Fig. 4
From: Deep neural networks for accurate predictions of crystal stability
Accuracy of stability classification. Plots of the accuracy of stability classification of the ANN models compared to DFT as a function of the Ehull threshold for a. garnets, and b. perovskites. The accuracy is defined as the sum of the true positive and true negative classification rates. A true positive (negative) means that the Ehull for a particular composition predicted from the optimized artificial neural network model and DFT are both below (above) the threshold. For the mixed compositions, an Ehull is calculated for all orderings (20, 7, and 18 orderings per composition for C-, A-, and D-mixed garnets, respectively, and ten orderings per composition for both A- and B-mixed perovskites)
