approxposterior

approxposterior is a Python package for efficient approximate Bayesian inference and Bayesian optimization of computationally-expensive models. approxposterior trains a Gaussian process (GP) surrogate model for the computationally-expensive model and employs an active learning approach to iteratively improve the GPs predictive performance while minimizing the number of calls to the expensive model required to generate the GP’s training set.

approxposterior implements variants of Bayesian Active Learning for Posterior Estimation (BAPE) by Kandasamy et al. (2017) and Adaptive Gaussian process approximation for Bayesian inference with expensive likelihood functions (AGP) by Wang & Li (2018). These active learning algorithms outline schemes for GP active learning that approxposterior uses for its Bayesian posterior inference. approxposterior implemented the Jones et al. (1998) Expected Utility function for Bayesian optimization.

For information on how to install approxposterior, numerous examples, and detailed API documentation, check out the Table of Contents to the left and below.

Contents:

• Approximate Bayesian Inference
  • Rosenbrock Function Example
• Bayesian Optimization
• MAP Estimation
• Fitting a Line
• Scaling and Accuracy
• API
  • approx.py
  • gmmUtils.py
  • gpUtils.py
  • mcmcUtils.py
  • likelihood.py
  • utility.py
• Installation
• FAQs
• Citation
• Github
• Submit an Issue

Indices and tables

• Index

• Module Index