"libsvm tools" and new version of libsvm

[Chih-Jen Lin]

Dear Colleagues:

We announce the new version 2.4 of the software libsvm:

libsvm: a library for support vector machines
http://www.csie.ntu.edu.tw/~cjlin/libsvm

and a new page "libsvm tools" containing additional
codes/data sets based on libsvm:

http://www.csie.ntu.edu.tw/~cjlin/libsvmtools

Due to popular requests, in this new page we include MANY our past and
current experimental codes/data sets such as svm probability output,
minimization of leave-one-out bounds, different multi-class svms,
reduced SVM, etc.  Moreover, we now provide an integrated development
environment to libsvm. Users can create and connect components (like
scaler, trainer, predictor, etc) in this environment.

Improvements of the new libsvm include

1. a script for users who know NOTHING about svm. It makes 
   everything automatic--from data scaling to parameter selection.
2. handling non-PSD kernels such as sigmoid and others. 
   Theoretical convergence for such cases has been established.
3. significant improvements of the python model selection tool.
4. improvements of the program for data scaling.

Any comments are very welcome.

Sincerely,
Chih-Chung Chang and Chih-Jen Lin
Department of Computer Science
National Taiwan University 
Taipei, Taiwan
cjlin at csie.ntu.edu.tw