Luke Duttweiler, Jonathan Klus, Brent Coull, Sally W. Thurston
{"title":"迹图变厚:非欧几里得空间的 MCMC 诊断方法","authors":"Luke Duttweiler, Jonathan Klus, Brent Coull, Sally W. Thurston","doi":"arxiv-2408.15392","DOIUrl":null,"url":null,"abstract":"MCMC algorithms are frequently used to perform inference under a Bayesian\nmodeling framework. Convergence diagnostics, such as traceplots, the\nGelman-Rubin potential scale reduction factor, and effective sample size, are\nused to visualize mixing and determine how long to run the sampler. However,\nthese classic diagnostics can be ineffective when the sample space of the\nalgorithm is highly discretized (eg. Bayesian Networks or Dirichlet Process\nMixture Models) or the sampler uses frequent non-Euclidean moves. In this\narticle, we develop novel generalized convergence diagnostics produced by\nmapping the original space to the real-line while respecting a relevant\ndistance function and then evaluating the convergence diagnostics on the mapped\nvalues. Simulated examples are provided that demonstrate the success of this\nmethod in identifying failures to converge that are missed or unavailable by\nother methods.","PeriodicalId":501215,"journal":{"name":"arXiv - STAT - Computation","volume":"9 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"The Traceplot Thickens: MCMC Diagnostics for Non-Euclidean Spaces\",\"authors\":\"Luke Duttweiler, Jonathan Klus, Brent Coull, Sally W. Thurston\",\"doi\":\"arxiv-2408.15392\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"MCMC algorithms are frequently used to perform inference under a Bayesian\\nmodeling framework. Convergence diagnostics, such as traceplots, the\\nGelman-Rubin potential scale reduction factor, and effective sample size, are\\nused to visualize mixing and determine how long to run the sampler. However,\\nthese classic diagnostics can be ineffective when the sample space of the\\nalgorithm is highly discretized (eg. Bayesian Networks or Dirichlet Process\\nMixture Models) or the sampler uses frequent non-Euclidean moves. In this\\narticle, we develop novel generalized convergence diagnostics produced by\\nmapping the original space to the real-line while respecting a relevant\\ndistance function and then evaluating the convergence diagnostics on the mapped\\nvalues. Simulated examples are provided that demonstrate the success of this\\nmethod in identifying failures to converge that are missed or unavailable by\\nother methods.\",\"PeriodicalId\":501215,\"journal\":{\"name\":\"arXiv - STAT - Computation\",\"volume\":\"9 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-08-27\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - STAT - Computation\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2408.15392\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - STAT - Computation","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2408.15392","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
The Traceplot Thickens: MCMC Diagnostics for Non-Euclidean Spaces
MCMC algorithms are frequently used to perform inference under a Bayesian
modeling framework. Convergence diagnostics, such as traceplots, the
Gelman-Rubin potential scale reduction factor, and effective sample size, are
used to visualize mixing and determine how long to run the sampler. However,
these classic diagnostics can be ineffective when the sample space of the
algorithm is highly discretized (eg. Bayesian Networks or Dirichlet Process
Mixture Models) or the sampler uses frequent non-Euclidean moves. In this
article, we develop novel generalized convergence diagnostics produced by
mapping the original space to the real-line while respecting a relevant
distance function and then evaluating the convergence diagnostics on the mapped
values. Simulated examples are provided that demonstrate the success of this
method in identifying failures to converge that are missed or unavailable by
other methods.