README.md

Stein Pi-Importance-Sampling

This code accompanies the paper "Stein $\Pi$-Importance Sampling". It was written in Python 3.10.4 and also tested in Python 3.10.4.

This package can be installed directly by pip in the root directory

python -m pip install .

Some of the experiments in this paper rely on the following packages

• Stein Thinning
• posteriordb v0.4.0
• posteriordb-python
• BridgeStan v1.0.2
• qpsolvers v3.4.0
• proxsuite v0.3.7
• Wasserstein v1.1.0

The original package for Stein Thinning has been modified, and it can be installed in the following way

cd stein_thinning
python -m pip install .

To reproduce the experiment, first navigate to the "experiment" folder

cd experiment

and then clone the posteriordb using

git clone https://github.com/stan-dev/posteriordb.git

run the relevant script finally. e.g. "gene_store_wksd.py" will run the benchmarking on PosteriorDB to store the numpy data files. "plot_KSDCurve.py" will plot the corresponding results.