子空间比较法研究拓扑块的变量关系及变量选择

Application of Subspace Comparison Method in Studying the Relationship Between Topological Block Variables and Variable Selection

用子空间比较法研究由不同拓扑指数构成的块变量之间的关系 ,提出了一种变量选择的新方法 .以5 3 0种烷烃的 5大类拓扑指数为样本进行比较 ,得到了反映拓扑块变量间线性相关关系的夹角余弦值 ,以及这些子空间所对应的结构信息 .结合子空间比较的结果选择变量 ,对上述烷烃的沸点值进行回归 ,R=0 .9948,S=4.0 8,交互检验预测误差 LOO=4.3

This paper cares about the relationship about the block variables instead of individual topological index so that we can use subspace comparison to study the high dimensional problem. The data are about five series of topological indices of 530 saturated hydrocarbons. By the cosines of the angles between any two block-variables, the linear correlation between different block variables is studied. It is a good model of boiling point of 530 saturated hydrocarbons with variables, which are selected by the subspace comparison. The subspace comparison is a new method to select variable.