The inference-tools package

Introduction

This package aims to provide a set of python-based tools for Bayesian data analysis which are simple to use, allowing them to applied quickly and easily.

Inference tools is not a framework for building Bayesian/probabilistic models - instead it provides tools to characterise arbitrary posterior distributions (given a function which maps model parameters to a log-probability) via MCMC sampling.

This type of ‘black-box’ functionality allows for inference without the requirement of first implementing the problem within a modelling framework.

Additionally, the package provides tools for analysing and plotting sampling results, as well as implementations of some useful applications of Gaussian processes.

Requests for features/improvements can be made via the issue tracker. If you have questions or are interested in getting involved with the development of this package, please contact me at chris.bowman.physics@gmail.com.

Contents:

• Getting started
  • Installation
  • Jupyter notebook demos
  • Example - Gaussian peak fitting
• Markov-chain Monte-Carlo sampling
  • GibbsChain
  • PcaChain
  • HamiltonianChain
  • EnsembleSampler
  • ParallelTempering
• Constructing Likelihoods, Priors and Posteriors
  • Likelihood classes
  • Prior classes
  • The Posterior class
• Density estimation and sample analysis
  • GaussianKDE
  • UnimodalPdf
  • sample_hdi
• Gaussian process regression, optimisation and inversion
  • GpRegressor - Gaussian process regression
  • GpOptimiser - Gaussian process optimisation
  • GpLinearInverter - Gaussian process linear inversion
  • Covariance functions
  • Acquisition functions
• Approximate inference
  • conditional_sample
  • get_conditionals
• Plotting and visualisation of inference results
  • matrix_plot
  • trace_plot
  • hdi_plot