Simulation-based regularized logistic regression
R package for regularized logistic
regression by Gibbs sampling. Ssubtly different MCMC schemes are tailored to the
data type (binary v. binomial, say) and the desired estimator
(regularized MLE, or Bayesian maximum a posteriori/posterior mean, etc.)
through a unified interface.
The package provides:
- logistic regression under lasso and ridge penalties via a coupled pair of data augmentation schemes
- power-posterior analysis for calculating MAP estimators
- Binomial logistic regression without data expansion (i.e., via a binarization)
New! The latest version supports:
- polychotomous (3+) outputs, i.e., multinomial logistic
- faster computation for sparse design matrices
Obtaining the package
Rfrom cran.r-project.org by selecting the version for your operating system.
- Install the
bootpackages, from within
R> install.packages(c("reglogit", "mvtnorm", "boot"))
- Optionally, install the
- Load the library as you would for any
- See the package documentation.
version of the reference manual, or help pages, is also available.
The help pages can be accessed from within
R. Try starting with:
R> ?reglogit # follow the examples
- For details on the use of this software for estimating player abilities in hockey, please see the paper linked below and our Chicago Hockey Analytics page.
- Simulation-based regularized logistic regression (2012) with Nicholas Polson; Bayesian Analysis, 7(3), pp. 567-590; preprint on arXiv:1005.3430
- Estimating player contribution in hockey with regularized logistic regression (2013) with Shane Jensen, and Matt Taddy. Journal of Quantitative Analysis in Sports, 9(1), pp. 97-111; preprint on arXiv:1209.5026