Tempest¶

Tempest is a Python implementation of the Persistent Sampling method for accelerated Bayesian inference.

What is Tempest?¶

Tempest is a Python package for fast Bayesian posterior and model evidence estimation. It leverages the Persistent Sampling (PS) algorithm, offering significant speed improvements over traditional methods like MCMC and Nested Sampling.

Key Features¶

Fast Posterior Sampling: Accelerated Bayesian inference using persistent proposals
Evidence Estimation: Reliable estimation of the Bayesian model evidence (marginal likelihood)
Multimodal Support: Effective handling of multimodal distributions with automatic clustering
Parallelization: Support for multiprocessing and MPI for large-scale problems
Flexible Priors: Easy-to-use prior distribution specification
Robust MCMC: Uses t-preconditioned Crank-Nicolson for efficient parameter space exploration

When to Use Tempest¶

Tempest is ideal for:

Large-scale scientific problems with expensive likelihood evaluations
Problems with non-linear correlations between parameters
Multimodal posterior distributions
Applications in cosmology, astronomy, and other scientific domains

Quick Example¶

Here's a simple example showing how to sample from a 10-dimensional Rosenbrock distribution:

import tempest as tp
import numpy as np
from scipy.stats import uniform

n_dim = 10  # Number of dimensions

# Define the prior transform: U(-10, 10) for each dimension
def prior_transform(u):
    return 20 * u - 10  # Transform [0,1] to [-10, 10]

# Define the log-likelihood function
def log_likelihood(x):
    return -np.sum(10.0 * (x[:, ::2]**2.0 - x[:, 1::2])**2.0 
                   + (x[:, ::2] - 1.0)**2.0, axis=1)

# Create the sampler
sampler = tp.Sampler(
    prior_transform=prior_transform,
    log_likelihood=log_likelihood,
    n_dim=n_dim,
    vectorize=True,
)

# Run the sampler
sampler.run()

# Get weighted posterior samples
samples, weights, logl = sampler.posterior()

# Get Bayesian evidence estimate
logz, logz_err = sampler.evidence()

Examples¶

Explore more applications of Tempest:

Rosenbrock Distribution - Sampling from a challenging banana-shaped distribution
Gaussian Mixture - Handling multimodal posteriors with clustering
Bayesian Model Comparison - Using Bayes factors to compare competing models
Working with Blobs - Storing auxiliary quantities from likelihood evaluations

Installation¶

Install Tempest using pip:

pip install tempest-sampler

Or from source:

git clone https://github.com/minaskar/tempest.git
cd tempest
pip install .

See the Installation Guide for more details.

License¶

Tempest is free software made available under the MIT License. For details see the LICENCE file.

Acknowledgements¶

If you use Tempest in your research, please cite the relevant papers. See the Citation page for details.