(Bayesian) Additive Voronoi Tessellations
Implements the Bayesian Additive Voronoi Tessellation
model for non-parametric regression, classification and machine learning as
introduced in Stone and Gosling (2025) .
This package provides a flexible alternative to BART (Bayesian Additive
Regression Trees) using Voronoi tessellations instead of trees. Users can
fit Bayesian regression and probit classification models, estimate the
associated posterior distributions and make predictions. It is particularly
useful for spatial data analysis, machine learning, complex function
approximation and Bayesian modelling where the underlying structure is
unknown.
AddiVortes: (Bayesian) Additive Voronoi Tessellations
Overview
AddiVortes implements the Bayesian Additive Voronoi Tessellation model for machine learning regression, classification and non-parametric statistical modelling. This R package provides a flexible alternative to BART (Bayesian Additive Regression Trees), using Voronoi tessellations instead of trees for spatial partitioning.
Key Features
- Machine Learning Regression: Advanced Bayesian regression modelling for complex datasets
- Classification: Binary and multi-category probit models, chosen automatically from the response
- Alternative to BART: Uses Voronoi tessellations instead of trees for more flexible spatial modelling
- Spatial Data Analysis: Excellent for geographic and spatial datasets
- Non-parametric Modelling: No assumptions about functional form
- Bayesian Framework: Full posterior inference with uncertainty quantification
- Complex Function Approximation: Captures non-linear relationships and interactions
Applications
AddiVortes is particularly well-suited for:
- Spatial regression and geographic data analysis
- Binary and multi-category classification with posterior class probabilities
- Machine learning tasks requiring interpretable models
- Non-parametric regression where the functional form is unknown
- Bayesian modelling with uncertainty quantification
- Complex surface modelling and function approximation
- Alternative to BART for researchers seeking different ensemble approaches
Installation
You can install the latest version of AddiVortes from GitHub with:
pak::pak("johnpaulgosling/AddiVortes")
Quick Start
library(AddiVortes)
# Load your data
# X <- your_predictors
# y <- your_response
# Fit the AddiVortes model (regression or classification, from y)
# model <- AddiVortes(y, X)
# Make predictions
# predictions <- predict(model, newdata = X_test)
# For classification, type = "response" gives probabilities
# and type = "class" gives labels
Documentation
Vignettes:
See also the function reference.
Comparison with BART
While BART (Bayesian Additive Regression Trees) uses tree-based partitioning, AddiVortes uses Voronoi tessellations, which can provide:
- More natural spatial partitioning
- Flexible geometric boundaries
- Alternative ensemble approach for machine learning
- Enhanced performance on spatial data
Cite Us
If you use this package in your research, please cite:
citation("AddiVortes")
References
Stone, A. and Gosling, J.P. (2025). AddiVortes: (Bayesian) additive Voronoi tessellations. Journal of Computational and Graphical Statistics, 34, 859–71. doi
Keywords
Bayesian machine learning, BART alternative, Voronoi tessellation, spatial regression, non-parametric regression, ensemble methods, statistical modelling, R package