坐标下降法求解LASSO问题.pdfVIP

The Annals of Applied Statistics 2008, Vol. 2, No. 1, 224–244 DOI: 10.1214/07-AOAS147 c Institute of Mathematical Statistics, 2008 COORDINATE DESCENT ALGORITHMS FOR LASSO PENALIZED REGRESSION1 8 0 By Tong Tong Wu and Kenneth Lange 0 2 University of Maryland, College Park and University of California, Los r Angeles a M Imposition of a lasso penalty shrinks parameter estimates toward zero and performs continuous model selection. Lasso penalized re- 7 gression is capable of handling linear regression problems where the 2 number of predictors far exceeds the number of cases. This paper ] tests two exceptionally fast algorithms for estimating regression co- P efficients with a lasso penalty. The previously known ℓ2 algorithm is A based on cyclic coordinate descent. Our new ℓ1 algorithm is based on . greedy coordinate descent and Edgeworth’s algorithm for ordinary t ℓ1 regression. Each algorithm relies on a tuning constant that can be a t chosen by cross-validation. In some regression problems it is natural s [ to group parameters and penalize parameters group by group rather than separately. If the group penalty i