torchns

Nested sampling in torch

What is it?

torchns is a nested sampler with a slice sampling exploration scheme based on torch library. It is designed to be integrated with the swyft sequential simulation-based inference code.

Its main features are:

• Vectorized evaluations of slice sampling chains to draw new live points.
• Vectorized evaluations of the log-likelihood.
• Functionality to define constrained prior regions, useful for sequential simulation-based inference applications.

Installation

• Change directory to wherever you would like to store the library, then run:

git clone https://github.com/undark-lab/torchns.git # for https client
[or git clone [email protected]:undark-lab/torchns.git # for ssh client]

• Making sure that the desired python environment is active, run the following installation code:

cd torchns/
pip install .

• This will install torchns in the current python environment that is active on your system and will be available via import torchns

Further information

• Source code: https://github.com/undark-lab/torchns
• Example usage: https://github.com/undark-lab/torchns/examples
• Support & discussion: https://github.com/undark-lab/torchns/discussions
• Bug reports: https://github.com/undark-lab/torchns/issues
• Related paper: torchns was first introduced in arxiv:2308.08597.

Release Details

• v0.0.1 | August 2023 | Initial release based on arxiv:2308.08597.

Relevant other nested sampling packages

• jaxns is a JAX based nested sampler.
• proxnest implements the proximal nested sampling algorithm.
• Polychord is a nested sampler that uses slice sampling.
• MultiNest is a nested sampler that uses ellipsoidal sampling.
• dynesty is a dynamic nested sampler.