This is an implementation of Graph-based Multi-ODE Neural Networks for Spatio-Temporal Traffic Forecasting
In this paper, we propose a novel architecture called Graph-based Multi-ODE Neural Networks GRAM-ODE which is designed with multiple connective ODE-GNN modules to learn better representations by capturing different views of complex local and global dynamic spatio-temporal dependencies. We also add some techniques to further improve the communication between different ODE-GNN modules towards the forecasting task. Extensive experiments conducted on six real-world datasets demonstrate the outperformance of GRAM-ODE compared with state-of-the-art baselines as well as the contribution of different GRAM-ODE components to the overall performance.
An overview of the multi ODE-GNN block which consists of three ODE modules, i.e., (a) global, (b) local, and (c) edge-based temporal dependencies as well as a new aggregation layer (d). The inputs and outputs of the multi ODE-GNN block are displayed with H and H' blocks on the left and right sides of the diagram. The shared weights among different ODE modules are marked in green, and a constraint to limit the divergence of embeddings is marked in red.
python run_stode.py
- python 3.8
- torch 1.9.0+cu111
- torchdiffeq 0.2.3
- fastdtw 0.3.4
PEMS03, PEMS04, PEMS07, and PEMS08 already downloaded and preprocessed in data folder
PEMS-BAY and METR-LA can be downloaded here this repo and this Google Drive
If you find the paper or the repo useful, please cite it with
@article{
liu2023graphbased,
title={Graph-based Multi-{ODE} Neural Networks for Spatio-Temporal Traffic Forecasting},
author={Zibo Liu and Parshin Shojaee and Chandan K. Reddy},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2023},
url={https://openreview.net/forum?id=Oq5XKRVYpQ},
note={}
}