An Affine Scaling Interior Point Method

for Continuous Knapsack Constraints

with Application to SVM


by


M. D. Gonzalez-Lima, W. W. Hager, and H. Zhang


  • The paper, July 25, 2009, revised July 9, 2010.

  • The code svmASL

  • Table of Numerical Results

  • Runfile for svmASL

  • Runfile for GPDT

  • Runfile for LIBSVM

  • September 30, 2010: A new code BLG, for solving optimization problems with bound and linear constraints using gradient methods, has been incorporated into the SVMsubspace code. BLG has greater flexibility than ASL in the choice of the search direction since either affine scaling, gradient projection, or Frank-Wolfe search directions can be selected. BLG improves on ASL by treating various numerical stability issues in a more systematic way. The new code is available here.

    Support Vector Machine (SVM) Test Problems

    Problem Name Description
    a7a Adult (predict whether a household has income greater than $50,000 based on census data)
    a8a Adult (predict whether a household has income greater than $50,000 based on census data)
    a9a Adult (predict whether a household has income greater than $50,000 based on census data)
    ijcnn1 IJCNN 2001 neural network competition
    w6a Website (predict whether web page belongs to a category based on the presence of 300 selected keywords on the page)
    w7a Website (predict whether web page belongs to a category based on the presence of 300 selected keywords on the page)
    real-sim UseNet articles from four discussion groups, for simulated auto racing, simulated aviation, real autos, real aviation. --separate real from simulated, and auto from aviation
    mnist Distinguish digit 8 from the other 9 digits


  • GNU GENERAL PUBLIC LICENSE