LIBSVM多分类模板数据存储结构优化

SVM is a new pattern recognition method developed in recent years on the basis of statistical learning theory, which was originally designed as a binary classification problem, when dealing with multi-classification problems, we need to construct an appropriate multi-class classifier. LIBSVM developed and designed by Lin Chih-Jen and others of National Taiwan University uses the one against one method which will construct k(k-1)/2 binary classifiers and ballot method to handle the multi-classification problems. Since a single binary SVM only uses part of the corresponding category support vector, most of the support vector coefficients are zero. After further study of LIBSVM model data storage structure, we propose a storage structure of LIBSVM model data with sparse storage to improve the defects of original storage structure.