stagedtrees Staged Event Trees

stagedtrees is an R package that implements staged event trees, a class of probabilistic graphical models for categorical random variables. Staged trees generalise discrete Bayesian networks: they can represent asymmetric, context-specific conditional independences, so that a dependence may hold for some groups of units and vanish for others.

With stagedtrees it is possible to learn (stratified) staged event trees from data with score-based and clustering algorithms, use them to compute probabilities, make predictions, visualize staged trees and chain event graphs, and compare different models. The package also implements simple and context-specific staged trees, staged tree classifiers and asymmetry-labeled DAGs.

The development version on GitHub, which will be released on CRAN as version 2.4.0, adds learning from incomplete data and tools for treatment effect estimation: potential outcomes, propensity-score stratification and checks of the positivity assumption.

The book

Staged Tree Models: With Applications in R, by Manuele Leonelli and Gherardo Varando, is forthcoming with CRC Press. See the book page for details.

Installation

# Stable version from CRAN
install.packages("stagedtrees")

# Development version from GitHub
remotes::install_github(
  "stagedtrees/stagedtrees"
)

A first example

library("stagedtrees")

tree <- Titanic |>
  full() |>
  stages_bhc() |>
  stndnaming(uniq = TRUE)

prob(tree, c(Survived = "Yes"),
     conditional_on = c(Age = "Adult"))
#> [1] 0.3107124

plot(tree, col = "stages")

Citation

To cite stagedtrees in publications use:

Carli F, Leonelli M, Riccomagno E, Varando G (2022). The R package stagedtrees for structural learning of stratified staged trees. Journal of Statistical Software, 102(6), 1-30. doi:10.18637/jss.v102.i06

The BibTeX entry is available on the publications page.

news