VIABEL: Variational Inference and Approximation Bounds that are Efficient and Lightweight

VIABEL is a library (still in early development) that provides two types of functionality:

1. A lightweight, flexible set of methods for variational inference that is agnostic to how the model is constructed. All that is required is a log density and its gradient.
2. Methods for computing bounds on the errors of the mean, standard deviation, and variance estimates produced by a continuous approximation to an (unnormalized) distribution. A canonical application is a variational approximation to a Bayesian posterior distribution.

Documentation

For examples and API documentation, see readthedocs.

Installation

You can install the latest stable version using pip install viabel. Alternatively, you can clone the repository and use the master branch to get the most up-to-date version.

Citing VIABEL

If you use this package for diagnostics, please cite:

Validated Variational Inference via Practical Posterior Error Bounds. Jonathan H. Huggins, Mikołaj Kasprzak, Trevor Campbell, Tamara Broderick. In Proc. of the 23rd International Conference on Artificial Intelligence and Statistics (AISTATS), Palermo, Italy. PMLR: Volume 108, 2020.

The equivalent BibTeX entry is:

@inproceedings{Huggins:2020:VI,
  author = {Huggins, Jonathan H and Kasprzak, Miko{\l}aj and Campbell, Trevor and Broderick, Tamara},
  title = {{Validated Variational Inference via Practical Posterior Error Bounds}},
  booktitle = {Proc. of the 23rd International Conference on Artificial Intelligence and Statistics (AISTATS)},
  year = {2020}
}

If you use this package for variational inference, please cite:

Robust, Automated, and Accurate Black-box Variational Inference. Manushi Welandawe, Michael Riis Andersen, Aki Vehtari, Jonathan H. Huggins. arXiv:2203.15945 [stat.ML] (2022).

The equivalent BibTeX entry is:

@article{Welandawe:2022:BBVI,
  author = {Welandawe, Manushi and Andersen, Michael Riis and Vehtari, Aki and Huggins, Jonathan H},
  title = {Robust, Automated, and Accurate Black-box Variational Inference},
  journal = {arXiv},
  volume = {arXiv:2203.15945 [stat.ML]},
  year = {2022}
}