DragonSR leaderboard runs

This dataset contains the experimental outputs used to produce the symbolic regression leaderboard presented in the paper “DragonSR: Directed Acyclic Graph Search for Equation Discovery”. Results are provided for DragonSR and PySR across 27 benchmark expressions, comprising Nguyen functions, physics equations, and remote-sensing spectral indices. Experiments were conducted without noise and with relative noise levels ε ∈ {0.01, 0.05, 0.10}. The archive is organized hierarchically by method, noise level, benchmark expression, and independent run: DragonSR/ and PySR/: symbolic regression methods; no_noise/, noise_001/, noise_005/, and noise_01/: experimental noise conditions; numbered benchmark directories: expressions ordered according to their presentation in the paper; run1/, run2/, etc.: independent experimental runs. The benchmark suite contains: Nguyen-4 to Nguyen-12; Hubble’s law, Newton’s law of gravitation, the Rydberg formula, the ideal gas law, Kepler’s third law, Bode’s law, the Schechter luminosity function, the Leavitt law, and Planck’s law; NDVI, WI2015, AWEI_sh, BAI, BSI, EVI2, VARI, SAVI, and NIRv remote-sensing indices. The deposited files provide the run-level outputs used to evaluate the discovered symbolic expressions and compute the aggregate results reported in the paper. The numerical prefixes attached to benchmark directory names correspond to their ordering in the paper and do not form part of the benchmark names.