Modeling Spatial, Temporal, and Periodical Dependencies in Traffic with Product Graphs

CS4350 Machine Learning for Graph Data (Group 27)

Authors

Aleksander Buszydlik

Karol Dobiczek

Francesco Piccoli

Edmundo Sanz-Gadea López

User Manual

• Clone repo:

git clone https://github.com/drobiu/periodic-graph-traffic-forecasting.git

• Setup conda env:

conda env create -f environment.yml

Repository structure

.
├── README.md
├── data
│   ├── ...
└── src
    ├── GLOBAL.py
    ├── __init__.py
    ├── models
    │   ├── CHEB.py
    │   ├── GCN.py
    │   ├── TAG.py
    │   ├── __init__.py
    │   ├── agcrn
    │   │   ├── ...
    │   ├── arima.py
    │   ├── dcrnn
    │   │   ├── ...
    ├── notebooks
    │   ├── basic_GNN.ipynb
    │   └── data_loading.ipynb
    ├── hyperparams_optimization.py
    ├── train.py
    └── utils.py

models directory contains the our model and the baselines utilized.

train.py file is used to train our model defined in the models folder.

utils.py file contains all the functions used to train and all the functions to create the product graph.

hyperparams_optimization.py file contains the code we used to perform hyperparams optimization.