discoverysamplers

Bridge interfaces connecting Discovery models to various sampling algorithms.

Author	Jonas El Gammal
Source	Source code on GitHub
Documentation	Documentation on Read the Docs
License	MIT
Support	For questions, use GitHub Issues or drop me an email
Installation	Clone from GitHub

Overview

discoverysamplers provides lightweight wrappers that adapt Discovery-style models to the APIs expected by different sampling backends.

Supported Samplers

Sampler	Type	Description
Eryn	MCMC	Ensemble MCMC with parallel tempering and reversible-jump support
Nessai	Nested Sampling	Flow-based nested sampling with importance sampling
JAX-NS	Nested Sampling	Pure JAX nested sampling with GPU support
GPry	GP Emulation	Gaussian process surrogate model via Cobaya framework

Installation

git clone https://github.com/jonaselgammal/discoverysamplers.git
cd discoverysamplers
pip install .

Quick Example

from discoverysamplers import DiscoveryNessaiBridge

# Define model and priors
def model(params):
    return -0.5 * (params['x']**2 + params['y']**2)

priors = {
    'x': ('uniform', -5.0, 5.0),
    'y': ('uniform', -5.0, 5.0),
}

# Run sampler
bridge = DiscoveryNessaiBridge(model, priors)
results = bridge.run_sampler(nlive=1000, output='output/')

Usage

MCMC with Eryn

from discoverysamplers import DiscoveryErynBridge

bridge = DiscoveryErynBridge(model, priors)
bridge.create_sampler(nwalkers=32)
bridge.run_sampler(nsteps=10000)
samples = bridge.return_all_samples()

Nested Sampling with JAX-NS

from discoverysamplers import DiscoveryJAXNSBridge

bridge = DiscoveryJAXNSBridge(model, priors)
results = bridge.run_sampler(nlive=1000, rng_seed=42)

Prior Specification

priors = {
    'mass': ('uniform', 1.0, 3.0),
    'distance': ('loguniform', 0.1, 100.0),
    'phase': ('normal', 0.0, 1.0),
    'fixed_param': ('fixed', 1.0),
}

Citation

@software{discoverysamplers,
  author = {El Gammal, Jonas},
  title = {discoverysamplers: Tools for Bayesian inference with Discovery},
  year = {2025},
  url = {https://github.com/jonaselgammal/discoverysamplers}
}

License

MIT License - see LICENSE for details.