MatBench MP-E_form (formation energy) matbench-mp-e-form Leaderboard
MatBench MP-E_form — predict the formation energy per atom of 132,752 inorganic crystals. Composition + structure. The most-popular matbench task, used as a gold-standard regression benchmark for materials GNNs. · Metric: mean MAE (lower is better)
| # | Model | mean MAE | Paper |
|---|---|---|---|
| 1 | coGN | 0.02 | link |
| 2 | coNGN | 0.02 | link |
| 3 | ALIGNN | 0.02 | link |
| 4 | SchNet (kgcnn v2.1.0) | 0.02 | link |
| 5 | DimeNet++ (kgcnn v2.1.0) | 0.02 | link |
| 6 | GN-OA v1 | 0.02 | link |
| 7 | MegNet (kgcnn v2.1.0) | 0.03 | link |
| 8 | CGCNN v2019 | 0.03 | link |
| 9 | DeeperGATGNN | 0.03 | link |
| 10 | Finder_v1.2 structure-based version | 0.03 | link |
| 11 | MODNet (v0.1.10) | 0.04 | link |
| 12 | MODNet (v0.1.12) | 0.04 | link |
| 13 | Finder_v1.2 composition-only version | 0.08 | link |
| 14 | CrabNet | 0.09 | link |
| 15 | RF-SCM/Magpie | 0.12 | link |
| 16 | AMMExpress v2020 | 0.17 | link |
| 17 | Lattice-XGBoost | 0.75 | link |
| 18 | Dummy | 1.01 | link |