The projpred R package performs the projection predictive variable selection for generalized linear models as well as generalized linear and additive multilevel models. The package is compatible with the rstanarm and brms packages, but custom reference models can also be used.
The projection predictive variable selection is based on the ideas of
Goutis and Robert (1998) and Dupuis and Robert (2003). The methods
implemented in projpred are described in detail in
Piironen et al. (2020) and Catalina et al. (2020). They are evaluated in
comparison to many other methods in Piironen and Vehtari (2017). Type
citation("projpred")
in R (or see the CITATION
file) for details on how to cite projpred.
Currently, the supported response distributions (objects of class
family
in R) are gaussian()
,
binomial()
(via the brms package,
brms::bernoulli()
is also supported), and
poisson()
.
The vignettes (currently, there is only a single one) illustrate how to use the projpred functions in conjunction. Details on the projpred functions as well as some shorter examples may be found in the documentation.
There are two ways for installing projpred: from CRAN or from GitHub. The GitHub version might be more recent than the CRAN version, but the CRAN version might be more stable.
install.packages("projpred")
This requires the devtools package, so if necessary, the following code will also install devtools (from CRAN):
if (!requireNamespace("devtools", quietly = TRUE)) {
install.packages("devtools")
}::install_github("stan-dev/projpred", build_vignettes = TRUE) devtools
To save time, you may omit build_vignettes = TRUE
.
Catalina, A., Bürkner, P.-C., and Vehtari, A. (2020). Projection predictive inference for generalized linear and additive multilevel models. arXiv:2010.06994. URL: https://arxiv.org/abs/2010.06994.
Dupuis, J. A. and Robert, C. P. (2003). Variable selection in qualitative models via an entropic explanatory power. Journal of Statistical Planning and Inference, 111(1-2):77–94. DOI: 10.1016/S0378-3758(02)00286-0.
Goutis, C. and Robert, C. P. (1998). Model choice in generalised linear models: A Bayesian approach via Kullback–Leibler projections. Biometrika, 85(1):29–37.
Piironen, J. and Vehtari, A. (2017). Comparison of Bayesian predictive methods for model selection. Statistics and Computing, 27(3):711-735. DOI: 10.1007/s11222-016-9649-y.
Piironen, J., Paasiniemi, M., and Vehtari, A. (2020). Projective inference in high-dimensional problems: Prediction and feature selection. Electronic Journal of Statistics, 14(1):2155-2197. DOI: 10.1214/20-EJS1711.