Dynamic nested sampling

This repository contains the code used for making the results and plots in the dynamic nested sampling paper (Higson et. al, 2019).

If you have any questions then feel free to email [email protected]. However, note that this is research code and is not actively maintained.

Requirements

The results in the paper (except those in Section 6.2) were run in Python 3.6 using:

• perfectns v2.0.1;
• dyPolyChord v0.0.0;
• PolyChord v1.14;
• nestcheck v0.1.0.

Later versions of the above software should give the same results. The signal reconstruction example in Section 6.2 were added in a later draft and used:

• dyPolyChord v0.0.5;
• PolyChord v1.15;
• nestcheck v0.1.9;
• bsr v0.0.0.

Aside from dependencies of the above modules (such as scipy, numpy, pandas and matplotlib), the only other package required is getdist; this is used for the triangle plots of Gaussian mixture posteriors.

Code

The code is divided into two Jupyter notebooks:

• perfectns_paper_results.ipynb contains perfect nested sampling results and plots. Some results tables are cached in the perfectns_results directory so this should not take long to run. Alternatively you can reproduce the nested sampling run data yourself using make_perfectns_results.py, although this is quite computationally intensive. numpy random seeding is used by default, so all results should be reproducible.
• dypolychord_paper_results.ipynb contains the code used for the Gaussian mixture model and signal reconstruction results. It requires nested sampling runs which can be generated using make_gaussian_mix_results.py and make_fit_results.py; see the module docstrings for more details, including about the random seeding used. This can be done with either the Python or C++ versions of the likelihood (the results are identical up to numerical precision errors, but latter runs much faster).

Attribution

If it is useful for your research, then please cite the dynamic nested sampling paper. The BibTeX is:

@article{Higson2019dynamic,
author={Higson, Edward and Handley, Will and Hobson, Michael and Lasenby, Anthony},
title={Dynamic nested sampling: an improved algorithm for parameter estimation and evidence calculation},
year={2019},
journal={Statistics and Computing},
volume={29},
number={5},
pages={891--913},
doi={10.1007/s11222-018-9844-0},
url={https://doi.org/10.1007/s11222-018-9844-0},
archivePrefix={arXiv},
arxivId={1704.03459}}