model_name: Voronoi RF
model_key: voronoi-rf
model_version: v1.1.2 # scikit learn version which implements the random forest
date_added: '2022-11-26'
date_published: '2017-07-14'
authors:
  - name: Logan Ward
    affiliation: Argonne National Laboratory
    email: lward@anl.gov
    orcid: https://orcid.org/0000-0002-1323-5939
  - name: Chris Wolverton
    affiliation: Northwestern University
    email: c-wolverton@northwestern.edu
    orcid: https://orcid.org/0000-0003-2248-474X
trained_by:
  - name: Janosh Riebesell
    affiliation: University of Cambridge, Lawrence Berkeley National Laboratory
    email: janosh.riebesell@gmail.com
    orcid: https://orcid.org/0000-0001-5233-3462

repo: https://github.com/janosh/matbench-discovery
doi: https://doi.org/10.1103/PhysRevB.96.024104
paper: https://doi.org/10.1103/PhysRevB.96.024104
pypi: https://pypi.org/project/scikit-learn
pr_url: https://github.com/janosh/matbench-discovery/commit/f28cc6d
# checkpoint page: https://wandb.ai/janosh/matbench-discovery/artifacts/model/voronoi_rf_model_IS2RE/v0/files
checkpoint_url: https://api.wandb.ai/artifactsV2/default/janosh/matbench-discovery/voronoi_rf_model_IS2RE/QXJ0aWZhY3Q6MTYwNjgyOTk3NA%3D%3D/fb0cf7ce8454e5fbe1bcd49d42c0e49b/voronoi_rf_model.joblib

license:
  code: MIT
  code_url: https://github.com/janosh/matbench-discovery/blob/7c0b089e7/license
  checkpoint: MIT
  checkpoint_url: https://github.com/janosh/matbench-discovery/blob/7c0b089e7/license

requirements:
  matminer: 0.8.0
  scikit-learn: 1.1.2
  pymatgen: 2022.10.22
  numpy: 1.24.0
  pandas: 1.5.1

openness: OSOD
trained_for_benchmark: true
train_task: RS2RE
test_task: IS2E
targets: E
model_type: Fingerprint
model_params: 26_243_464
n_estimators: 1

training_set: [MP 2022]
training_cost: missing

notes:
  Description: A random forest trained to map the combo of composition-based Magpie features and structure-based relaxation-invariant Voronoi tessellation features (bond angles, coordination numbers, ...) to DFT formation energies.
  Long: This is an old model that predates most deep learning for materials but significantly improved over Coulomb matrix and partial radial distribution function methods. It therefore serves as a good baseline model to see what modern ML buys us.

metrics:
  phonons: not applicable # model doesn't predict forces
  geo_opt: not applicable
  discovery:
    pred_file: models/voronoi_rf/2022-11-27-train-test/e-form-preds-IS2RE.csv.gz
    pred_file_url: https://figshare.com/files/52057550
    pred_col: e_form_per_atom_voronoi_rf
    full_test_set:
      F1: 0.344 # fraction
      DAF: 1.509 # dimensionless
      Precision: 0.259 # fraction
      Recall: 0.511 # fraction
      Accuracy: 0.665 # fraction
      TPR: 0.511 # fraction
      FPR: 0.303 # fraction
      TNR: 0.697 # fraction
      FNR: 0.489 # fraction
      TP: 22517.0 # count
      FP: 64431.0 # count
      TN: 148440.0 # count
      FN: 21575.0 # count
      MAE: 0.141 # eV/atom
      RMSE: 0.206 # eV/atom
      R2: -0.316 # dimensionless
      missing_preds: 19 # count
    most_stable_10k:
      F1: 0.551 # fraction
      DAF: 2.487 # dimensionless
      Precision: 0.38 # fraction
      Recall: 1.0 # fraction
      Accuracy: 0.38 # fraction
      TPR: 1.0 # fraction
      FPR: 1.0 # fraction
      TNR: 0.0 # fraction
      FNR: 0.0 # fraction
      TP: 3802.0 # count
      FP: 6198.0 # count
      TN: 0.0 # count
      FN: 0.0 # count
      MAE: 0.349 # eV/atom
      RMSE: 0.417 # eV/atom
      R2: -1.012 # dimensionless
      missing_preds: 0 # count
    unique_prototypes:
      F1: 0.333 # fraction
      DAF: 1.579 # dimensionless
      Precision: 0.241 # fraction
      Recall: 0.535 # fraction
      Accuracy: 0.668 # fraction
      TPR: 0.535 # fraction
      FPR: 0.308 # fraction
      TNR: 0.692 # fraction
      FNR: 0.465 # fraction
      TP: 17854.0 # count
      FP: 56122.0 # count
      TN: 125992.0 # count
      FN: 15520.0 # count
      MAE: 0.148 # eV/atom
      RMSE: 0.212 # eV/atom
      R2: -0.329 # dimensionless
      missing_preds: 2 # count