PySR
High-performance symbolic regression and equation discovery in Python and Julia.
High-performance genetic algorithms for interpretable machine learning in Python and Julia.
PySR is an open-source tool for symbolic regression that searches for interpretable mathematical expressions fitting your data. It suits data scientists and researchers who need explainable models rather than opaque black-box predictions.
What it is
The tool uses evolutionary algorithms backed by a high-performance Julia search engine to discover explicit analytic equations. It features scikit-learn compatible syntax, custom operators, loss functions, and outputs formulas in SymPy, JAX, or PyTorch formats.
Who it's for
It suits data scientists working on low-dimensional datasets or performing symbolic distillation of neural networks. It does not suit teams looking for traditional gradient-boosting tables or those unable to manage mixed Python and Julia runtime dependencies.
Key features
Scikit-Learn Interface
Fits models using standard fit and predict syntax familiar to Python developers.
Multi-Format Export
Converts discovered equations directly into SymPy, JAX, or PyTorch formats.
Custom Operators
Supports user-defined mathematical operators and custom loss functions written in Julia syntax.
Symbolic Distillation
Converts trained neural networks into analytic equations to interpret deep learning models.
Up and running in 3 steps
Our honest verdict
What's great
- Dramatically faster searches powered by an underlying Julia execution engine
- Outputs fully differentiable expressions compatible with JAX and PyTorch
- Allows custom operator and custom loss function injection
- Includes warm-start capability to resume searches without losing state
Where it falls short
- Requires both Python and Julia runtimes, which can introduce complex environment issues
- Prone to hard import crashes if system library paths like libstdc++ conflict
- Symbolic regression scales poorly on high-dimensional datasets without prior feature reduction
Pricing & license
Free under its Apache-2.0 license. The only cost is the local compute power and time required to run the evolutionary search.
Open remote AI-training roles
876 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
Why does PySR require Julia?
PySR uses a high-performance Julia library called SymbolicRegression.jl as its underlying search engine to execute millions of genetic algorithm mutations rapidly.
How can I export equations to neural network frameworks?
You can call model.jax() or model.pytorch() on a fitted model to get differentiable expressions ready for deep learning pipelines.
