Hierarchical models: Local proposal variances for RWM-within-Gibbs and MALA-within-Gibbs

详细信息    查看全文

• 作者：Mylè ; ne Bé ; dard ^{bedard@dms.umontreal.ca}
• 关键词：Efficiency ; Gaussian proposal distribution ; Inhomogeneous proposal variances ; Langevin diffusion ; Location/scale parameters ; MALA
• 刊名：Computational Statistics & Data Analysis
• 出版年：2017
• 出版时间：May 2017
• 年：2017
• 卷：109
• 期：Complete
• 页码：231-246
• 全文大小：917 K
• 卷排序：109

文摘

Locally optimal proposal variances are introduced for RWM-within-Gibbs algorithms. These locally optimal tunings are shown to theoretically outperform constant ones. Similar state-dependent step sizes are discussed for MALA-within-Gibbs samplers. MALA-within-Gibbs constitutes an efficient, yet computationally affordable option. Efficiency of local tunings depends on the variability in the hierarchical target.