This project is focused on traffic imputation and is based on two key research papers and their associated code repositories:

• GRIN (Graph Recurrent Imputation Network) - Repository
• CSDI (Code Synthesis for Data Imputation) - Repository

• Implement diffusion model (CSDI) using Pytorch lightning (view csdi and diffmodels in https://github.com/willtryagain/traffic-imputation/tree/main/lib/nn/models, train_csdi script)
• Use mode-parallel training (FSDP) after manual wrapping
• Experimentation and visualization using wandb
• Parameter selection based on computational constraints and performance

Traffic evolution gif will take some time to load.

• PEMS-BAY
• Traffic Evolution

python -m scripts.train_csdi --config config/csdi/bay_point.yaml

conda env create -f grinc.yml

• GRIN Paper
• CSDI Paper
• Feature Propagation Paper
• DRCNN Paper
• LCR Paper