Accompanying data files (Melbourne, Washington DC, Singapore, and NYC-Manhattan)

Supporting files to implement GNN training for Melbourne, Singapore, Washington DC, and NYC-Manhattan. Each folder contains the training and validation data, graph adjacency information, compiled reported energy consumption data for each city, and generated predictions.

Each zipped folder consists the following files:

• Graph data - City object nodes (.parquet) and COO format edges (.txt)
• predictions.txt (model predictions from GraphSAGE model)
• final_energy.parquet (Compiled training and validation building energy data)

The provided files are supplementary to the code repository which provides Python notebooks stepping through the data preprocessing, GNN training, and satellite imagery download processes. The data should be used hand-in-hand with the complete code workflow provided in: https://doi.org/10.6084/m9.figshare.28188242.v1.

For any questions or clarifications, please contact: winyap@mit.edu.