zeus is a Python implementation of the Ensemble Slice Sampling method.

• Fast & Robust Bayesian Inference,
• Efficient Markov Chain Monte Carlo (MCMC),
• Black-box inference, no hand-tuning,
• Excellent performance in terms of autocorrelation time and convergence rate,
• Scale to multiple CPUs without any extra effort.

Example

For instance, if you wanted to draw samples from a 10-dimensional Gaussian, you would do something like:

import zeus
import numpy as np

def log_prob(x, ivar):
    return - 0.5 * np.sum(ivar * x**2.0)

nsteps, nwalkers, ndim = 1000, 100, 10
ivar = 1.0 / np.random.rand(ndim)
start = np.random.randn(nwalkers,ndim)

sampler = zeus.EnsembleSampler(nwalkers, ndim, log_prob, args=[ivar])
sampler.run_mcmc(start, nsteps)
chain = sampler.get_chain(flat=True)

Documentation

Read the docs at zeus-mcmc.readthedocs.io

Installation

To install zeus using pip run:

pip install zeus-mcmc

Attribution

Please cite the following papers if you found this code useful in your research:

@article{zeus,
        title={zeus: A Python Implementation of the Ensemble Slice Sampling method},
        author={Minas Karamanis and Florian Beutler},
        year={2021},
        note={in prep}
    }

@article{ess,
      title={Ensemble Slice Sampling},
      author={Minas Karamanis and Florian Beutler},
      year={2020},
      eprint={2002.06212},
      archivePrefix={arXiv},
      primaryClass={stat.ML}
}

Licence

Copyright 2019-2021 Minas Karamanis and contributors.

zeus is free software made available under the GPL-3.0 License. For details see the LICENSE file.