bayes-kit

bayes-kit is an open-source Python package for black-box Bayesian inference with minimial dependencies for maximal flexiblity.

Bayesian models involve specifying a posterior log density for parameters conditioned on data. Bayesian models are typically specified in terms of a prior distribution over parameters and a sampling distribution which generates observed data conditioned on the parameters. Users are free to define models using arbitrary Python code, such as

• direct implementation in NumPy,
• Python's foreign function interface to Fortran, C++, etc., or
• automatic differentiation libraries such as
  • Stan,
  • PyTorch,
  • JAX, or
  • TensorFlow Probability.

Bayesian inference involves conditioning on data and averaging over uncertainty. Specifically, bayes-kit can compute * parameter estimates, * posterior predictions for new observations, and * event probability forecasts, all with Bayesian uncertainty quantification.

Black-box algorithms are agnostic to model structure. Algorithms in bayes-kit may require only log densities, whereas the high-performance algorithms further require gradients of the log density function.

Markov chain Monte Carlo samplers

Markov chain Monte Carlo (MCMC) samplers provide a sequence of random draws from target log density, which may be used for Monte Carlo estimates of posterior expectations and quantiles for uncertainty quantification.

Hamiltonian Monte Carlo sampler

Hamiltonian Monte Carlo simulates Hamiltonian dynamics with a potential energy function equal to the negative log density. It requires a target log density, gradient function, and optionally a metric.

Random-walk Metropolis sampler

Random-walk Metropolis is a diffusive sampler that requires a target log density function and a symmetric pseudorandom proposal generator.

Dependencies

bayes-kit has minimal dependencies, requiring only

• NumPy.

Licensing

• Code in this repository is released under the MIT License.
• Documentation in this repository is released under the CC BY 4.0 License.