barrett

barrett is a software package meant to process and visualise the output of the nested sampling algorithm MultiNest. There are several packages already on the market for this, but barrett's main differential feature is out-of-core processing so the code can handle very large datasets.

Specific technologies: HDF5, and Python (h5py, scipy, numpy, matplotlib).

Usage

barrett is split into four submodules:

• barrett.data implements methods for modifying data (e.g. log, change units) or calculate depended variables (e.g. mean squark mass)
• barrett.posterior is for calculating and plot the one or two dimensional marginal posterior distribution.
• barrett.profilelikelihood is for calculating and plot the one or two dimensional profile likelihood.
• barrett.util contain various utility functions most notable convert_chain() which converts the plain text MultiNest output to the HDF5 format used by barrett.

As for parallelisation; writing to the same hdf5 file is strongly discouraged. Reading the file is however perfectly fine. So posterior/profilelikelihood module is perfectly parallelisable. The code itself is not parallelised, instead I recommend using Python's multiprocessing module to producing several plots in parallell. In most system tested the plotting is CPU bound, your mileage may vary.

Installation

Barrett is available on PyPI and can be installed using pip

pip install barrett

Cite

If you use barrett in your research please cite arXiv:1608.00990:

@article{2016arXiv160800990L,

author = {Liem, Sebastian}, title = "{Barrett: out-of-core processing of MultiNest output}", archivePrefix = "arXiv", eprint = {1608.00990}, primaryClass = "stat.CO", year = 2016

}

Example

Please check the example directory for plot.py for an, you guessed it, example.