This repository is a fork of DSTAR from Seoin Back (https://github.com/SeoinBack/DSTAR), and currently under development due to an issue where the finalized models cannot be exported as pickle files.
In addition to the original models in DSTAR, I have integrated DNN, LightGBM, and CatBoost models. Furthermore, 247 simple ensemble models, combining GBR, KRR, ELN, SVR, XGBoost, DNN, LightGBM, and CatBoost, have been added for enhanced performance evaluation.
This repository contains codes and notebooks used to create results in our paper.
For more details, check out this paper (https://pubs.acs.org/doi/pdf/10.1021/acs.jcim.1c00726).
-
Generalized Adsorption Simulator for Python (GASpy)
-
Additional packages required for gaspy enviroment:
- scikit-learn (0.24.2)
- pymatgen (2021.3.3)
See DSTAR_Guide_.pdf.
DSTAR can be utilized to discover catalyst for various electrochemical reactions. You can reproduce the catalyst discovery for each reaction with the following descriptions.
To reproduce DSTAR application for CO2RR, please refer to three ipynb files in application/CO2RR/. Each ipynb file will do the following:
01_Scaler.ipynb will generates scaler to normalize the productivity.
02_Heatmap.ipynb will visualize productivty heatmap.
03_Selectivity_Plot.ipynb will plot the productivity for each product corresponding to applied potential, composition and coordination number.
More details are in each ipynb files.
All predicted CO* / H* / OH* binding energies and coordination number of prototype surface used for application can be found in application/CO2RR/data/energy and application/CO2RR/data/CN_dict.pkl, and the boundary conditions of selectivity map are in application/CO2RR/script/condition.py.