一种支持向量机的训练方案

Although several programs to training Support Vector Machines may be found over the Internet, the main reason to develop a new one was to create a framework where not only new kernels could be added, as happens with the majority of SVMs programs, but also it was allowed to add new training algorithms and QP solver strategies. SVMBR was developed from scratch, in C++, after a careful class modeling and can be compiled to run on any operational system with a standard C++ compiler. At this moment, SMO and EDR are the only training methods available and there are four kernel functions supported: linear, polynomial, RBF and sigmoid. Reference: SVMs training using re-sampling based on error and a priori strategies for sample selection. PhD Thesis. Marcelo Barros de Almeida. Universidade Federal de Minas Gerais. 2002.