Fig. 2: Performance of machine-learned models for formation energy.
From: A critical examination of compound stability predictions from machine-learned formation energies
Parity plot for formation enthalpy predictions using six different machine learning models that take as input the chemical formula and output ΔHf. ElFrac refers to a baseline representation that parametrizes each formula only by the stoichiometric coefficient of each element. Meredig, Magpie, AutoMat, ElemNet, and Roost refer to the representations published in refs. 20,21,22,23,24, respectively. ΔHf,pred corresponds with ML predictions for aggregated hold-out sets during five-fold cross-validation of the Materials Project dataset (see “Methods” for details). ΔHf,MP refers to the formation energy per atom in the MP database. The absolute error on ΔHf is shown as the colorbar and the mean absolute error (MAE) is shown within each panel.
