methods
Bgolearn
What it does
Bgolearn provides a modular Bayesian optimization and active learning framework for materials discovery, connecting predictive models with the choice of the next experiment.
Approach & capabilities
Supports Bayesian optimization and active learning for materials discovery.
Handles regression and classification with single- and multiple-target workflows.
Combines acquisition strategies with virtual screening and multi-objective experimental design.
Read more
Bgolearn is the first active learning framework tailored for the materials science community. Official homepage: https://github.com/Bgolearn.
Since its release, Bgolearn has gained significant traction, with over 120,000 downloads and growing adoption in both academia and industry (see: npj Computational Materials, https://arxiv.org/abs/2601.06820). It provides a lightweight yet extensible Python package for Bayesian global optimization, purpose-built for accelerating materials discovery and intelligent design workflows.
The framework supports regression and classification tasks in both single- and multi-target scenarios. It implements diverse acquisition strategies and offers a modular pipeline for virtual screening, active learning, and multi-objective optimization. Bgolearn includes nine ready-to-use utility functions, making it easy to integrate into a wide range of research applications.
Key Features:
- Built-in support for regression/classification
- Single/multi-objective and multi-target design
- Seamless integration with high-throughput and virtual screening workflows
- Lightweight design with extensible modules
> PyPI: pip install Bgolearn
> Tutorial (BiliBili): Watch here
> Try it online: Google Colab Demo
In 2025, I led the development of BgoFace, a user interface for Bgolearn published in MGR Advances. BgoFace makes Bayesian global optimization accessible for materials research by simplifying workflows and bridging experimental-computational gaps. With an intuitive design and support for real-world constraints, it enables rapid discovery without requiring ML expertise.