STSM

Official code for the paper 'Spatial-temporal Forecasting for Regions without Observations', EDBT 2024

Good News

Our recent paper "Generalising Traffic Forecasting to Regions without Traffic Observations"(paper and code) focused on same task setting, has been accepted by AAAI26.

Requirements

-torch -pandas -numpy -tables -geographiclib -scikit-learn -tqdm

The details are in the requirement.txt

Dataset

-Dataset: including the AirQ and PEMS08's traffic data, and unobserved-valid-observed idxs for each split -data: (1) AirQ and PEMS08 contain the temporal adjacent matrix (2) each 1-hop sub-graph's region graph are in the floder "region_graph"

Down data and Dataset from this link: https://drive.google.com/drive/folders/1IfwQRto6yFWRC6lwrKVCzMSm7Ivae8Qh?usp=sharing

Document

• models\cl_gcc_cnn_all.py
• preprocess\split_dat_set.py horizontally or vertically split the dataset

Train the model

In RWOF

• chmod +x ./pems08.sh (chmod +x ./airq.sh) ./pems08.sh (./airq.sh)

• python -u run_model.py --unknown_ratio 0.5 --dataset pems08 --ada 1 --a_sg_nk 0.5 --lweight 0.5 --k 35

a_sg_nk is 𝜖𝑠𝑔 in paper; lweight is lambda in paper

Questions

If you have any quesitons, please contact me at suxs3@student.unimelb.edu.au