带L1规划的广义线性模型中的一种系数估计法  

L1 Regularization Path Algorithm for Generalized Linear Models

作　　者：温育芳[1] 宋向东[1] 张海森[1] 于尚洋[1] 杨婧[1] 

机构地区：[1]燕山大学理学院,河北秦皇岛066004

出　　处：《佳木斯大学学报（自然科学版）》2008年第1期75-77,共3页Journal of Jiamusi University：Natural Science Edition

摘　　要：介绍了L1规划广义线性模型(GLM)的一种系数估计法,估计系数的同时进行变量选择,从而确立模型.事实上,此算法用来选择变量更有用.L1规划法按照对系数一范数的惩罚来选择变量,是向前选择变量法的一种改进,运用凸优化的预测—修正法,GLM系数估计法可有效地算出系数,其中规划参数的步长对控制系数精确性至关重要,本文对三种步长选择法作了相应的比较和分析,接着给出算法的推广,最后总结算法并且对这种估计法的应用范围进行了展望.An estimation algorithm of coefficient to select variables for L1 regularized generalized linear models was introduced. Variables were selected with L1 regularization procedure according to the amount of penalization on the L1 norm of the coefficients. This method is much mere efficient than forward selection/backward deletion. The GLM estimation algorithm of coefficient can efficiently compute solutions along the entire regularization path using the predictor-corrector method of convex-optimization. Selecting the step length of the regularization parameter is critical in controlling the accuracy of coefficient. The corrsponding comparison and analysis about three step length selection were presented. Then, the algorithm was expanded. The prospect application expansion of the estimate algorithm was given in the end.

关 键 词：广义线性模型 预测 修正算法 KKT最优化条件 

分 类 号：O212.1[理学—概率论与数理统计]