路径抽样法在贝叶斯模型选择中的应用

Application of the Path Sampling to the Model Selection by Means of Bayesian Method

为采用贝叶斯分析方法解决模型选择问题 ,针对传统的Box Cox模型线性与非线性的选择问题 ,将路径抽样法应用于贝叶斯因子的计算 ,引进一个连续的路径参数并且假定它满足一定的概率分布 ,利用该路径参数连接待选择的模型 ,使计算贝叶斯因子的工作主要集中于马尔可夫链蒙特卡洛 (MCMC)抽样上 ,从而简化了使用贝叶斯分析方法的计算过程 ,实现了路径抽样法在模型选择中的具体应用 .

In order to solve the problem of model selection by means of Bayesian method and in view of the selection of the traditional linear Box-Cox model and the nonlinear Box-Cox model, the method of path sampling was adopted to calculate the Bayesian factors, A continuous path parameter with a certain assumed probability distribution was then introduced to connect different models to be selected. Thus, the main calculation of the Bayesian factors was focused on the sampling of Markov Chain Monte Carlo (MCMC) and the calculation in the analytical process by means of Bayesian method was simplified. It is concluded that the method of path sampling can be applied to model selection in practice.