Multicategory SVM PATH
This is an add-on R package to svmpath designed to compute the entire regularized solution path of the multicategory SVM. It generalizes svmpath to handle more than two classes.
- Package source: msvmpath_0.1-1.tar.gz | msvmpath_0.1-1.zip
- Reference manual: msvmpath.pdf
- msvmpath is developed and maintained by Zhenhuan Cui.
For any questions about this package, send email to Zhenhuan.Cui@sas.com.
- Characterizing the Solution Path of Multicategory Support Vector Machines - Lee, Y. and Cui, Z., Statistica Sinica, vol. 16, No. 2, 391-409, 2006.
This software in R computes a solution to the structured multicategory SVM (SMSVM), which incorporates feature selection with the SVM using the functional ANOVA decomposition. The lasso type L1 penalty on the shrinkage factors of functional components enforces sparsity of the solution. The ordinary MSVM is a special case of the SMSVM when there is no component shrinkage.
- SMSVM v1.2.1: R functions written by SangJun Lee, Yuwon Kim, Yoonkyung Lee, and Ja-Yong Koo.
- Documentation: a brief description of R functions (User's guide to be posted)
- Structured Multicategory Support Vector Machine with ANOVA decomposition (ps) - Lee, Y., Kim, Y., Lee, S., and Koo, J.-Y., October 2004, Technical Report No. 743, Department of Statistics, The Ohio State University. Biometrika, vol. 93, No. 3, 555-571, 2006.
- Multicategory Support Vector Machines, Theory, and Application to the Classification of Microarray Data and Satellite Radiance Data (ps) - Lee, Y., Lin, Y., and Wahba, G., Revised August 2003, Technical Report 714, Department of Statistics, The Ohio State University. Journal of the American Statistical Association, vol. 99, 67-81, 2004.
- The software is under continuous development and any updates
will be announced on this website.
Detailed documentation is to follow.
- [Sep.10.2005] SMSVM v1.2 has been posted. Authors' notes (restricted access).
- [Sep.15.2005] Data splitting process for cross-validation has been refined to handle leave-one-out case. With the exception of leave-one-out cross-validation, the fold size should not be smaller than the minimum class size. To the effect, data.split function has been newly added. SMSVM 1.2.1 reflects this change.
- [Nov.11.2005] Description of some outputs of R functions has been added to Documentation, and the R bug of converting a row vector of x.test into a column vector when the test set consists of only one case has been fixed. This affects class.code, predict.msvm, predict.smsvm, make.anovaKernel, and eval.kernel.
LICENSE: The software is distributed free of charge for academic use if its use is properly acknowledged, and it may not be used for any commercial purposes without the authors' agreement. For any comments about this software, email to yklee at [stat domain].