Code for the AISTATS 2022 paper titled "Discovering Inductive Bias with Gibbs Priors: A Diagnostic Tool for Approximate Bayesian Inference" by Luca Rendsburg, Agustinus Kristiadi, Philipp Hennig, and Ulrike von Luxburg.
- Install CUDA >=11 and CuDNN >=8.2: https://gist.github.com/kmhofmann/cee7c0053da8cc09d62d74a6a4c1c5e4.
- Install Miniconda: https://docs.conda.io/en/latest/miniconda.html.
- Create a virtual environment:
conda create --name {ENV_NAME} --file requirements.txt
, for an arbitrary{ENV_NAME}
. - Activate:
conda activate {ENV_NAME}
.
- Choose a folder in experiments/ to reproduce the experiments for the log-normals model in Section 5.1 (experiments/log_normal), for the volatility model in Section 5.2 (experiments/volatility), for the baseline (experiments/baseline), or for convergence monitoring (experiments/convergence)
- Run the
run_%.ipynb
file(s) to generate the data, which are saved in the corresponding folder inresults/
- Run the
eval_%.ipynb
file to create the figures used in the paper.
@inproceedings{rendsburg2022gibbs-prior,
title={Discovering Inductive Bias with {G}ibbs Priors: A Diagnostic Tool for Approximate {B}ayesian Inference},
author={Rendsburg, Luca and Kristiadi, Agustinus and Hennig, Philipp and von Luxburg, Ulrike},
booktitle={AISTATS},
year={2022}
}