Causality Enhanced Global-Local Graph Neural Network for Bioprocess Factor Forecasting

Requirements

This work is based on BasicTS with torch==1.10.0+cu111 and easy-torch==1.2.10. Other dependencies can be seen in requirements.txt.

Train CEGLo-GNN

1. Run Decomposition/run_DYG_doz.py to perform global-local decomposition. The result is saved in Decomposition/FXL_DYG_doz.
2. Run ceglognn/Encoder_DYG_doz.py to perform embedding. Move the best checkpoints to encoder_ckpt
3. Run ceglognn/CEGLo_DYG_doz.py to perform graph generation and prediction.