The latest results of the Center for Urban Science and Computational Research, Department of Electronic Engineering, Tsinghua University "Spatio-Temporal Few-Shot Learning via Diffusive Neural Network Generation"Accepted by ICLR2024, this study proposed the GPD (Generative Pre-Trained Diffusion) model to achieve spatiotemporal learning in data sparse scenarios.

This dataset is the open source data and code of the paper. The data used for training and evaluation can be found inTime Series DataFound in (already included in the HyperNeural seed file "Time-series-data").

After downloading the data, move them to ./Data .

For each city, the research team provided the following data:

• Graph data: Records the adjacency matrix of the space-time graph.
• Time series data: Record the time sequence data of each node.

The research team provides two time series datasets: crowd flow (including DC, BM, man) and traffic speed (including metr-la, pems-bay, shenzhen, hengdu_m).