Bikes-SCEG

---

tensorflow version: TF 2.0

datasets: (bikes demands in each stations)

• citibike: https://www.citibikenyc.com/system-data
• capitalbike: https://www.capitalbikeshare.com/system-data

urban feature:

• NYC weather data: https://www.kaggle.com/selfishgene/historical-hourly-weather-data
• Washington weather data: https://www.kaggle.com/marklvl/bike-sharing-dataset
• holiday data (e.g., workday, holiday):https://www.opm.gov/policy-data-oversight/pay-leave/federal-holidays

Files:

• datapreocess.py: Load data from files with a sliding-window
• bs.py: Main function. Run the file to train and test the SCEG model
• Egcn.py: Framework for time-evolving station embedding(E-GCN) and community-informed staiton embedding(B-GCN)
• GCN_layer.py: details for GCN and Evolve-GCN (Evolvegcn: Evolving graph convolutional networks for dynamicgraphs. In: AAAI’20)
• vaeTL.py:
  • encoder: latent representation for time-evolving station embedding and community-informed staiton embedding
  • decoder: output stations's demands
• Cluster.py: cluster stations to communities
• Metrics.py: MAPE and RMSPE for all stations\ settled stations \ new stations

Please cite: Qianru Wang, Bin Guo, YiOuyang, Kai Shu, ZhiwenYu, and Huan Liu. Spatial Community-Informed Evolving Graphsfor Demand Prediction. ECML2020(accepted)