Projection predictive variable selection
The projpred package performs projection predictive variable selection for generalized linear models. Currently projpred is most easily compatible with rstanarm but other reference models can also be used.
The methods implemented in the package are discussed in detail in
- Juho Piironen, Markus Paasiniemi and Aki Vehtari (2020). Projective inference in high-dimensional problems: prediction and feature selection. Electronic Journal of Statistics, 14(1): 2155-2197.
- Alejandro Catalina, Paul-Christian Bürkner and Aki Vehtari (2020). Projection Predictive Inference for Generalized Linear and Additive Multilevel Models (preprint).
The projective approach is evaluated in comparison to many other methods in
- Juho Piironen and Aki Vehtari (2017). Comparison of Bayesian predictive methods for model selection. Statistics and Computing, 27(3):711-735. doi:10.1007/s11222-016-9649-y. (online).
Currently, the supported models (family objects in R) include Gaussian, Binomial and Poisson families. See the projpred vignette for examples.
To get started see the projpred vignettes.
Install the latest release from CRAN
Install the latest development version from GitHub