Code supporting “Bayesian Generalized Horseshoe Estimation of Generalized Linear Models”
Paper published in the Proceedings of the European Conference on Machine Learning, 2019.
The arxiv pre-print (with additional text/experimental results) is also available.
The code required to recreate the experimental results presented in this paper is available in download form from here. The main folder contains a branch of the Bayesreg package which implements the following additional functionality:
- Poisson regression using the mGrad-1, mGrad-2, pre-conditioned Crank-Nicolson and generalized elliptical slice sampling algorithms;
- binomial regression using the mGrad-1 and mGrad-2 samplers;
- generalized horseshoe priors using the rejection sampler, inverse gamma-inverse gamma and gamma-gamma samplers.
The experiments\ subfolder contains the MATLAB and R scripts (for the NUTS sampler using RStan) required to run the experiments and recreate Tables 2 and 3 in the paper. Note, the paper was produced using a Surface Pro 2016, so timings may (will) vary depending on the particular hardware you are using.