说明：  处理lasso regression。 使用stochastic Markov property。
(Algorithm for solving the Lasso problem: 0.5* (y- X*beta) *(y- X*beta)+ lambda* ||beta||_1 where ||beta||_1 is the L_1 norm i.e., ||beta||_1 = sum(abs( beta )) We use the method proposed by Fu et. al based on single co-ordinate descent. For more details see GP s notes or the following paper: Penalized Regressions: The Bridge Versus the Lasso Wenjiang J. FU, Journal of Computational and Graphical Statistics, Volume 7, Number 3, Pages 397?416, 1998 )