DeepPySR -- A Symbolic Regression Framework with Dynamic Pruning, Pareto Selection, and Hierarchical Composition for Real-World Scientific Discovery

2026-07-09 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

DeepPySR is a new symbolic regression framework designed to discover analytical equations from data, producing glass-box models with direct interpretability, unlike black-box methods. This transparency is vital in clinical medicine and social science. DeepPySR addresses three key challenges in symbolic regression: high-dimensional inputs, principled Pareto-front formula selection, and data irregularities like multicollinearity. It achieves this through a dynamic variable-pruning schedule, an exponential Pareto selection criterion, and a multi-layer architecture for hierarchical symbolic composition. Benchmarked on Feynman physics, biomedical, and social-science datasets, DeepPySR significantly outperforms PySR and other baselines. For instance, it achieved an R$^2$ of 0.794 vs. 0.702 for body fat, an F1 of 0.898 vs. 0.787 for heart disease, an R$^2$ of 0.964 vs. 0.948 for student performance, and an R$^2$ of 0.525 vs. 0.370 for Raine BMI, yielding interpretable formulas aligned with domain risk factors.

Key takeaway

For Research Scientists or Machine Learning Engineers seeking highly interpretable models for scientific discovery or clinical applications, DeepPySR offers a significant advancement. Its dynamic pruning, exponential Pareto selection, and hierarchical composition directly address common symbolic regression limitations, yielding superior performance and transparent formulas. You should consider integrating DeepPySR when your project demands glass-box models and robust handling of high-dimensional or irregular data, especially where domain interpretability is paramount.

Key insights

DeepPySR advances symbolic regression for scientific discovery by integrating dynamic pruning, Pareto selection, and hierarchical composition to overcome key challenges.

Principles

• Glass-box models offer crucial transparency in sensitive domains.
• Addressing SR challenges improves model accuracy and interpretability.
• Hierarchical composition enhances symbolic formula discovery.

Method

DeepPySR employs a dynamic variable-pruning schedule, an exponential Pareto selection criterion, and a multi-layer architecture for hierarchical symbolic composition to find interpretable equations.

In practice

• Apply to clinical medicine for interpretable risk factor models.
• Use for social science research to discover analytical equations.
• Improve R$^2$ and F1 scores on complex real-world datasets.

Topics

• Symbolic Regression
• DeepPySR
• Machine Learning
• Model Interpretability
• Scientific Discovery
• Pareto Optimization
• Feature Pruning

Best for: AI Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• FunctionEvolve: Structure-Guided Symbolic Regression with LLMs — Structure-guided symbolic regression with LLMs significantly improves exact form recovery by using expression trees.
• Automating the Expert Eye: A System-Agnostic Deep Learning Framework for Rare Event Discovery in Imbalanced Force Spectroscopy — Deep learning automates rare event discovery in SMFS, overcoming extreme data imbalance with high accuracy and interpretability.
• Efficient Learning of Deep State Space Models via Importance Smoothing — Parallel variational Monte Carlo (PVMC) robustly and efficiently trains Deep State Space Models for both discriminative and generative tasks.
• Shortcut to Nowhere: Demystifying Deep Spurious Regression — Addresses continuous spurious correlations in regression tasks to prevent catastrophic model failures under deployment shifts.
• All you need is SAMPAT — SAMPAT offers a provably interpretable neural architecture by generating closed-form algebraic expressions for function approximation.
• Quick Paper Review: "There Will Be a Scientific Theory of Deep Learning" — Learning mechanics, a theory of deep learning training dynamics, is proposed as an emerging scientific framework.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.