原文地址：http://www.csie.ntu.edu.tw/~cjlin/libsvm/index.html

Chih-Chung Chang and Chih-Jen Lin

 Version 3.21 released on December 14, 2015. It conducts some minor fixes. 
 LIBSVM tools provides many extensions of LIBSVM. Please check it if you need some functions not supported in LIBSVM. 
 We now have a nice page LIBSVM data sets providing problems in LIBSVM format. 
 A practical guide to SVM classification is available now! (mainly written for beginners) 
We now have an easy script (easy.py) for users who know NOTHING about SVM. It makes everything automatic--from data scaling to parameter selection. 
The parameter selection tool grid.py generates the following contour of cross-validation accuracy. To use this tool, you also need to install python and gnuplot.

To see the importance of parameter selection, please see our guide for beginners. 
 Using libsvm, our group is the winner of IJCNN 2001 Challenge (two of the three competitions), EUNITE world wide competition on electricity load prediction, NIPS 2003 feature selection challenge (third place), WCCI 2008 Causation and Prediction challenge (one of the two winners), and Active Learning Challenge 2010 (2nd place).

Introduction

LIBSVM is an integrated software for support vector classification, (C-SVC, nu-SVC), regression (epsilon-SVR, nu-SVR) and distribution estimation (one-class SVM). It supports multi-class classification.

Since version 2.8, it implements an SMO-type algorithm proposed in this paper:
R.-E. Fan, P.-H. Chen, and C.-J. Lin. Working set selection using second order information for training SVM. Journal of Machine Learning Research 6, 1889-1918, 2005. You can also find a pseudo code there. (how to cite LIBSVM)

Our goal is to help users from other fields to easily use SVM as a tool. LIBSVM provides a simple interface where users can easily link it with their own programs. Main features of LIBSVM include

• Different SVM formulations
• Efficient multi-class classification
• Cross validation for model selection
• Probability estimates
• Various kernels (including precomputed kernel matrix)
• Weighted SVM for unbalanced data
• Both C++ and Java sources
• GUI demonstrating SVM classification and regression
• Python, R, MATLAB, Perl, Ruby, Weka, Common LISP, CLISP, Haskell, OCaml, LabVIEW, and PHP interfaces. C# .NET code and CUDA extension is available. 
  It's also included in some data mining environments: RapidMiner, PCP, and LIONsolver.
• Automatic model selection which can generate contour of cross validation accuracy.

Download LIBSVM

The current release (Version 3.21, December 2015) of LIBSVM can be obtained by downloading the zip file or tar.gz file. You can also check this github directory. Please e-mail us if you have problems to download the file.

The package includes the source code of the library in C++ and Java, and a simple program for scaling training data. A README file with detailed explanation is provided. For MS Windows users, there is a sub-directory in the zip file containing binary executable files. Precompiled Java class archive is also included.

Please read the COPYRIGHT notice before using LIBSVM.

Graphic Interface

Here is a simple applet demonstrating SVM classification and regression.
Click on the drawing area and use ``Change'' to change class of data. Then use ``Run'' to see the results.

Change Run Clear

Examples of options: -s 0 -c 10 -t 1 -g 1 -r 1 -d 3 
Classify a binary data with polynomial kernel (u'v+1)^3 and C = 10

options:
-s svm_type : set type of SVM (default 0)0 -- C-SVC1 -- nu-SVC2 -- one-class SVM3 -- epsilon-SVR4 -- nu-SVR
-t kernel_type : set type of kernel function (default 2)0 -- linear: u'*v1 -- polynomial: (gamma*u'*v + coef0)^degree2 -- radial basis function: exp(-gamma*|u-v|^2)3 -- sigmoid: tanh(gamma*u'*v + coef0)
-d degree : set degree in kernel function (default 3)
-g gamma : set gamma in kernel function (default 1/num_features)
-r coef0 : set coef0 in kernel function (default 0)
-c cost : set the parameter C of C-SVC, epsilon-SVR, and nu-SVR (default 1)
-n nu : set the parameter nu of nu-SVC, one-class SVM, and nu-SVR (default 0.5)
-p epsilon : set the epsilon in loss function of epsilon-SVR (default 0.1)
-m cachesize : set cache memory size in MB (default 100)
-e epsilon : set tolerance of termination criterion (default 0.001)
-h shrinking: whether to use the shrinking heuristics, 0 or 1 (default 1)
-b probability_estimates: whether to train a SVC or SVR model for probability estimates, 0 or 1 (default 0)
-wi weight: set the parameter C of class i to weight*C, for C-SVC (default 1)The k in the -g option means the number of attributes in the input data.

To install this tool, please read the README file in the package. There are Windows, X, and Java versions in the package.

Additional Information (how to cite LIBSVM)

Frequently Asked Questions (FAQ) and Change log

References of LIBSVM:

• Official implementation document: 
  C.-C. Chang and C.-J. Lin. LIBSVM : a library for support vector machines. ACM Transactions on Intelligent Systems and Technology, 2:27:1--27:27, 2011. 
  pdf, ps.gz, ACM digital lib.
• Instructions for using LIBSVM are in the README files in the main directory and some sub-directories.
• A guide for beginners: 
  C.-W. Hsu, C.-C. Chang, C.-J. Lin. A practical guide to support vector classification
• An introductory video for windows users.
• Other implementation documents: 
  R.-E. Fan, P.-H. Chen, and C.-J. Lin. Working set selection using the second order information for training SVM. Journal of Machine Learning Research 6, 1889-1918, 2005. You can also find a pseudo code there.
• Other documents written by users. (including some non-English documents)

For more information about nu-SVM and one-class SVM , please see

• B. Schölkopf, A. Smola, R. Williamson, and P. L. Bartlett. New support vector algorithms. Neural Computation, 12, 2000, 1207-1245.
• B. Schölkopf, J. Platt, J. Shawe-Taylor, A. J. Smola, and R. C. Williamson. Estimating the support of a high-dimensional distribution. Neural Computation, 13, 2001, 1443-1471.

Interfaces and extensions to LIBSVM

Acknowledgments: This work was supported in part by the National Science Council of Taiwan via the grant NSC 89-2213-E-002-106. The authors thank their group members and users for helpful discussion and comments. Please send comments and suggestions to Chih-Jen Lin.