PySR

Why we included this project

Symbolic regression is a practical way to turn raw data into equations you can actually read and reason about, and PySR is one of the most complete open-source implementations of that idea. Instead of training a black-box model, it evolves compact mathematical expressions such as simple power laws, rational functions, or combinations of a few operators to optimize whatever objective you define, so what you ship is a formula you can inspect and hand to a domain expert for validation. That suits physics, engineering, materials science, and quantitative finance workflows where the goal is uncovering an underlying relationship rather than squeezing out raw accuracy. The scikit-learn-style Python front end keeps the learning curve low, while the search itself runs on a high-performance Julia backend that scales to distributed compute for large runs. Teams used to classical modeling will find it approachable, and researchers doing equation discovery get a mature, actively developed engine backed by a published algorithm.