https://github.com/thjashin/mirror-stein-samplers

Sampling with Mirrored Stein Operators
https://github.com/thjashin/mirror-stein-samplers

Last synced: 27 days ago
JSON representation

Sampling with Mirrored Stein Operators

• Host: GitHub
• URL: https://github.com/thjashin/mirror-stein-samplers
• Owner: thjashin
• Created: 2021-06-17T20:23:40.000Z (about 3 years ago)
• Default Branch: main
• Last Pushed: 2023-02-14T00:48:16.000Z (over 1 year ago)
• Last Synced: 2024-02-20T14:00:32.234Z (4 months ago)
• Language: Jupyter Notebook
• Size: 20 MB
• Stars: 6
• Watchers: 1
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Lists

• awesome-stars - thjashin/mirror-stein-samplers - Sampling with Mirrored Stein Operators (Jupyter Notebook)

README

        
# Sampling with Mirrored Stein Operators

Code for reproducing the experimental results in [https://arxiv.org/abs/2106.12506](https://arxiv.org/abs/2106.12506)

---

### Update (2023/2/13):

We added a new example of sampling from a uniform distribution in a $[-1, 1]^d$ rectangle in [uniform.ipynb](/uniform.ipynb). 

---

## Requirements

- R >= 4.0.4 (only needed for the non-toy post-selection inference experiments)

- python >= 3.8

```

rpy2 >= 3.4.4

tensorflow >= 2.4.0

tensorflow_probability >= 0.12

numpy >= 1.19.5

scipy >= 1.6.3

matplotlib >= 3.4.1

pandas >= 1.2.4

seaborn >= 0.11.1

scikit-learn >= 0.24.2

tqdm

absl-py

```

## Experiments

### Approximation quality on the simplex

* Sparse Dirichlet: [dirichlet.ipynb](/dirichlet.ipynb)

* Quadratic: [quad.ipynb](/quad.ipynb)

### Confidence intervals for post-selection inference

Note: The `R_HOME` variable must be set correctly before running the scripts.

- Simulation

    - 2D example: [selective_inf.ipynb](/selective_inf.ipynb)

    - Nominal coverage vs. actual coverage: [coverage.py](/coverage.py)

    - Coverage vs. number of samples: [coverage_wrt_k.py](/coverage_wrt_k.py)

    - Plotting: [plot_coverage.ipynb](/plot_coverage.ipynb)

- HIV drug resistance: [hiv.ipynb](/hiv.ipynb)

### Large-scale posterior inference with non-Euclidean geometry

- Run script: [lr.py](/lr.py)

- Plotting: [plot_lr.ipynb](/plot_lr.ipynb)

## Citation

If you find this repository useful, please cite:

```

@article{shi2021sampling,

  title={Sampling with Mirrored {S}tein Operators}, 

  author={Jiaxin Shi and Chang Liu and Lester Mackey},

  journal={International Conference on Learning Representations},

  year={2022}

}

```