JRuby library for working with libsvm
Java Ruby
Switch branches/tags
Nothing to show
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
bin
examples
java
lib
src/main/java
test
LICENSE.txt
README.rdoc
Rakefile.rb
buildfile.rb
svm_toolkit.gemspec

README.rdoc

svm_toolkit

Support-vector machines are a popular tool in data mining. This package includes an amended version of the Java implementation of the libsvm library (version 3.11). Additional methods and examples are provided to support standard training techniques, such as cross-validation, and simple visualisations. Training/testing of models can use a variety of built-in or user-defined evaluation methods, including overall accuracy, geometric mean, precision and recall.

Copyright © 2011-12, Peter Lane

Install

This software works with JRuby, in 1.9 mode.

To install:

$ jruby -S gem install svm_toolkit

Features

Current

  • All features of LibSVM 3.11 are supported, and many are augmented with Ruby wrappers.

  • Loading Problem definitions from file in Svmlight, Csv or Arff (simple subset) format.

  • Creating Problem definitions from values supplied programmatically in arrays.

  • Rescaling of feature values.

  • Integrated cost/gamma search for model with RBF kernel, taking advantage of multiple cores.

  • Contour plot visualisation of cost/gamma search results.

  • Model provides value of w-squared for hyperplane.

  • svm-demo application, a version of the svm_toy applet which comes with libsvm.

  • Model stores indices of training instances used as support vectors.

  • User-selected evaluation techniques supported in Model#evaluate_dataset and Svm.cross_validation_search.

  • Library provides evaluation classes for OverallAccuracy, GeometricMean, ClassPrecision, ClassRecall, MatthewsCorrelationCoefficient.

Planned

  • splitting problem sets for train/cross/test

  • support for sampling, SMOTE and related processes (perhaps in separate package)

  • active-learning

License

svm_toolkit is free software: you can redistribute it and/or modify it under the terms of the Open Works License.

See file 'LICENSE.txt' for more details.

Acknowledgements

The svm_toolkit is based on LibSVM, which is available from: www.csie.ntu.edu.tw/~cjlin/libsvm/

The contour plot uses the PlotPackage library, available from: thehuwaldtfamily.org/java/Packages/Plot/PlotPackage.html

Contributor: