zeus is a pure-Python implementation of the Ensemble Slice Sampling method.
Doing Bayesian Inference with zeus is both simple and fast, since there is no need to hand-tune any hyperparameters or provide a proposal distribution. The algorithm exhibits excellent performance in terms of autocorrelation time and convergence rate. zeus works out-of-the-box and can scale to multiple CPUs without any extra effort.
For instance, if you wanted to draw samples from a 10-dimensional Gaussian, you would do:
import numpy as np
import zeus
def logprob(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.sampler(logprob, nwalkers, ndim, args=[ivar])
sampler.run(start, nsteps)
print(sampler.chain)
To install zeus using pip run:
pip install git+https://github.com/minaskar/zeus
- See the :doc:`cookbook` page to learn how to perform Bayesian Inference using zeus.
- See the :doc:`faq` page for frequently asked questions about zeus' operation.
- See the :doc:`api` page for detailed API documentation.
Copyright 2019 Minas Karamanis and contributors.
zeus is free software made available under the GPL-3.0 License.
.. toctree:: :maxdepth: 2 :hidden: cookbook faq api notebooks/normal_distribution.ipynb