…