jax-decision-making is an early-stage library which aims to provide algorithms for a
variety of sequential decision-making problems. It currently provides implementations
of several acquisition/utility functions for Bayesian optimisation, including
probability of improvement, expected improvement and Thompson sampling. The
implementations are built upon the JAX library, enabling automatic differentiation,
vectorisation, and just-in-time (JIT) compilation for high performance on CPUs, GPUs, and TPUs. This allows for efficient research, development, and deployment of decision-making agents.
Initially, jax-decision-making was created as a sub-package within the GPJax library, but it has now been separated
into its own package. Currently, Gaussian processes (implemented in GPJax) are the
primary surrogate model around which the library has been developed. Nonetheless, now
that the packages have been decoupled, we are happy to increase support for alternative
surrogate models, such as Bayesian neural networks etc. Please feel free to open issues
for features you would like to see implemented. Thise might include:
- Support for alternative sequential decision-making paradigms (e.g. reinforcement learning).
- Support for additional acquisition functions/tricks for Bayesian optimisation and experimental design (e.g. trust-regions for high-dimensional problems).
- Support for alternative surrogate models beyond Gaussian processes (e.g. Bayesian neural networks).
pip install jax-decision-makingjax-decision-making is distributed under the terms of the MIT license.