astroautomata/PySR ? reverse-engineered prompt

Build me a Python library for symbolic regression that can search for simple, readable equations from data and feel easy to use like scikit learn.

I want to be able to pass in X and y, train a model, and then get back the best equations it found, along with predictions and a way to inspect the candidate formulas it discovered. It should support custom math operators and custom loss functions, and it should work fast by using Julia under the hood for the search. Make it pleasant for real users, with a clean Python API, a basic example script, and docs that show how to install it, fit a model, and interpret the results.

Please also include support for running on common setups like pip, conda, Docker, and notebooks, and make sure the project structure is ready for tests, examples, and documentation. If you need current Julia or package details, look them up online before wiring anything up.