bayesnest 0.1.4

Lightweight log-space nested sampling for Bayesian inference

BayesNest

BayesNest is a lightweight, log-space nested sampler for Bayesian inference, supporting multimodal and non-Gaussian posteriors. It includes both naive rejection sampling and ellipsoidal constrained sampling (inspired by MultiNest), and computes both posterior samples and Bayesian model evidence.

---

Features

• Log evidence computation
• Posterior sampling with log-weights
• Support for multimodal inference via ellipsoidal sampling
• Corner plot visualization of posteriors
• Convergence diagnostics for log-evidence

---

Installation

pip install bayesnest

Example: Linear Gaussian Model

Once installed, you can run the examples found in the github repo

python examples/linear_gaussian_example.py

This will:

• Simulate data from a linear Gaussian model

• Use nested sampling to estimate the posterior distribution over slope a and intercept b

• Print the posterior mean and log evidence

• Generate a 2D corner plot of the posterior distribution plot

• Show the evolution of the log-evidence estimate over iterations

Example: Multimodal Posterior

A synthetic example with two separated Gaussian modes is also found in the github repo

python examples/multimodal_example.py

This example will:

• Define a 2D likelihood with two disjoint Gaussian peaks

• Sample using ellipsoidal decomposition

• Reveal both modes in the posterior distribution plot

• Demonstrate ellpisoid nested sampling’s strength in handling multimodal distributions

Expected output:

• A bimodal posterior in the corner plot

• A log-evidence trace that reflects both modes properly being integrated