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:
- 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.
- 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, 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}
}