model_name: AlchemBERT
model_key: alchembert
model_version: v2024.12.25
date_added: '2024-12-25'
date_published: '2024-12-11'
authors:
  - name: Xiaotong Liu
    affiliation: Beijing Information Science and Technology University
    email: liuxiaotong@bistu.edu.cn
  - name: Yuhang Wang
    affiliation: Beijing Information Science and Technology University
    email: 2024020669@bistu.edu.cn

repo: https://gitee.com/liuxiaotong15/alchemBERT
doi: https://doi.org/10.26434/chemrxiv-2024-r4dnl
paper: https://chemrxiv.org/engage/chemrxiv/article-details/67540a28085116a133a62b85
pr_url: https://github.com/janosh/matbench-discovery/pull/187
checkpoint_url: https://figshare.com/ndownloader/files/53298683

license:
  code: GPL-3.0
  code_url: https://gitee.com/liuxiaotong15/alchemBERT/blob/master/LICENSE
  checkpoint: CC-BY-4.0
  checkpoint_url: https://figshare.com/articles/dataset/28690583?file=53298683

targets: E
train_task: RS2RE
test_task: IS2RE
trained_for_benchmark: true
openness: OSOD

requirements:
  torch: 2.5.1
  lightning: 2.4.0
  transformers: 4.46.3
  pymatgen: 2024.11.13

model_type: Transformer
model_params: 110_000_000
n_estimators: 1

training_set: [MP 2022]
training_cost: missing

notes:
  Description: We tested BERT on the IS2RE task of Matbench Discovery.

metrics:
  discovery:
    pred_file: models/alchembert/2025-2-7-alchembert-wbm-IS2RE.csv.gz
    pred_file_url: https://figshare.com/files/52874099
    pred_col: e_form_per_atom_alchembert
    full_test_set:
      F1: 0.436 # fraction
      DAF: 1.9 # dimensionless
      Precision: 0.326 # fraction
      Recall: 0.656 # fraction
      Accuracy: 0.708 # fraction
      TPR: 0.656 # fraction
      FPR: 0.281 # fraction
      TNR: 0.719 # fraction
      FNR: 0.344 # fraction
      TP: 28937.0 # count
      FP: 59830.0 # count
      TN: 153041.0 # count
      FN: 15155.0 # count
      MAE: 0.111 # eV/atom
      RMSE: 0.171 # eV/atom
      R2: 0.102 # dimensionless
      missing_preds: 2 # count
    most_stable_10k:
      F1: 0.603 # fraction
      DAF: 2.821 # dimensionless
      Precision: 0.431 # fraction
      Recall: 1.0 # fraction
      Accuracy: 0.431 # fraction
      TPR: 1.0 # fraction
      FPR: 1.0 # fraction
      TNR: 0.0 # fraction
      FNR: 0.0 # fraction
      TP: 4313.0 # count
      FP: 5687.0 # count
      TN: 0.0 # count
      FN: 0.0 # count
      MAE: 0.272 # eV/atom
      RMSE: 0.334 # eV/atom
      R2: -0.47 # dimensionless
      missing_preds: 0 # count
    unique_prototypes:
      F1: 0.421 # fraction
      DAF: 2.001 # dimensionless
      Precision: 0.306 # fraction
      Recall: 0.673 # fraction
      Accuracy: 0.713 # fraction
      TPR: 0.673 # fraction
      FPR: 0.28 # fraction
      TNR: 0.72 # fraction
      FNR: 0.327 # fraction
      TP: 22473.0 # count
      FP: 50988.0 # count
      TN: 131126.0 # count
      FN: 10901.0 # count
      MAE: 0.117 # eV/atom
      RMSE: 0.175 # eV/atom
      R2: 0.096 # dimensionless
      missing_preds: 0 # count