Table 7 Absent-important ratio and present-ignored ratio for each dataset. Absent features are scarcely ever important (absent-important ratio is close to zero) while present features are typically used by models (present-ignored ratio has low values)
From: Extended study on atomic featurization in graph neural networks for molecular property prediction
Dataset | Absent-important ratio | Present-ignored ratio |
|---|---|---|
ESOL (random) | 0.00 | 0.22 |
ESOL (scaffold) | 0.01 | 0.18 |
Human | 0.00 | 0.27 |
QM9 | 0.11 | 0.31 |
Rat | 0.00 | 0.25 |