This SVM+ code was originally written by [Dmitry Pechyony](http://www.cs.technion.ac.il/~pechyony/). I downloaded it on 2016/05/17 and fixed some minor bugs to make it runnable on Ubuntu 14.04LTS x64. I did't change any core function, so it should give the exactly same result as the original one [CODE](http://www.cs.technion.ac.il/~pechyony/conj_svm.tar.gz). > The original code is also provided [conj_svm.tar.gz](conj_svm.tar.gz) just in case the above link expired. Simple examples are provided under [example/binary-cls](example/binary-cls), plase make sure you have compiled [libsvm-3.21](libsvm-3.21) before trying. > I use the LIBSVM-3.21 for svm prediction, since the SVM+'s prediction program cannot give a resonable reuslt and may have some bugs. - - - - - - - - - - Original REAMNE - - - - - - - - - - This implementation is an extension of libsvm and implements alternating SMO and conjugate alternating SMO for solving SVM+, described in [1] and [2]. To generate executables run "make all". As in libsvm, you can see the description of all parameters by simply running ./svm-train . To speed up the experiments, I usually ran SVM+ with a relatively loose stopping criterion: -e 1.9 . Feel free to ask any questions that you might have! Dmitry Pechyony pechyony@gmail.com [1] D. Pechyony, R. Izmailov, A. Vashist and V. Vapnik. SMO-style Algorithms for Learning using Privileged Information . DMIN 2010. [2] D. Pechyony and V. Vapnik. Fast Optimization Algorithms for Solving SVM+. Chapter in Statistical Learning and Data Science, 2011.