Graph WaveNet for Deep Spatial-Temporal Graph Modeling

This is the original pytorch implementation of Graph WaveNet in the following paper: [Graph WaveNet for Deep Spatial-Temporal Graph Modeling, IJCAI 2019] (https://arxiv.org/abs/1906.00121).  A nice improvement over GraphWavenet is presented by Shleifer et al. paper code.

Requirements

• python 3

• see requirements.txt

Data Preparation

Step1: Download METR-LA and PEMS-BAY data from Google Drive or Baidu Yun links provided by DCRNN.

Step2: Process raw data

# Create data directories
mkdir -p data/{METR-LA,PEMS-BAY}

# METR-LA
python generate_training_data.py --output_dir=data/METR-LA --traffic_df_filename=data/metr-la.h5

# PEMS-BAY
python generate_training_data.py --output_dir=data/PEMS-BAY --traffic_df_filename=data/pems-bay.h5

Train Commands

python train.py --gcn_bool --adjtype doubletransition --addaptadj  --randomadj