aglebov / gradient-free-mcmc-postprocessing

Gradient-free optimal postprocessing of MCMC output

Gradient-Free Optimal Postprocessing of MCMC Output

The project aims to extend the work in

1. Riabiz, M., Chen, W. Y., Cockayne, J., Swietach, P., Niederer, S. A., Mackey, L., Oates, C. J. (2022). Optimal thinning of MCMC output. Journal of the Royal Statistical Society Series B: Statistical Methodology, 84(4), 1059-1081.

by implementing the proposals presented in

2. Fisher, M. A., Oates, C. (2022). Gradient-free kernel Stein discrepancy. arXiv preprint arXiv:2207.02636.

and

3. Huang, C., Joseph, V. R. (2023). Enhancing Sample Quality through Minimum Energy Importance Weights. arXiv preprint arXiv:2310.07953.

We replicate the results presented from [1] for the Lotka-Volterra model in code/Lotka-Volterra.ipynb.

code/Gradient_free_thinning.ipynb demonstrates using gradient-free kernel Stein density as proposed in [2] for a bivariate Gaussian mixture.

The code/examples directory also contains several examples of using the relevant Python packages.