Data.SVM

This module provides a safe bindings to libsvm functions and structures with implicit memory handling.

Documentation

type Vector = IntMap Double

Vector type provides a sparse implementation of vector. It uses IntMap as underlying implementation.

type Problem = [(Double, Vector)]

SVM problem is a list of maps from training vectors to 1.0 or -1.0

data KernelType

Kernel function for SVM algorithm.

Constructors

Linear

Linear kernel function, i.e. dot product

RBF

Gaussian radial basis function with parameter gamma

Fields

• gamma :: Double

Sigmoid

Sigmoid kernel function

Fields

• gamma :: Double
• coef0 :: Double

Poly

Inhomogeneous polynomial function

Fields

• gamma :: Double
• coef0 :: Double
• degree :: Int

data Algorithm

SVM Algorithm with parameters

Constructors

CSvc

c-SVC algorithm

Fields

• c :: Double

NuSvc

nu-SVC algorithm

Fields

• nu :: Double

NuSvr

nu-SVR algorithm

Fields

• nu :: Double
• c :: Double

EpsilonSvr

eps-SVR algorithm

Fields

• epsilon :: Double
• c :: Double

OneClassSvm

One class SVM

Fields

• nu :: Double

data ExtraParam

Extra parameters of SVM implementation

Constructors

ExtraParam

Fields

• cacheSize :: Double
• shrinking :: Int
• probability :: Int

data Model

Model is a wrapper over foreign pointer to CSvmModel

train :: Algorithm -> KernelType -> Problem -> IO Model

The train function allows training a Model starting from a Problem by specifying an Algorithm and a KernelType

train' :: ExtraParam -> Algorithm -> KernelType -> Problem -> IO Model

Like train but with extra parameters

crossValidate :: Algorithm -> KernelType -> Problem -> Int -> IO [Double]

Stratified cross validation

crossValidate' :: ExtraParam -> Algorithm -> KernelType -> Problem -> Int -> IO [Double]

Like crossvalidate but with extra parameters

loadModel :: FilePath -> IO Model

Load model from the file

saveModel :: Model -> FilePath -> IO ()

Save model to the file

predict :: Model -> Vector -> IO Double

Predict a value for Vector by using Model

withPrintFn :: CSvmPrintFn -> IO a -> IO a

Wrapper to change the libsvm output reporting function.

libsvm by default writes some statistics to stdout. If you don't want any output from libsvm, you can do e.g.:

>>> withPrintFn (\_ -> return ()) $ train (NuSvc 0.25) (RBF 1) feats

type CSvmPrintFn = CString -> IO ()