Superalloy symbolic regression: data and code

This record contains the dataset and code for the article “Interpretable Symbolic Regression for Predicting Solidification Behavior, Phase Stability, and Cracking Susceptibility in Superalloys” (M. Younes Araghi, M. Hayder, I. Ghamarian, C.M. Stewart, S. Xu; submitted to the Journal of Materials Engineering and Performance). The dataset (superalloy_symbolic_regression_dataset.xlsx) covers 555 Ni-, Co- and Fe-based superalloys. For each alloy it gives the composition in mass fractions, composition-averaged descriptors, and six CALPHAD-derived properties: liquidus, solidus and γ′ solvus temperatures, γ′ volume fraction, Clyne–Davies crack susceptibility coefficient and freezing range. It also gives the composition-grouped training/selection/test split used in the paper. A README sheet defines every column and unit. The CALPHAD values originate from Younes Araghi and Xu, MRS Communications (2026), doi:10.1557/s43579-026-01047-1. The code (superalloy_symbolic_regression_pipelines.py) contains, in one file, the five pipelines used in the paper: PySR symbolic regression, a SISSO-style screening and sparsification method, deep symbolic optimization, a benchmark of seven black-box regressors, and the script that produces the tables and figures. Each pipeline runs from the command line, for example --model pysr or --model all; see README.md. Requirements: Python ≥ 3.10 with numpy, pandas, scipy, scikit-learn, sympy, matplotlib, openpyxl and joblib; PySR additionally needs Julia.