论文部分内容阅读
关于聚类问题现在已经有很多方法可以实现,但大多数基于有限混合模型的聚类方法需要预先估计聚类的个数,因而聚类的准确性和泛化性会受到一定影响。本文则提出了一种基于无线混合模型——中国餐馆模型(CRP)的聚类方法,CRP模型是Dirichlet过程的一种表示方法,基于Dirichlet无线混合模型找出其后验分布,利用Gibbs采样MCMC方法估计出模型中各个参数以及潜在的聚类个数,并在MATLAB环境下进行一个小实验来验证聚类的效果。
There are many methods to solve the clustering problem nowadays. However, most of the clustering methods based on the finite mixture model need to estimate the number of clusters in advance, so the accuracy and generalization of the clustering will be affected. In this paper, we propose a clustering method based on wireless hybrid model - Chinese restaurant model (CRP). CRP model is a representation method of Dirichlet process. Based on the Dirichlet wireless hybrid model, the posterior distribution is found, and the MCMC The method estimates each parameter in the model and the number of potential clusters, and carries out a small experiment in MATLAB environment to verify the clustering effect.