zezhishao / STEP

Code for our SIGKDD'22 paper Pre-training-Enhanced Spatial-Temporal Graph Neural Network For Multivariate Time Series Forecasting.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

About Figure 1 in the Paper

underdoc-wang opened this issue · comments

Hi Zezhi,
Thanks for the great work. I've just got this question when I read carefully - for Figure 1, you mention that "sensor 20 and 301 are closely connected on the traffic network". I wonder whether you made this judgement based on the distance matrix, or you do have the latitude/longitude or connectivity information of the sensors. I've been looking for this information for while, but I haven't found it for PEMS04/07/08. Would you mind share it if you have?

commented

Hi, Zhaonan,
In the paper, "sensor 20 and 301 are closely connected on the traffic network" means that they are one-hop connected in graph.
PEMS03/04/07/08 do have connectivity information. You can download the datasets mentioned in BasicTS and find a csv file, e.g., datasets/raw_data/PEMS07/PEMS07.csv.
image

Hi Zezhi,
Thanks for the prompt response. Yes, that makes sense. I guess I am more interested in knowing the coordinates or IDs of the sensors. I found it for PEMS03 but not for the other three datasets. Just curious if you happen to know.

commented

Sorry for the misunderstanding, but I also do not know about such meta data.

No worry at all. Thanks for letting me know.