DD_tools: mpm_dd

MPM_DD Minimax prob. machine.

      W = MPM_DD(X,FRACREJ,SIGMA,LAMBDA)

 Computes the minimax probability machine of Lanckriet, using the RBF
 kernel with kernel-width SIGMA and quantile FRACREJ. It tries to find
 the linear classifier that separates the data from the origin,
 rejecting maximally FRACREJ of the target data. Unfortunately, it does
 not really work, and the rejection threshold is actually re-derived
 from the target data.

 For this method an inverse of the covariance matrix is required, and
 that might be regularized. This regularisation constant is LAMBDA.

      W = MPM_DD(X,FRACREJ,SIGMA,LAMBDA,NU,RHO)

 The method can be made a bit more robust by introducing NU>0 and RHO>0
 that allow to move the mean and covariance matrix around a bit. (See
 their paper in NIPS2002)

See also: svdd lpdd 



Copyright: D.M.J. Tax, D.M.J.Tax@prtools.org
Faculty EWI, Delft University of Technology
P.O. Box 5031, 2600 GA Delft, The Netherlands