摘要
文献[1]给出了一种从小于最优岭参数k0的初值出发逐步改进岭参数的方法。这种方法改进了Hoerl和Kennard的结果。本文给出了另外一种从大于最优岭参数k0的初值出发逐步改进岭参数的方法。在实际应用中,这2种方法互为补充。
In this paper,the author considered linear model Yn×1=Xn×mβm×1+εn×1,E(Y)=Xβ,Var(Y)=σ2In,R(X)=m.Its canonical model was Yn×1=Zn×mαm×1+εn×1.Where Z′Z=Λ=diag(λ1,…,λm),λ1≥0,…,λm≥0 were the eigenvalues of X′X.The ridge estimator of α was α∧(k)=(Λ+kI)-1Z′Y,and the ridge estimator of β was P′α∧(k),where P was orthogonal matrix.So that P′X′XP=Λ.Paper (1) gave a new method to determine ridge parameter k.That method has improved the Hoerl-kennard formula.In this paper,another method to determine ridge parameter k in ridge regression was given.
出处
《江苏工业学院学报》
2003年第1期39-42,共4页
Journal of Jiangsu Polytechnic University
基金
江苏省教育厅自然科学研究基金资助(02KJD110002)
江苏工业学院科技基金资助
