Table 2 ECE and ENCE performance of ensemble models before and after post-hoc calibration for different datasets
From: Explainable uncertainty quantifications for deep learning-based molecular property prediction
Dataset | Uncertainty model | Aleatoric uncertainty | |||
|---|---|---|---|---|---|
ECE | ENCE | ||||
Before calibration | After calibration | Before calibration | After calibration | ||
QM9 | AtomUnc | 0.1635 | 0.0129 | 0.3094 | 0.2120 |
MolUnc | 0.1270 | 0.0700 | 0.2507 | 0.1772 | |
Zinc15 | AtomUnc | 0.3952 | 0.0722 | 0.8148 | 0.1721 |
MolUnc | 0.3950 | 0.3139 | 0.7415 | 0.6200 | |
Lipophilicity | AtomUnc | 0.0413 | 0.0396 | 0.3683 | 0.3704 |
MolUnc | 0.0157 | 0.0119 | 0.2452 | 0.2441 | |
Delaney (ESOL) | AtomUnc | 0.2118 | 0.0622 | 0.6414 | 0.5578 |
MolUnc | 0.1994 | 0.1121 | 0.6122 | 0.5468 | |