MetalloSR: Curated Metallurgical Equations, Structural Annotations, and Synthetic Benchmarks for Symbolic Regression

<p dir="ltr">This dataset supports the development and evaluation of MetalloSR, a metallurgy-specific symbolic regression framework that uses structural information from established metallurgical equations to guide the search for interpretable mathematical relationships.</p><p dir="ltr">The core corpus comprises 1,324 equations curated from textbooks and reference materials in metallurgy and materials science. The equations cover topics including mechanical behaviour, thermodynamics, kinetics, diffusion, phase transformations, metal forming, welding, and casting and solidification.

Equations are represented in machine-readable form using LaTeX expressions and associated metadata.</p><p dir="ltr">Structural annotations describe mathematical operators, functions, exponent patterns, and equation families. The accompanying feature representations and assignments to 13 structural clusters support the construction of cluster-specific probabilistic grammars for symbolic regression with PySR. Parsing methods and annotation fields should be considered when interpreting the structural information, because automatically extracted representations may contain parsing limitations.</p><p dir="ltr">The accompanying synthetic benchmark datasets are generated from known physical and metallurgical relationships and are intended to evaluate equation recovery and regression performance.