通过逐元Gibbs抽样加速随机搜索变量选择.pdfVIP

£á £££ ááá ë|¢C˛¿J(George and McCulloch(1993))¥Ç5£8.•Å61 C˛¿Jê{Éò ©z •ÈU?ŸOé«â ÈıÛä ß,˘}£å ıÍ—UCk ìd˙™ß˘¥kî˙ †ÿ©Ã á5:¥3±kìd˙™ÿCcJeßJpë|¢C˛¿JOé « ˘¥œLÿ”uGeorge and McCulloch (1993)ò´#Gibbsƒ¢y ßTê{òáwÕA:¥ß¥±ò´ê™l £8XÍ©Ÿ•ƒ ߢ;ùGeorge and McCulloch (1993)J —é{•E›¶_å $é˛ ,ßduk ìd˙™vkUCß^Tê{?1ë|¢C˛ ¿J3¿.ê°ÜGeorge and McCulloch (1993)¥”k©•ò Íä(Jå ±|±˘*: Öcµ ìdC˛¿JGibbsƒÇ5£8ë|¢C˛¿Já⁄ O I 万方数据 Abstract Abstract The stochastic search variable selection proposed by George and McCulloch (1993) is one of the most popular variable selection methods for linear regression models. Many efforts have been proposed in the literature to improve its computa- tional efficiency. However, most of these efforts change its original Bayesian formu- lation, thus the comparisons are not fair. This work focuses on how to improve the computational efficiency of the stochastic search variable selection, but remains its original Bayesian formulation unchanged. The improvement is achieved by devel- oping a new Gibbs sampling scheme different from that of George and McCulloch (1993). A remarkable feature of the proposed Gibbs sampling scheme is that, it samples the regression coefficients from their posterior distributions in a compo- nentwise manner, so that the expensive computation of the inverse of information matrix, which is involved in the algorithm of George and McCulloch (1993), can be avoided. Moreover, since the origi