libsvm-go: Support Vector Machine

This is a full port of LIBSVM in the Go programming language. LIBSVM is a suite of tools and an API library for support vector classification, regression, and distribution estimation. This port implements the libsvm library in the form of a Go package called libsvm. It also implements the svm-train and svm-predict command line tools.

This port has no external package dependencies, and uses only the native standard library.

Installation

go get github.com/vsekhar/libsvm-go/pkg/libsvm
make

Compatibility Notes

I have tried to make the Go implementation of svm-train and svm-predict plug-in compatibile with the original LIBSVM 3.18 distribution. This is to allow you to use the other tools available in the original distribution, like easy.py and grid.py.

svm-predict should be 100% plug-in compatibile. However, svm-train is plug-in compatible with one exception. The exception is the parameter weight flag used in the command. In this implementation, the flag is

-w i,weight : set the parameter C of class i to weight*C, for C-SVC (default 1)

For full documentation of the svm-train and svm-predict commands, please refer to the original LIBSVM web site.

API Example

Training

import "github.com/vsekhar/libsvm-go/pkg/libsvm"
    
param := libsvm.NewParameter()      // Create a parameter object with default values
param.KernelType = libsvm.POLY      // Use the polynomial kernel
    
model := libsvm.NewModel(param)     // Create a model object from the parameter attributes
    
// Create a problem specification from the training data and parameter attributes
problem, err := libsvm.NewProblem("a9a.train", param) 
    
model.Train(problem)                // Train the model from the problem specification
    
model.Dump("a9a.model")             // Dump the model into a user-specified file

Predicting

import "github.com/vsekhar/libsvm-go/pkg/libsvm"
    
// Create a model object from the model file generated from training
model := libsvm.NewModelFromFile("a9a.model")  
    
x := make(map[int]float64)
// Populate x with the test vector
    
predictLabel := model.Predict(x)    // Predicts a float64 label given the test vector

vsekhar / libsvm-go