MatBench MP-Gap (DFT band gap) matbench-mp-gap Leaderboard
MatBench MP-Gap — predict the DFT band gap of 106,113 inorganic crystals from the Materials Project. Composition + structure regression task. Mean MAE (eV) across 5 folds; lower is better. Long-running materials-science benchmark from automatminer/matbench (v0.1). · Metric: mean MAE (lower is better)
| # | Model | mean MAE | Paper |
|---|---|---|---|
| 1 | coGN | 0.16 | link |
| 2 | DeeperGATGNN | 0.17 | link |
| 3 | coNGN | 0.17 | link |
| 4 | ALIGNN | 0.19 | link |
| 5 | MegNet (kgcnn v2.1.0) | 0.19 | link |
| 6 | DimeNet++ (kgcnn v2.1.0) | 0.20 | link |
| 7 | Finder_v1.2 structure-based version | 0.22 | link |
| 8 | MODNet (v0.1.10) | 0.22 | link |
| 9 | MODNet (v0.1.12) | 0.22 | link |
| 10 | Finder_v1.2 composition-only version | 0.23 | link |
| 11 | SchNet (kgcnn v2.1.0) | 0.24 | link |
| 12 | CrabNet | 0.27 | link |
| 13 | AMMExpress v2020 | 0.28 | link |
| 14 | CGCNN v2019 | 0.30 | link |
| 15 | RF-SCM/Magpie | 0.35 | link |
| 16 | Dummy | 1.33 | link |