A comparison of computational approaches for maximum likelihood estimation of the Dirichlet parameters on high-dimensional data

Likelihood estimates of the Dirichlet distribution parameters can be obtained only through numerical algorithms. Such algorithms can provide estimates outside the correct range for the parameters and/or can require a large amount of iterations to reach convergence. These problems can be aggravated i...

Descripción completa

Detalles Bibliográficos
Autores: Giordan, Marco, Wehrens, Ron
Tipo de recurso: artículo
Fecha de publicación:2015
País:España
Institución:Universitat Autònoma de Barcelona
Repositorio:Dipòsit Digital de Documents de la UAB
Idioma:inglés
OAI Identifier:oai:ddd.uab.cat:132930
Acceso en línea:https://ddd.uab.cat/record/132930
Access Level:acceso abierto
Palabra clave:Levenberg-marquardt algorithm
Re-parametrization
Starting values
Metabolomics data
Descripción
Sumario:Likelihood estimates of the Dirichlet distribution parameters can be obtained only through numerical algorithms. Such algorithms can provide estimates outside the correct range for the parameters and/or can require a large amount of iterations to reach convergence. These problems can be aggravated if good starting values are not provided. In this paper we discuss several approaches that can partially avoid these problems providing a good trade-off between efficiency and stability. The performances of these approaches are compared on high-dimensional real and simulated data.