MatBench Phonons (last phdos peak) matbench-phonons Leaderboard
MatBench Phonons — predict the frequency of the highest-frequency optical phonon mode from crystal structure (1,265 samples). Smaller dataset, structure-only regression. Good probe for SOTA equivariant + GNN methods. · Metric: mean MAE (lower is better)
| # | Model | mean MAE | Paper |
|---|---|---|---|
| 1 | MegNet (kgcnn v2.1.0) | 28.76 | link |
| 2 | coNGN | 28.89 | link |
| 3 | ALIGNN | 29.54 | link |
| 4 | coGN | 29.71 | link |
| 5 | DeeperGATGNN | 33.12 | link |
| 6 | MODNet (v0.1.12) | 34.28 | link |
| 7 | DimeNet++ (kgcnn v2.1.0) | 37.46 | link |
| 8 | MODNet (v0.1.10) | 38.75 | link |
| 9 | SchNet (kgcnn v2.1.0) | 38.96 | link |
| 10 | Finder_v1.2 composition-only version | 46.58 | link |
| 11 | Finder_v1.2 structure-based version | 50.74 | link |
| 12 | CrabNet | 55.11 | link |
| 13 | AMMExpress v2020 | 56.17 | link |
| 14 | CGCNN v2019 | 57.76 | link |
| 15 | RF-SCM/Magpie | 67.61 | link |
| 16 | Dummy | 323.98 | link |