nestcheck

Nested sampling is a popular numerical method for Bayesian computation, which simultaneously generates samples from the posterior distribution and an estimate of the Bayesian evidence for a given likelihood and prior. nestcheck provides Python utilities for analysing samples produced by nested sampling, and estimating uncertainties on nested sampling calculations (which have different statistical properties to calculations using other numerical methods). This includes implementations of the diagnostic tests and plots described in:

• “Sampling errors in nested sampling parameter estimation” (Higson et al., 2018);

• “nestcheck: diagnostic tests for nested sampling calculations” (Higson et al., 2019).

To get started, see the installation instructions and the quickstart demo. More examples of nestcheck’s use can be found in the code used to make the results and plots in Higson et al. (2019) at https://github.com/ejhigson/diagnostic.

Compatible nested sampling software

Currently nestcheck.data_processing has functions to load results from:

• PolyChord >=v1.14;

• MultiNest >=v3.11;

• dyPolyChord (same output format as PolyChord);

• dynesty;

• perfectns.

You can easily add input functions for other nested sampling software packages. Note that nestcheck requires information about the iso-likelihood contours within which dead points were sampled (“born”), which is needed to split nested sampling runs into their constituent single live point runs (“threads”); see Higson et al. (2018) for more details. nestcheck is fully compatible with dynamic nested sampling, in which the number of live points is varied to increase calculation accuracy.

Documentation contents

• Installation
  • Dependencies
  • Tests
• Quickstart demo
  • Loading nested sampling runs
  • Evidence and parameter estimation calculations from runs
  • Bootstrap sampling error estimates
  • Diagrams of uncertainties on posterior distributions using bootstrap resamples
  • Diagrams of samples in log X
  • Calculating errors due to implementation-specific effects
  • Tests for implementation specific effects using only 2 nested sampling runs
• Detailed API documentation
  • data_processing
  • ns_run_utils
  • error_analysis
  • diagnostics_tables
  • plots
  • estimators
  • io_utils
  • parallel_utils
  • pandas_functions
  • dummy_data
  • write_polychord_output

Attribution

If nestcheck is useful for your academic research, please cite the three papers introducing the software and the methods it implements. The BibTeX is:

@article{higson2019diagnostic,
title={nestcheck: diagnostic tests for nested sampling calculations},
author={Higson, Edward and Handley, Will and Hobson, Mike and Lasenby, Anthony},
journal={Monthly Notices of the Royal Astronomical Society},
year={2019}
volume={483},
number={2},
pages={2044--2056},
doi={10.1093/mnras/sty3090},
url={http://doi.org/10.1093/mnras/sty3090},
archivePrefix={arXiv},
arxivId={1804.06406}}

@article{higson2018sampling,
title={Sampling Errors in Nested Sampling Parameter Estimation},
author={Higson, Edward and Handley, Will and Hobson, Mike and Lasenby, Anthony},
year={2018}
journal={Bayesian Analysis},
number={3},
volume={13},
pages={873--896},
doi={10.1214/17-BA1075},
url={https://doi.org/10.1214/17-BA1075}}

@article{higson2018nestcheck,
title={nestcheck: error analysis, diagnostic tests and plots for nested sampling calculations},
author={Higson, Edward},
year={2018},
journal={Journal of Open Source Software},
number={29},
pages={916},
volume={3},
doi={10.21105/joss.00916},
url={http://joss.theoj.org/papers/10.21105/joss.00916}}

Changelog

The changelog for each release can be found at https://github.com/ejhigson/nestcheck/releases.

Contributions

Contributions are welcome! Development takes place on github:

• source code: https://github.com/ejhigson/nestcheck;

• issue tracker: https://github.com/ejhigson/nestcheck/issues.

When creating a pull request, please try to make sure the tests pass and use numpy-style docstrings.

If you have any questions or suggestions please get in touch (e.higson@mrao.cam.ac.uk).

Authors & License

Copyright 2018-Present Edward Higson and contributors (MIT license).