[MADLIB-901] Implement new SVM module

• Type: Epic
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: v1.9
• Component/s: Module: Support Vector Machines
• Labels: SVM
• Assignee: Xiaocheng Tang
• Reporter: Frank McQuillan
• Created: 02/Nov/15 21:27
• Updated: 21/Mar/16 18:33
• Resolved: 21/Mar/16 18:33

Description

The current version of SVM in MADlib is characterized as an "Early Stage Development" module, with some known deficiencies. The goal of this jira is to rewrite SVM from scratch and add new capability.

Main features desired:

• linear SVM
• non-linear SVM for Gaussian and polynomial kernels
• support classification, regression and novelty detection