INVESTIGADORES
MARTIN Osvaldo Antonio
artículos
Título:
Bayesian additive regression trees for probabilistic programming
Autor/es:
MIRIANA QUIROGA; GARAY, PABLO G.; JUAN MANUEL ALONSO; JUAN M LOYOLA; MARTÍN, OSVALDO ANTONIO
Revista:
arxiv
Editorial:
arXiv
Referencias:
Año: 2022
ISSN:
2331-8422
Resumen:
Bayesian additive regression trees (BART) is a non-parametric method toapproximate functions. It is a black-box method based on the sum of many treeswhere priors are used to regularize inference, mainly by restricting trees´learning capacity so that no individual tree is able to explain the data, butrather the sum of trees. We discuss BART in the context of probabilisticprogramming languages (PPLs), specifically we introduce a BART implementationextending PyMC, a Python library for probabilistic programming. We present afew examples of models that can be built using this probabilisticprogramming-oriented version of BART, discuss recommendations for samplediagnostics and selection of model hyperparameters, and finally we close withlimitations of the current approach and future extensions.
