BoTorch is a powerful framework for efficient Monte-Carlo Bayesian optimization. It offers key features such as modular plug-ins for new models, acquisition functions, and optimizers. Built on PyTorch, it allows easy integration of neural network modules and has native GPU support with autograd. BoTorch is scalable and supports scalable GPs via GPyTorch, allowing code to be run on multiple devices. The framework provides references to papers that have used BoTorch. To get started, users can install BoTorch via Conda or pip and then fit a model using the provided code. BoTorch also allows the construction of acquisition functions and optimization of these functions to find optimal candidates.
https://botorch.org/