Year: 2026
Type: preprint
Abstract: Context tree models, or variable-length Markov chains, describe symbolic sequences whose dependence structure varies with context length. Bayesian formulations of these models enable uncertainty quantification and regularization through priors on tree structures and transition probabilities. Despite their theoretical appeal, practical implementations have been fragmented and model-specific. The bacontrees package for R provides a unified framework for Bayesian Context Trees by defining core data structures and a consistent interface for building, modifying and manipulating trees. Rather than implementing a single inference method, bacontrees offers the computational foundation upon which diverse Bayesian and frequentist algorithms can be developed. This separation of representation and inference promotes reproducibility and facilitates experimentation with new modeling strategies. The package aims to support applications in information theory, linguistics, and other domains where context-based probabilistic modeling is central.
Source: arXiv (Cornell University)
Authors: Freguglia, Victor, Paulichen, Thiago
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225188171