Abstract

Symbolic regression libraries present a framework for automatically discovering mathematical models directly from data, bridging the gap between data-driven methods and analytical science. PySR is an open-source library that provides a high-performance framework for symbolic regression, pairing a flexible multi-population evolutionary algorithm with its lightning-fast backend, SymbolicRegression.jl. This architecture allows users to seamlessly integrate PySR with existing Python or Julia workflows—whether in a laptop setup or distributed across a cluster. In this talk, I will introduce PySR’s modular ecosystem of symbolic regression modules, and show how it can be “plugged into” existing Julia libraries to automatically learn closed-form equations tailored to the user’s domain. I will also highlight PySR’s new template expressions feature, which enables learning within a specific functional form, thus allowing scientists a flexible way of embedding domain knowledge in the search. I will also discuss PySR’s interface with deep learning as an interpretability tool.

To join remotely use the following link: https://cam-ac-uk.zoom.us/j/82860255019?pwd=ONrFS2La00nTh8iVQQdLl2JYzGattP.1 > Passcode: 071461