lilac / svm Goto Github PK

INTRODUCTION:

svm is a library for doing least squares support vector regression.  It is
implemented in the Haskell programming language.  The library is set up as
a Cabal package and can be downloaded from github.com/andrewdougherty/svm
or hackage.haskell.org/package/svm.

Currently the library implements:

least squares support vector regression

The following kernel functions are included:

linear kernel function (featureless space)
multilayer perceptron (similar to a neural net)
polynomial kernel function (polynomial fit of the data)
radial basis function (Gaussian basis functions)
reciprocal kernel function (decaying exponential basis functions)
spline kernel function

For least squares support vector regression, the solution for a set of
points is given by:

|y> = K |a> + b |1>

A conjugate gradient algorithm (CGA) is used to find the optimal set of
dual weights |a>.

USAGE:

Given a set of training points {point, value} least squares support vector
regression is done with the command:

dataSet = DataSet <points> <values>
svm = LSSVM (KernelFunction <kernelFunction>) <cost> <kernelParams>
solution = solve svm dataSet <epsilon> <iterNum>

where the variables in the angles brackets are:

points  :: Array Int [Double]  -- The point in the feature space.
values  :: UArray Int Double   -- The value at the corresponding point.
epsilon :: Double              -- A cutoff value for the step size of the CGA.
iterNum :: Int                 -- The max number of iterations for the CGA.