SPIDER The Spider Objects

Probabilistic outputs for SVM

   A=PLATT(C,H) returns a platt object initialized with hyperparameters H and
               based on algorithm C. 
   Converts a real valued margin producing pattern recognition
   algorithm into a conditional probability estimator. It thus
   requires a pattern recognition approach that can produce a real
   valued output before thresholding, e.g SVMs. It achieves this
   by finding the coefficients of a sigmoid by using cross validation.
   Hyperparameters, and their defaults
    child = svm          -- real valued margin algorithm to be used 
    folds = 3            -- number of cv folds
    A                -- 
    B                -- Parameters of the sigmoid: 1/(1+exp(A*f(x)+B)
    training, testing 
     rr.X  contains now the probability of being class 2

Reference : Probabilistic Outputs for Support Vector Likelihood Methods Machines and Comparisons to Regularized
Author : John Platt
Link : http://research.microsoft.com/users/jplatt/SVMprob.ps.gz