. 2024 Jan 26;128(5):945–957. doi: 10.1021/acs.jpca.3c07129

Table 2. Results of 5-fold Cross-Validation for the Active Learning Strategies Evaluated by Training cNNPs with an Oxalic Acid CSP Landscape^a.

strategy	energy MAE (kJ mol^–¹)	energy RMSE (kJ mol^–¹)	final dataset size
random	0.86 (0.04)	1.20 (0.13)	320 (36)
highest uncertainty	0.97 (0.06)	1.29 (0.10)	252 (45)
highest uncertainty FPS	0.96 (0.10)	1.33 (0.16)	288 (67)

All cases used the active learning hyperparameters in entry 6 of Table 1 with the uncertainty threshold of 1 kJ mol^–1.

Table 2. Results of 5-fold Cross-Validation for the Active Learning Strategies Evaluated by Training cNNPs with an Oxalic Acid CSP Landscapea.