基于Gev的贝叶斯模型研究及应用

发布时间：2018-01-19 12:10

本文关键词： 基于GEV的贝叶斯模型分布式贝叶斯算法人类行为海量数据　出处：《华东理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：广义极值分布(Generalized Extreme Value Distribution)作为极值变量的渐进分布具有参数少、灵活性高等优点,它可以通过设定形状参数或尺寸参数的值来实现分布对称性或者偏度的变化。GEV分布的这些优势可以很好地用来弥补一些模型的不足之处。另一方面,贝叶斯参数估计的优势在于先验信息的有效利用以及在复杂模型的参数估计问题上可利用的诸多算法,如ABC、MH等。通过结合GEV与贝叶斯估计方法,本文着重研究了 GEV作为连接函数的二元响应变量模型的优势及其贝叶斯方法改进,通过提出分布式贝叶斯方法克服了海量数据处理的低效问题,并将其应用到海量的游戏数据中,实际的预测误差仅为3.5%。本文还讨论了基于GEV的贝叶斯分层模型的优势以及几种变形,并通过手机数据这个载体去研究人类行为。
[Abstract]:Generalized Extreme Value distribution has few parameters as the asymptotic distribution of extreme variables. High flexibility and other advantages. It can realize the distribution symmetry or skewness change by setting the value of shape parameter or dimension parameter. These advantages can be used to make up for the shortcomings of some models. On the other hand. The advantage of Bayesian parameter estimation lies in the efficient use of prior information and the many algorithms that can be used in the problem of parameter estimation of complex models, such as GEV and Bayesian estimation. This paper focuses on the advantages of the binary response variable model of GEV as a connection function and the improvement of the Bayesian method, and proposes a distributed Bayesian method to overcome the inefficient problem of massive data processing. The actual prediction error is only 3.5. This paper also discusses the advantages of Bayesian hierarchical model based on GEV and several kinds of deformation. And through the mobile phone data this carrier to study human behavior.
【学位授予单位】：华东理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O212.8

【相似文献】