Overview

seqinf is a Python library for performing sequential simulation-based inference. It aims to complement the popular sbi package by providing top-level abstractions that enable running methods like SNPE/SNLE/etc without additional boilerplate. It also provides additional inference diagnostics and support for Bayesian NDEs and active methods (e.g., ASNPE).

Install

Installation is available through PyPI:

pip install seqinf

Dependencies

Note the version clash between sbibm's requirement for sbi, and the latest sbi. This can more or less be ignored, and if this presents an issue when installing locally, just make sure to install sbi after sbibm.

Citing this package

Although seqinf is intended for general pipelines, it was principally written to wrap ASNPE. If you use this library in your work, please cite the following:

@misc{griesemer2024activesequentialposteriorestimation,
      title={Active Sequential Posterior Estimation for Sample-Efficient Simulation-Based Inference}, 
      author={Sam Griesemer and Defu Cao and Zijun Cui and Carolina Osorio and Yan Liu},
      year={2024},
      eprint={2412.05590},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2412.05590}, 
}