Leaderboard
Sort models by different metrics (thermodynamic stability classification, convex hull distance regressions or tun time).
Sort best models- Discovery Acceleration Factor
- Mean Absolute Error
- Root Mean Squared Error
- True Negative Rate
- True Positive Rate
MACE
Added 2023-07-14 Published 2022-05-13 4.69M params Missing preds: 38 (0.01%) Training set: MPTrj (1.58M structures from 146k materials)
Metrics
- 0.88
- 3.78
- 0.67
- 0.06 eV / atom
- 0.58
- 0.7
- 0.1 eV / atom
- 0.89
- 0.8
- 111.9 h
CHGNet
Added 2023-03-03 Published 2023-03-01 413k params Missing preds: 0 (0.00%) Training set: MPTrj (1.58M structures from 146k materials)
Metrics
- 0.85
- 3.36
- 0.61
- 0.06 eV / atom
- 0.51
- 0.69
- 0.1 eV / atom
- 0.87
- 0.76
- 142.48 h
ALIGNN
Added 2023-06-02 Published 2021-02-22 4.03M params Missing preds: 0 (0.00%) Training set: MP v2022.10.28 (155k structures)
Metrics
- 0.84
- 3.21
- 0.57
- 0.09 eV / atom
- 0.49
- 0.3
- 0.15 eV / atom
- 0.87
- 0.67
- 0 h
M3GNet
Added 2022-09-20 Published 2022-02-05 228k params Missing preds: 353 (0.14%) Training set: MPF.2021.2.8 (188k structures from 62.8k materials)
Metrics
- 0.81
- 2.88
- 0.57
- 0.08 eV / atom
- 0.44
- 0.58
- 0.12 eV / atom
- 0.81
- 0.8
- 83.65 h
CGCNN
Added 2022-12-28 Published 2017-10-27 128k params Missing preds: 2 (0.00%) Training set: MP v2022.10.28 (155k structures)
Metrics
- 0.82
- 2.86
- 0.51
- 0.14 eV / atom
- 0.44
- -0.6
- 0.23 eV / atom
- 0.86
- 0.6
- 11.52 h
MEGNet
Added 2022-11-14 Published 2021-12-18 168k params Missing preds: 0 (0.00%) Training set: Graphs of MP 2019 (133k structures)
Metrics
- 0.83
- 2.96
- 0.51
- 0.13 eV / atom
- 0.45
- -0.25
- 0.21 eV / atom
- 0.87
- 0.58
- 3.44 h
CGCNN+P
Added 2023-02-03 Published 2022-02-28 128k params Missing preds: 2 (0.00%) Training set: MP v2022.10.28 (155k structures)
Metrics
- 0.79
- 2.56
- 0.5
- 0.11 eV / atom
- 0.39
- 0.02
- 0.18 eV / atom
- 0.8
- 0.69
- 0 h
Wrenformer
Added 2022-11-26 Published 2021-06-21 5.17M params Missing preds: 0 (0.00%) Training set: MP v2022.10.28 (155k structures)
Metrics
- 0.74
- 2.26
- 0.47
- 0.11 eV / atom
- 0.34
- -0.04
- 0.19 eV / atom
- 0.75
- 0.72
- 58.34 h
BOWSR
Added 2022-11-17 Published 2021-04-20 168k params Missing preds: 6,184 (2.41%) Training set: Graphs of MP 2019 (133k structures)
Metrics
- 0.71
- 1.96
- 0.42
- 0.12 eV / atom
- 0.3
- 0.15
- 0.17 eV / atom
- 0.69
- 0.72
- 2705.45 h
Voronoi RF
Added 2022-11-26 Published 2017-07-14 0 params Missing preds: 17 (0.01%) Training set: MP v2022.10.28 (155k structures)
Metrics
- 0.67
- 1.58
- 0.33
- 0.15 eV / atom
- 0.24
- -0.33
- 0.21 eV / atom
- 0.69
- 0.54
- 204.7 h
Per-Element Model Error Heatmaps
This periodic table heatmap shows the MAE of model-predicted convex hull distance projected onto each element. The errors for every structure in the test set are projected onto the fraction of each element in the composition and averaged over all structures. The error is the absolute difference per atom between predicted and actual energy distance to the convex hull.
- MACE