PyBOP

PyBOP is an open-source Python library designed to facilitate Bayesian optimization of expensive black-box objective functions. It provides a modular toolkit for building surrogate models, selecting evaluation points, and managing optimization campaigns. The core idea is to use a probabilistic surrogate, typically a Gaussian process, to model the objective and to choose future samples by optimizing an acquisition function such as Expected Improvement, Upper Confidence Bound, or Probability of Improvement.

PyBOP emphasizes modularity: users can swap surrogate models, kernels, and acquisition strategies without changing the surrounding optimization loop. The library supports common features found in Bayesian optimization workflows, including handling noisy evaluations, bounding constraints on parameters, and optional parallel or batched evaluations to accelerate the search.

Typical usage involves defining a wrapper around the objective function, configuring the search space through bounds or categorical options, and initiating an optimization run with a specified number of iterations or a time budget. The library maintains a history of evaluated points and observed values, updates the surrogate model after each evaluation, and returns the best observed parameters along with the corresponding objective value.

PyBOP is intended for researchers and practitioners who need a lightweight, extensible platform for hyperparameter tuning, experimental design, or automated parameter searches. It interoperates with NumPy and SciPy for numerical routines and can be extended with custom kernels, priors, or acquisition strategies as needed. Documentation and source code are typically hosted on a public repository with examples and tutorials to illustrate common workflows.