Skip to content

Paper experiments

This repository implements the regularized Stein thinning algorithm introduced in Kernel Stein Discrepancy thinning: a theoretical perspective of pathologies and a practical fix with regularization.

If you use this library, please consider citing: 

@article{benard2023kernel,
  title={Kernel Stein Discrepancy thinning: a theoretical perspective of pathologies and a practical fix with regularization},
  author={B{\'e}nard, Cl{\'e}ment and Staber, Brian and Da Veiga, S{\'e}bastien},
  journal={arXiv preprint arXiv:2301.13528},
  year={2023}
}

All numerical experiments presented in the paper can be reproduced using the scripts in the example/ folder.

In particular:

  • Figures 1–3: example/mog_randn.py
  • Section 4 and Appendix 1:
    • Gaussian mixture: example/mog4_mcmc/ and example/mog4_mcmc_dim/
    • Mixture of banana-shaped distributions: example/mobt2_mcmc/ and example/mobt2_mcmc_dim/
    • Bayesian logistic regression: example/logistic_regression.py
  • Supplementary material:
    • Figure 2: example/mog_weight_weights.py
    • Figure 6: example/mog4_mcmc_lambda