model_name: MEGNet
model_key: megnet
model_version: v2022.9.20
date_added: '2022-11-14'
date_published: '2021-12-18'
authors:
  - name: Chi Chen
    affiliation: UC San Diego
    orcid: https://orcid.org/0000-0001-8008-7043
  - name: Weike Ye
    affiliation: UC San Diego
    orcid: https://orcid.org/0000-0002-9541-7006
  - name: Yunxing Zuo
    affiliation: UC San Diego
    orcid: https://orcid.org/0000-0002-2734-7720
  - name: Chen Zheng
    affiliation: UC San Diego
    orcid: https://orcid.org/0000-0002-2344-5892
  - name: Shyue Ping Ong
    affiliation: UC San Diego
    orcid: https://orcid.org/0000-0001-5726-2587
    email: ongsp@ucsd.edu

repo: https://github.com/materialsvirtuallab/megnet
url: https://materialsvirtuallab.github.io/megnet
doi: https://doi.org/10.1021/acs.chemmater.9b01294
paper: https://arxiv.org/abs/1812.05055
pypi: https://pypi.org/project/megnet
pr_url: https://github.com/janosh/matbench-discovery/pull/85
# checkpoint used: Eform_MP_2019 MAML checkpoint
# https://github.com/materialsvirtuallab/megnet/blob/f37057602/megnet/utils/models.py#L21
checkpoint_url: https://figshare.com/files/22291785

license:
  code: BSD-3-Clause
  code_url: https://github.com/materialsvirtuallab/megnet/blob/f3705760/LICENSE.md
  checkpoint: BSD-3-Clause
  checkpoint_url: https://github.com/materialsvirtuallab/megnet/blob/f3705760/LICENSE.md

requirements:
  megnet: 1.3.2
  pymatgen: 2022.10.22
  numpy: 1.24.0
  pandas: 1.5.1

openness: OSOD
trained_for_benchmark: false
train_task: RS2RE
test_task: IS2E
targets: E
model_type: GNN
model_params: 167_761
n_estimators: 1

# we tested the Eform_MP_2019 checkpoint of MEGNet, the original 2018 version was trained on 'Graphs of MP 2018'
# title: Graphs of MP 2018
# url: https://figshare.com/articles/dataset/7451351
# n_structures: 69_239
training_set: [MP Graphs]
training_cost: missing

hyperparams:
  graph_construction_radius: 4.0 # Å, from https://github.com/materialsvirtuallab/megnet/blob/f370576026/README.md#data-sets
  max_neighbors: .inf

notes:
  Description: MatErials Graph Network is another GNN for material properties of relaxed structure which showed that learned element embeddings encode periodic chemical trends and can be transfer-learned from large data sets (formation energies) to predictions on small data properties (band gaps, elastic moduli).
  Training: Using pre-trained model released with paper. Was only trained on `MP-crystals-2018.6.1` dataset [available on Figshare](https://figshare.com/articles/Graphs_of_materials_project/7451351).

metrics:
  phonons: not applicable # model doesn't predict forces
  geo_opt: not applicable
  discovery:
    pred_file: models/megnet/2022-11-18-megnet-wbm-IS2RE.csv.gz
    pred_file_url: https://figshare.com/files/51607286
    pred_col: e_form_per_atom_megnet
    full_test_set:
      F1: 0.513 # fraction
      DAF: 2.699 # dimensionless
      Precision: 0.463 # fraction
      Recall: 0.574 # fraction
      Accuracy: 0.813 # fraction
      TPR: 0.574 # fraction
      FPR: 0.138 # fraction
      TNR: 0.862 # fraction
      FNR: 0.426 # fraction
      TP: 25311.0 # count
      FP: 29342.0 # count
      TN: 183529.0 # count
      FN: 18781.0 # count
      MAE: 0.128 # eV/atom
      RMSE: 0.204 # eV/atom
      R2: -0.277 # dimensionless
      missing_preds: 1 # count
    most_stable_10k:
      F1: 0.632 # fraction
      DAF: 3.022 # dimensionless
      Precision: 0.462 # fraction
      Recall: 1.0 # fraction
      Accuracy: 0.462 # fraction
      TPR: 1.0 # fraction
      FPR: 1.0 # fraction
      TNR: 0.0 # fraction
      FNR: 0.0 # fraction
      TP: 4620.0 # count
      FP: 5380.0 # count
      TN: 0.0 # count
      FN: 0.0 # count
      MAE: 0.304 # eV/atom
      RMSE: 0.336 # eV/atom
      R2: -0.908 # dimensionless
      missing_preds: 0 # count
    unique_prototypes:
      F1: 0.51 # fraction
      DAF: 2.959 # dimensionless
      Precision: 0.452 # fraction
      Recall: 0.585 # fraction
      Accuracy: 0.826 # fraction
      TPR: 0.585 # fraction
      FPR: 0.13 # fraction
      TNR: 0.87 # fraction
      FNR: 0.415 # fraction
      TP: 19537.0 # count
      FP: 23651.0 # count
      TN: 158463.0 # count
      FN: 13837.0 # count
      MAE: 0.13 # eV/atom
      RMSE: 0.206 # eV/atom
      R2: -0.248 # dimensionless
      missing_preds: 0 # count