R package for fitting Brand: Bayesian Robust Adaptive Novelty Detector. Brand is a two-stage Bayesian semiparametric model. In the first stage, it learns the main characteristics of the known classes from the labeled dataset using robust procedures. In the second phase, a Bayesian semiparametric mixture of known groups and a novelty term is fitted to the test set. Training insights is used to elicit informative priors for the known components. The novelty term is instead captured via a flexible Dirichlet Process mixture.
The package provides efficient Slice Samplers for handling both multivariate and functional data.
The repository is associated with the paper Denti, Cappozzo, Greselin (2021) A two-stage Bayesian semiparametric model for novelty detection with robust prior information. https://doi.org/10.1007/s11222-021-10017-7
For replicating the simulated and real data analyses reported in the paper, please referer to this repository.
You can install the development version of brand from GitHub with:
# install.packages("remotes")
remotes::install_github("Fradenti/Brand")