Skip to content

Latest commit

 

History

History
21 lines (16 loc) · 2.13 KB

README.md

File metadata and controls

21 lines (16 loc) · 2.13 KB
Raw

Many-body Expansion Based Machine Learning Models for Octahedral Transition Metal Complexes

This repository contains all the data and code to reproduce the results and figures from the main paper. To install all utilities and dependencies run:

pip install -e .

(tested on Python--3.8.19)

The individual steps can be executed in any order since all the intermediate data that is generated in each step already contained in this repository. However, to evalute/reprodce everything from scratch the scripts and notebooks have to be evaluated in the following order:

  1. The /raw_data folder contains a .csv of the target properties and the .xyz files for all four data sets. The notebook /scripts_and_notebooks/preprocessing/featurize_raw_data.ipynb evaluates the standard-RACs and ligand-RACs feature vectors and saves the results in four .csv files in the /data folder which are used to train the models in the following step.
  2. The two scripts in /scripts_and_notebooks/training/ can be used to retrain the different KRR and NN models by changing the values of model_type = ModelType.STANDARD_RACS and target = TargetProperty.SSE in the main function call at the very bottom of the scripts. The resulting models are saved in the /models folder.
  3. The model performance scores in the tables of the manuscript are generated from the 3 notebooks in /scripts_and_notebooks/scoring/
  4. /scripts_and_notebooks/plotting contains notebooks to generate all figures in the main text and supporting material. The four main figures are generated in the following notebooks:
    • Fig. 1: parity_plot_sse_train_val_horizontal.ipynb (output: plots/parity_plot_sse_val_horizontal.pdf)
    • Fig. 2: interpolation_plot_sse_horizontal.ipynb (output: plots/interpolation_plot_sse_horizontal_krr_1.pdf)
    • Fig. 3: parity_plot_sse_ligand_test_horizontal.ipynb (output: plots/parity_plot_see_horizontal_lig_test_nn.pdf)
    • Fig. 4: interpolation_plot_orbitals.ipynb (output: plots/interpolation_plot_homo_ls_nn_1.pdf)
    • Fig. 5: feature_importance_nn_two_body.ipynb (output: plots/feature_importance_nn_two_body.pdf)