Inf-FS是图论方法利用矩阵的幂级数的收敛性能评估的重要性特性对所有其他的在一起。事实上,Inf-FS配方,每个特性映射在一个关联图,节点代表特性和加权边缘之间的关系。一定长度的每条路径图被视为一个可能的选择功能。因此,这些路径不同,让他们倾向于无限许可调查的重要性每个可能的特征子集。Inf-FS分配每个特性的初始设置的最后得分;分数的多少有关给定的特性是一个很好的候选人关于分类的任务。因此,在后代排名顺序Inf-FS允许我们执行的结果特征子集选择在选型阶段来确定特征选择的数量。
= = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
参考:无限的特征选择
链接:http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=7410835
ResearchGate网站:https://www.researchgate.net/publication/282576688_Infinite_Feature_Selection
= = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =