cavalab/srbench

What it solves

SRBench addresses the lack of standardized, reproducible, and strong benchmarks in the field of symbolic regression (SR). It aims to bridge the gap between the genetic programming (GP) and machine learning (ML) communities by providing a public, open-source framework to compare modern SR methods against strong baselines using large, diverse datasets.

How it works

SRBench provides a unified experimental setup where contributing methods must implement a scikit-learn compatible API. The benchmark evaluates methods using Docker containers to ensure reproducibility, performing hyperparameter tuning and multiple independent runs across various datasets, including those from the Penn Machine Learning Benchmarks (PMLB) and first-principles regression problems.

Who it’s for

This tool is designed for ML researchers and developers working on symbolic regression and genetic programming who need to evaluate their new methods against existing state-of-the-art techniques.

Highlights

• Extensive Method Roster: Evaluates up to 25 different symbolic regression methods in the 2025 edition.
• Reproducible Environment: Uses Docker containers for each method to ensure consistent execution.
• Diverse Datasets: Utilizes a wide range of datasets, including PMLB and first-principles problems.
• Open Standards: Requires a scikit-learn compatible API for easy integration of new methods.