Home

Awesome

FC-GAGA

This repo provides an implementation of the FC-GAGA algorithm introduced in https://arxiv.org/abs/2007.15531 and reproduces the experimental results presented in the paper.

<p align="center"> <img width="600" src=./fig/model.png> </p>

Citation

If you use FC-GAGA in any context, please cite the following paper:

@inproceedings{
  oreshkin2020fcgaga,
  title={{FC-GAGA}: Fully Connected Gated Graph Architecture for Spatio-Temporal Traffic Forecasting},
  author={Boris N. Oreshkin and Arezou Amini and Lucy Coyle and Mark J. Coates},
  booktitle={AAAI},
  year={2021},
}

COLAB based demo

Open In Colab

Standalone Docker based demo

This workflow can be used to reproduce the FC-GAGA results without relying on the Google Colab environment. All necessary dependencies are captured in Dockerfile and requirements.txt

Clone this repository

mkdir workspace
cd workspace
git clone git@github.com:boreshkinai/fc-gaga.git

Build docker image

cd fc-gaga
docker build -f Dockerfile -t fc-gaga:$USER .

Start docker container

nvidia-docker run -p 8888:8888 -v ~/workspace/fc-gaga:/workspace/fc-gaga -t -d --shm-size="1g" --name fc_gaga_$USER fc-gaga:$USER

Go inside the container and run the main script

docker exec -i -t fc_gaga_$USER /bin/bash 
python run.py

The script run.py reproduces all the computations you can see in the Colab notebook.