nessai (/ˈnɛsi/): Nested Sampling with Artificial Intelligence

nessai is a nested sampling algorithm for Bayesian Inference that incorporates normalisings flows. It is designed for applications where the Bayesian likelihood is computationally expensive.

nessai can be installed using pip:

pip install nessai

or via conda

conda install -c conda-forge -c pytorch nessai

By default the version of PyTorch will not necessarily match the drivers on your system, to install a different version with the correct CUDA support see the PyTorch homepage for instructions: https://pytorch.org/.

As of bilby version 1.1.0, nessai is now supported by default but it is still an optional requirement. See the bilby documentation for installation instructions for bilby

See the examples included with nessai for how to run nessai via bilby.

Documentation is available at: nessai.readthedocs.io

Please see the guidelines here.

The core nested sampling code, model design and code for computing the posterior in nessai was based on cpnest with permission from the authors.

The normalising flows implemented in nessai are all either directly imported from nflows or heavily based on it.

Other code snippets that draw on existing code reference the source in their corresponding doc-strings.

The authors also thank Christian Chapman-Bird, Laurence Datrier, Fergus Hayes, Jethro Linley and Simon Tait for their feedback and help finding bugs in nessai.

If you find nessai useful in your work please cite the DOI for this code and our paper:

@software{nessai,
  author       = {Michael J. Williams},
  title        = {nessai: Nested Sampling with Artificial Intelligence},
  month        = feb,
  year         = 2021,
  publisher    = {Zenodo},
  version      = {latest},
  doi          = {10.5281/zenodo.4550693},
  url          = {https://doi.org/10.5281/zenodo.4550693}
}

@article{PhysRevD.103.103006,
  title = {Nested sampling with normalizing flows for gravitational-wave inference},
  author = {Williams, Michael J. and Veitch, John and Messenger, Chris},
  journal = {Phys. Rev. D},
  volume = {103},
  issue = {10},
  pages = {103006},
  numpages = {19},
  year = {2021},
  month = {May},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevD.103.103006},
  url = {https://link.aps.org/doi/10.1103/PhysRevD.103.103006}
}