In article <b0qmhh$1jo$[EMAIL PROTECTED]>, rif <[EMAIL PROTECTED]> wrote:
>This problem has been studied fairly extensively. Perhaps the most >references and best-known work on this is the paper by John Platt: > >http://research.microsoft.com/~jplatt/SVMprob.ps.gz > >The basic idea of this (and other) works is to fit a density function >to the real-valued SVM output, thereby mapping the arbitrary >real-valued outputs to a normalized probability scale. ... Or you could just use the somewhat-similar Gaussian process models, which directly model the probabilites, work fine with any number of classes, allow you to choose the "kernel" by maximum likelihood or Bayesian methods, and are free of "learning theory" mumbo-jumbo. These Gaussian process models *maybe* require more computation, but this probably won't be an issue unless you have at least a thousand or more cases. See the following references: http://www.cs.toronto.edu/~radford/mc-gp.abstract.html http://www.dai.ed.ac.uk/homes/ckiw/postscript/pami_final.ps.gz http://www.inference.phy.cam.ac.uk/mackay/abstracts/vgc.html ---------------------------------------------------------------------------- Radford M. Neal [EMAIL PROTECTED] Dept. of Statistics and Dept. of Computer Science [EMAIL PROTECTED] University of Toronto http://www.cs.utoronto.ca/~radford ---------------------------------------------------------------------------- . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
