probayes
Probability package supporting multiple Bayesian methods including MCMC
Unlike existing libraries, probayes adopts a model-driven approach with full flexibility constrained only by the rules of probability. Since probayes is its infancy and in a state of flux, there is no manual. Currently probayes supports the following:
- Multiple random variable sampling in untransformed and transformed domain space.
- Transitional simulation, including random walks, using Markov chain conditionals.
- Discrete grid exact inference.
- Ordinary Monte Carlo random sampling.
- Ordinary Monte Carlo rejection sampling.
- Metropolis-Hastings MCMC sampling.
- Limited support for multivariate normal-covariance Gibbs sampling.
In the near-future, it is intended to expand the scope of probayes to include:
- Code initial support for approximate inference using using dense mean field messaging.
- Support derivative-based updates (HMC, gradient ascent/descent optimisation).
A quickstart is also intended, but for now there are examples in the examples/ subdirectories:
- checks/ Simple check scripts
- rv_examples/ Random variable examples
- markov/ Markov chain examples
- cov_examples/ Examples of using covariance matrices
- dgei/ Discrete grid exact inference examples
- omc/ Ordinary Monte-Carlo examples
- mcmc/ Markov chain Monte Carlo examples (Metropolis-Hastings, Gibbs...)