Crystal Graph Convolutional Neural Network (CGCNN)

Workflow

Data Preparation

python utils/feature_extraction.py
python utils/clean_id_prop.py
python utils/shuffle_id_prop.py

Training and Testing

python main.py

Results Visualization

python utils/create_graph.py

Repository Structure

• data.py: Contains data utilities and dataset class for loading and handling data.
• cgcnn.py: Main implementation of the Crystal Graph Convolutional Neural Network (CGCNN).
• main.py: Wrapper script for training and running the model.

utils/

Contains utility scripts for various tasks related to data collection, preprocessing, and analysis.

• cif.py: Functions to download CIF files.
• clean_data.py: Selects entries where both CIF and property data are present.
• clean_id_prop.py: Cleans the id_prop table by removing outliers and random values.
• create_graph.py: Generates plots for model predictions versus true values, calculates Mean Absolute Error (MAE) vs epoch, and evaluates the R² score.
• data_download.py: Script to download property data for training.
• exploring_cif.py: Explores how to work with CIF files using pymatgen.
• shuffle_id_prop.py: Shuffles the id_prop data for randomization in training.

models/

Contains the best-performing models saved during training.

results/

Stores the results of training and evaluation, including metrics, plots, and logs.